Analyzing pre-emptive price cuts upon threat of entry : Comparing the effects for business and leisure travelers within the airline industry C

(1)

Master’s Thesis

Analyzing pre-emptive price cuts upon threat

of entry

Comparing the effects for business and leisure travelers

within the airline industry

Lars Dedding

10214925

Business Economics: Managerial Economics and Strategy

15EC

(2)

2

Statement of Originality

This document is written by Lars Dedding who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

3

Abstract

This research aimed at providing a better understanding behind pre-emptive price cuts from incumbent airline carriers upon threat of entry. By making a distinction between business and tourist routes, it is tested whether routes containing a higher proportion of loyal business travelers experience greater pre-emptive price cuts. This would be in line with the view that incumbents lower their prices in order to lock in loyal business customers. Using an extended dataset, insufficient significant results were found to suggest a difference between business and tourist routes. This research was able to verify that pre-emptive price cutting did not occur on routes were entry was pre-announced. This suggests that the motive of incumbents to lower their prices upon threat of entry was not locking-in valuable customers but deterrence of entry.

(4)

4

1. Introduction

The term “Southwest Effect” was introduced in 1993 by the Bureau of Transportation (Bennet and Craun, 1993). Within the airline industry, this phrase describes the negative effect on fare prices and positive effect on airline traffic on any route where Southwest Airlines entered, being one of the first low-cost carriers in the U.S. market. This effect from the actual entry of Southwest Airlines has been documented numerous times (Morrison, 2001; Pitfield, 2008). Besides the actual entry, the threat of entry is reported to have a downward sloping effect on fares as well. Goolsbee and Syverson (2008) studied this by defining threat of entry as the occurrence when Southwest Airlines starts flying at both endpoints of a certain route, while not entering that route yet. In figure I, this is explained using an example of the Boston to Philadelphia route. In this example, Southwest was already flying from Boston airport since 2004. When it started flying to Philadelphia in 2009 it threatened entry on the Boston to Philadelphia route.

Within the dataset of Goolsbee and Syverson (2008), ranging from 1993 to 2004, they find that the probability of Southwest Airlines entering a route with at one end a Southwest hub is small but significantly larger than the probability of entering a route with no bordering Southwest hub. However, when both endpoints of that route become Southwest airports, this probability is increased with a factor 70, making entry significantly more likely to occur. Goolsbee and Syverson (2008) found that incumbent firms pre-emptively lowered their prices in anticipation of Southwest entering that route. However, they did not find conclusive evidence on the rationale behind the incumbents’ behavior, which seems

counterintuitive from a standard economic perspective. This, since ticket prices are reduced in the period before the entrant joins the route, while not altering the prices would lead to higher profits in these periods. Goolsbee and Syverson (2008) suggest that the behavior could be related to an effort to lock-in valuable loyal customers.

This research aims at providing evidence for the lowering of loyal customers’ fares in an attempt to stop them from shifting to Southwest Airlines once Southwest enters. It poses

(6)

6 the following research question: Is the price decrease of airline industry incumbents’ due to

threat of entry based on locking-in loyal customers?

In order to answer this question, the research from Goolsbee and Syverson (2008) is

elaborated upon by making a distinction between business and leisure travelers. Within the literature regarding the airline industry, this is a commonly made distinction due to the difference in demand elasticities (Brons et al. 2002). Besides the difference in demand elasticity, Cairns and Galbraith (1990) suggest that business travelers tend to be relatively more loyal customers to an airline company and more likely to participate in frequent flyer programs than leisure travelers. To make the distinction between these two types of customers, a methodology similar to Gerardi and Shapiro (2009) is used. They separate the routes of the itineraries into “tourist routes” and “big-city routes”. They assume that tourist routes contain mostly leisure travelers, who have a relatively high elasticity of demand. Between big cities, a larger proportion of more price-insensitive business travelers is assumed to travel along leisure travelers (Gerardi and Shapiro, 2009). By comparing the effects of Southwest Airlines threatening entry on these routes, one can compare the effects on incumbents’ fares of loyal and non-loyal customers.

By extending the dataset of Goolsbee and Syverson (2008), additional evidence for the price decreasing Southwest effect on actual entry was found. Also, significant price decreases upon the threat of entry of Southwest indicate pre-emptive price decreases. When comparing the effect of threatened entry on fares between 303 business routes and 229 leisure routes, using an interacted model, there are insufficient significant results to indicate that this effect was greater for business routes. However, with the use of the extended dataset, evidence for entry deterrence as motive for pre-emptive price cuts was found. This was achieved by comparing pre-announced entry to threatened entry, where only

significant decreases were found in the latter, where incumbents have an incentive to cut prices to deter entry.

The setup of this research is as follows: the literature review will be commenced with, where relevant literature regarding threat of entry will be discussed. After discussing several theoretical predictions and reviewing other relevant empirical studies, the hypotheses will

(7)

7 be formulated. Then the data and methodology section will carefully describe how the dataset was realized in order to identify threat of entry. After that the results will be

discussed, as well as the validity of the research. This will be followed by concluding remarks and recommendations for future research.

2. Literature and hypotheses

This section will commence with theoretical predictions regarding threat of entry, in which the main reasons for pre-emptive price cutting are listed. Then empirical literature

concerning the airline industry, and the Southwest effect in particular, is discussed.

Hereafter, the main influence to this research, namely the paper of Goolsbee and Syverson (2008) on pre-emptive price cutting in the airline industry is explained, and the way this research expands upon it. From this, the hypotheses are formed.

The bulk of the literature regarding pricing under threat of entry relates to entry deterrence, which consists of several underlying theories. Gaskins (1971) proposed a model in which a monopolist faces entry by competitors. The response of the incumbent can either be to enjoy short-run monopoly profits and to ignore the entry this encourages, or to lower the price towards the entry deterring limit price. Although the optimal choice for the incumbent is dependent on several factors such as cost advantages and market growth, in this model it is vital that potential entrants view the current pre-entry price as a proxy for the future post-entry price.

This assumption has been a subject of discussion (Dixit, 1980; Klemperer, 1987). According to Dixit (1980), possible entrants are aware that incumbents are likely to lower their prices in order to deter entry which makes holding the post-entry price near the limit price an incredible threat. Therefore Dixit proposes a Cournot-based view in which new entrants do not assume the pre-entry quantity to be a determinant for the post-entry quantity, but see post-entry market directions as exogenous. The incumbent can still benefit if this were to hold, by altering the pre-entry market circumstances. By committing itself to capacity

(8)

8 increases pre-entry, portraying excess capacity in this period, the incumbent is able to signal a credible threat to potential entrants that it is able to lower prices by increasing output post-entry. Milgrom and Roberts (1962) also introduce a model in which the post-entry price acts as a signal to potential entrants. Herein, amidst imperfect information, the pre-entry price is set as such to signal the cost structure of the incumbent firm. By doing so, the incumbent firm is able to lower the probability of deterrence of entry.

Klemperer (1987) opposes the view that signalling is the reasoning behind pre-emptive price cutting, since it would be too costly to do so in this way. The rationale that Klemperer

proposes is that firms cut prices in order to lock-in customers to their firm, which decreases potential demand for entrants. The underlying assumption for locking in customers to a firm by decreasing their price is related to switching costs, which are to be incurred by customers if they purchase an identical product from different suppliers. In the case of the airline industry, this argument holds due the several loyalty programs which are in place (Borenstein, 1989). Frequent Flyer Programmes (FFPs) are designed in order to reward travellers that commit to travelling with one carrier, by providing these customers with points that can be exchanged for free travel for example. In doing so, loyalty is generated between passengers and their carrier of choice (Lederman, 2007). Borenstein (1989) states that FFPs are especially effective in locking in repeat customers, including lucrative business travelers. Cairns and Galbraith (1990) assent to this view, stating that FFPs are generally designed to reward business-class travellers more generously than travellers in economy-class.

The reason that carriers have additional incentive to lock in business travellers, is because of their relatively low elasticity of demand (Brons et al, 2002). This allows the carrier to ask relatively higher fares from business travellers then from leisure travellers, who tend to be more price-sensitive. This difference in elasticities of demand is widely documented (Brons et al, 2002; Gerardi and Shapiro, 2009; Cairns and Galbraith, 1990), and is used as

(9)

9 The Southwest Effect as described earlier is well-established within empirical research, decreases in fares are reported at 21, 48 and 55 percent once Southwest starts flying a certain route (Goolsbee and Syverson, 2008; Windle and Dresner, 1995; Bennet and Craun, 1993, respectively). Along with the decrease in price, these scholars also find substantial increases in airline traffic from 40 percent to a two- to six-fold increase, respectively. The reason Southwest has been a major subject for research in the airline industry, has mainly been due to its low-cost structure and consequential low prices, but also because it does not operate on a hub-and-spoke system like most competitors (Bennet and Craun, 1993).

Different to the hub-and-spoke system, which offers flights to and from a central hub airport, Southwest specializes in frequent direct flights. This provides more direct routes than carriers operating with a hub-and-spoke system can offer.

Due to the difficulty of properly measuring threat of entry, empirical research in this field is scarce. Most of the literature studies the threat of entry related to entry deterrence (Dafny, 2005; Ellison & Ellison, 2011). Their findings suggest that incumbents lower prices in

advance to potentially deter entry. The only study found that relates the pre-emptive price upon threat of entry to locking in loyal customers, is that of Goolsbee and Syverson (2008).

Goolsbee and Syverson (2008) proposed a methodology to proxy threat of entry that focusses on the period before Southwest entered a particular route. The threat of entry is induced by Southwest starting service at both endpoints of a route, without flying that route. Goolsbee and Syverson (2008) computed that the possibility of Southwest entering a route where they offered service at both endpoints increased with factor 70 compared to where Southwest offered service at only one of the endpoints. This increased probability of entry provided a method to test the effect of threat of entry on incumbent prices. The period before Southwest entered both endpoints was also taken into consideration since Southwest announced their arrival several months upon starting service. Goolsbee and Syverson (2008) explain that Southwest did so in order to facilitate their arrival (i.e. hire staff) and by means of selling tickets. Because of this, the incumbents would have had access to information regarding the expansion pattern of Southwest before actual entry.

(10)

10 Goolsbee and Syverson (2008) find evidence of pre-emptive fare decreases for both the period in which Southwest started service in both endpoints as for the period before. The authors only find suggestive evidence to explain this anticipatory behavior. Pre-emptive price cutting appears to be more apparent in routes with a larger proportion of business travelers as opposed to leisure travelers. Since business travelers are known to be more loyal than leisure travelers (Cairns and Galbraith, 1990), this suggests that incumbent firms lower their prices in order to lock-in valuable customers.

Since no data is available on the distribution of business and leisure travellers within flights, it is common within the literature to proxy the proportion of travellers based on the route (Borenstein, 1989; Gerardi and Shapiro, 2009). Goolsbee and Syverson (2008) make use of a relatively rudimentary proxy for leisure travellers, based on Borenstein (1989), which

displays two shortcomings in identifying tourist and business travellers.

Firstly, although Borenstein (1989) uses the ratio of accommodation earnings divided by total earnings per Metropolitan Statistical Areas to classify tourist routes, Goolsbee and Syverson (2008) use a state-wide measure. By doing so, it fails to distinguish between the multiple airports that states contain, and therefore the variation within the accommodation earnings of the cities served by the airports is not taken into account. For example, Nashville International Airport and Memphis International Airport, the two largest airports within the state of Tennessee, vary significantly in their accommodation ratio. When using the

Metropolitan Statistical Area, which represents the area surrounding a city over 50,000 inhabitants including suburbs (Federal Register, 2010), their accommodation ratios are 0.66% and 1.76%, respectively. Specifics on these computations are provided in the methodology section.

The second shortcoming is that Goolsbee and Syverson (2008) denote all routes that are not leisure intensive as business routes. More precisely, all routes where at least one endpoint is in a state which has an accommodation ratio of over 1 percent, are labelled tourist routes and all other routes are regarded business routes. This indirectly assumes an equal

proportion of business travellers between all non-leisure endpoints, which is unlikely to hold since these endpoints consist out of a wide variation of small and big cities.

(11)

11 Gerardi and Shapiro (2009) propose a proxy that labels business routes as routes between the 30 largest Metropolitan State Areas within the United States based on population. This research will follow the methodology of Gerardi and Shapiro (2009) to distinguish business and leisure travellers, since it mitigates both issues of Goolsbee and Syverson’s (2008) methodology. To do so, all routes are divided among three categories: business routes, tourist routes or neither, where business routes exhibiting high accommodation ratios are not labelled as business routes to create disjunctive routes. Even though this research makes use of a methodology that disregards data belonging in neither of the groups, it does so aiming to improve the identification of business and leisure travellers in comparison to Goolsbee and Syverson (2008). This research also functions as a replication study, increasing the time period from 1993-2004 to 1993-2016, to study whether the pre-emptive price cutting effect still holds when the dataset is expanded.

In this research it will be studied whether pre-emptive price cuts by incumbent carriers upon threat of entry of Southwest differs between leisure and business travelers,

succeeding the research of Goolsbee and Syverson (2008). Besides expanding the original dataset, the goal of this research is to determine whether prices are lowered pre-emptively in order to lock-in loyal customers. Cairns and Galbraith (1990) suggest that business travelers exhibit more loyal behavior than leisure travelers. Using the methodology of Gerardi and Shapiro (2009) to identify between business and leisure travelers allows a more precise separation of the two. This is done by separating routes into tourist routes and big-city routes, where a higher proportion of business travelers is assumed in the latter. The decrease of ticket fares upon threat of entry by Southwest is hypothesized to remain negative when extending the dataset. Within this result, the effect on business routes is hypothesized to be greater than on tourist routes due to the loyal nature of business customers.

If the pre-emptive price cutting were to be larger on business routes than on tourist routes, then the negative effect of actual entry on the fares is hypothesized to be greater on tourist routes than on business routes. This is because prices on tourist routes would have been

(12)

12 lowered less than prices on business routes in the period before entry, therefore prices on tourist routes would need to drop more than business routes upon entry of Southwest.

3. Data & Methodology

This section provides information on how the dataset was formed and gives a description of the methodology used.

Similar to Goolsbee & Syverson (2008) and Gerardi & Shapiro (2009), data on the U.S. airline industry is retrieved from the DB1B database from the Bureau of Transportation Statistics. Herein, a 10 percent sample of all domestic itineraries are provided per quarter ranging from the beginning of 1993 to the last quarter of 2016. All available data is taken into

account since Southwest Airlines kept expanding their routes and number of airports, as can be inferred from Southwest Media (2017). Southwest Media provides the exact dates on which it entered which airport. The latest addition to the list of Southwest Airlines’ airports is Long Beach Airport, which is added on June 5, 2016. This is one of the 48 airports from which Southwest started service during the time period of the dataset. Goolsbee & Syverson (2008) made use of data ranging from 1993 to the last quarter of 2004, during which 22 airports were entered. Southwest expanded to 26 more airports since then, substantially increasing the size of the dataset in comparison.

Goolsbee & Syverson (2008) made use of Borenstein’s (2010) dataset named “Market Data”, which consists of cleaned and aggregated DB1B-data. Although this particular dataset was inaccessible for this research, Borenstein’s (2010) methodology was closely followed in order to replicate a similar yet largely extended dataset. By merging quarterly market and ticket-data from the DB1B, a large dataset containing quarterly route-specific data was obtained. To provide a sense of scale on the dataset, the first quarter of 1993 contained 2.8 million route-carrier-specific observations. For example, one observation herein is an United Airways itinerary in the first quarter of 2005 from Los Angeles LAX airport to Washington IAD airport. For this observation, there is information regarding fares, the origin and

(13)

13 destination of the route (including locations of stops for non-direct flights), the non-stop distance in miles between the two endpoints as well as data on whether it was a one-way or round-trip journey.

In order to recreate the data as used by Goolsbee and Syverson (2008), several adjustments had to be made to the DB1B data. In the end, the data will consist of average logged fares per aircraft carrier at the route-level, where a route is defined by its airports of origin and destination. To mitigate any issues with connecting flights, only direct flights are taken into account.

As is common in the literature (Gerardi & Shapiro, 2009), and similar to Borenstein’s (2010) dataset: fares of round-trip itineraries were halved to correct for the difference in one-way and two-way tickets. Then the flight was split into two separate one-way flights. Data with ticket prices below $20, above $9998 or with low dollar credibility (according to the Bureau of Transportation Statistics) were removed to increase the reliability of the data. In line with Goolsbee and Syverson (2008), only data on legacy carriers was used and only the 86

airports where Southwest offered service over the studied time period are included. After these adjustments were made for all quarters, the data was collapsed to carrier-route-quarter level, providing 612,369 observations. One observation herein is the average fare from United Airways itineraries from LAX to IAD in the first quarter of 2005.

Between 1993 and 2016, 1328 routes were entered in total by Southwest. Out of these routes, 523 routes were entered immediately upon Southwest providing service in both endpoints. On 968 routes, Southwest did not start flying immediately upon Southwest presence in both endpoints, these routes will be referred to as threatened routes. Out of these 968 routes, Southwest eventually entered the route by the end of the dataset on 911 occasions.

Table I provides additional summary statistics regarding the ticket fares and their standard deviation. These statistics are comparable to those of Goolsbee and Syverson (2008), whose standard deviation of mean logged fares on threatened routes is 0.45, equal to the standard deviation of threatened routes in this dataset. Over 1993 to 2004, they find 704 threatened

(14)

14 routes, out of which 533 were eventually entered by Southwest in 2004 providing over 19,000 observations. This research extends the time period up to 2016 providing over 66,000 observations. Similar to Goolsbee and Syverson (2008), only observations from 12 months before Southwest threatened entry up to 12 months after Southwest entered the route are taken into consideration. By doing so, the reference period is kept the same for all threats. This is elaborated upon in the methodology section.

In order to determine whether fare prices were reduced more for loyal customers, the methodology of Gerardi and Shapiro (2009) is closely followed. In their research routes are segmented into business routes, tourist routes or neither. Business routes and tourist routes are designed in such a way that they are mutually exclusive, meaning that a route cannot qualify as both business and tourist at the same time.

Each airport in the sample is matched to the Metropolitan Statistical Area (MSA) in which it is located. To define the leisure endpoints, the ratio of accommodation earnings to total non-farm earnings are computed for each MSA. These calculations are performed for all years after which the mean is calculated. Although Gerardi and Shapiro (2009) compute this using data of a single year, this research calculates the average population over the 1993 to 2016 period as inferred from U.S. Census Bureau data to increase accuracy. Similar to Gerardi and Shapiro (2009) the airports within an MSA above the 85th_{percentile of the ratio}

are defined as leisure endpoints. All threatened routes to and from a leisure endpoint are defined as tourist routes. 229 routes qualify as tourist route, providing a total of 17,344 observations.

Big-city endpoints are defined as airports located in one of the 30 most populated MSAs out of the 383 MSAs in the United States. Similarly, the average population over the 1993 to 2016 period as inferred from U.S. Census Bureau data is calculated to increase accuracy. A big-city route is defined as a threatened route with a big-city endpoint at either side.

Gerardi and Shapiro exclude several airports, which were already included in tourist routes. Four airports in leisure endpoints within this dataset were excluded from the business routes for that reason, namely Fort Lauderdale, San Diego, Palm Beach and Orlando. This provides a total of 303 routes within the dataset, with a total of 24,671 observations.

(15)

15 Table I displays summary statistics of the data which compares overall, threatened, tourist and business routes. The underlying assumption behind the division in tourist and business routes is that the latter should contain a larger proportion of business travelers. Gerardi and Shapiro (2009) argue that if this holds, then the dispersion of tourist route fares should be smaller than the dispersion of business route fares. If tourist routes contain predominantly leisure travelers, then the fares of this price-sensitive group should be unimodally

distributed. If business routes were to contain two types of traveler, price-sensitive and price-insensitive, the distribution of their fares should be bimodal.

From table I the level of dispersion cannot be inferred, since the adjusted dataset is already collapsed to the average fare level. Computing the relative standard deviation, as measure of dispersion, from table I for the business and tourist routes would result in similar

outcomes. However, calculating this with the dataset before collapsing gives an correct representation of the price dispersion. Computing the relative standard deviation, by dividing the standard deviation over the mean, for the first quarter of 1993 results in values of 0.65 and 0.76 for the tourist and business routes respectively1_{. To verify this result,}

another measure of dispersion is calculated. For this, the Gini-coefficient is used as is

regularly done in the airline industry (Borenstein & Rose, 2004; Gerardi & Shapiro, 2009. For the business routes, the Gini-coefficient was 0.375 and for the tourist routes the GINI-coefficient was 0.315. Since a higher Gini-GINI-coefficient displays higher dispersion, business routes are more likely to serve both business and tourist travelers.

Based upon the methodology of Goolsbee and Syverson (2008), longitudinal regressions are done to implement an event study on all threatened routes within the dataset.

This is implemented in regression model (I): (I) 𝑦𝑖𝑟𝑡 = 𝑎𝑟𝑖+ 𝛽𝑖𝑡 + ∑ 𝛾 12 𝑘=−8 (SW_both_endpoints)𝑟,𝑡0+𝑘 + ∑ 𝛿 12 𝑙=0 (SW_flying)𝑟,𝑡𝑒+𝑙 + θX𝑖𝑟𝑡 + 𝜀_𝑖𝑟𝑡

1_{Since the full dataset without collapsing is too large to work with, only the results for the first quarter of 1993} is shown. Similar results hold if the data of other quarters are used.

(16)

16 Herein, i represents the carrier, r represents the route and t represents the time in yearly quarters. The dependent variable in this specification is y, representing the mean logged fares of incumbent carriers. A logarithmic transformation is used in order to mitigate bias caused by the skewness of average fares. Inferred from a skewness and kurtosis test for normality, the hypothesis that average fares were normally distributed was rejected at the 1 percent level.

The studied time period varies for each threatened route, depending on the moment Southwest started service in both endpoints but did not yet start flying (denoted as t0), and

the moment Southwest started flying the route (denoted as te). Included in the data for

each threatened route are the three years before t0, the three years after t0 (provided that

Southwest did not yet start flying), up to the three years after te. Dummy variables are

generated for when Southwest starts flying (SW_flying) and for when Southwest becomes active in both endpoints (SW_both_endpoints) for all but the 9th_{to 12}th_{quarter before t}

0.

Therefore, the fares in this third year before Southwest started threatening entry act as reference period. Figure 2 provides a timeline for the Philadelphia to Boston route to illustrate this. Here, t0 occurred in the third quarter of 2009 and te occurred in the second

quarter of 2010. In this timeline, all the shaded quarters are included in the data, yet only dummies are made for the SW_both_endpoints and SW_flying quarters. These dummies are constructed to be mutually exclusive, so that the coefficient of one of the dummies can be interpreted as the relative change to the reference period.

Besides these dummies, X represents a vector of control variables that account for the operating costs in airports as will be expanded upon. Besides these control variables the regressions will account for fixed effects on carrier-route (αri) and carrier-quarter (βit) level.

For each of the regressions, the standard errors will be clustered by route-carrier, this is done to mitigate intertemporal correlation in the error term (Goolsbee & Syverson, 2008).

As discussed in the hypotheses, it is assumed that pre-emptive fare cuts remain when the dataset is expanded. If this were to hold then the regression results should portrait negative coefficients for the SW_both_endpoints-variables. To verify whether this effect is greater on

(17)

17 business routes than it is on tourist routes, a separate regression model (II) is used that includes interaction effects. Model (II) is constructed as follows:

(II) 𝑦𝑖𝑟𝑡 = 𝑎𝑟𝑖 + 𝛽𝑖𝑡+ ∑ 𝛾 12 𝑘=−8 (SW_both_endpoints)_𝑟,𝑡_0+𝑘 + ∑ 𝛿 12 𝑙=0 (SW_flying)_r,𝑡_𝑒+𝑙 + 𝑡𝑜𝑢𝑟𝑖𝑠𝑡 ∗ ∑ 𝜂 12 𝑘=−8 (SW_both_endpoints)_r,t_0+𝑘 + tourist ∗ ∑ 𝜇 12 𝑙=0 (SW_flying)r,t𝑒+𝑙 + θX𝑖𝑟𝑡+ 𝜀𝑖𝑟𝑡

The only difference between (I) and (II) is the addition of the tourist dummy interaction terms, which equals one for tourist routes. This model will be used on a restricted dataset consisting of only business and tourist routes. Doing so enables this model to test whether business routes exhibit larger fare reductions upon threat of entry than tourist routes. In order to confirm the hypothesis, negative values are expected for the γ and δ coefficients, while positive values are expected for the η and μ coefficients.

However, caution is required when interpreting the results. If the expansion pattern of Southwest was based upon the operating costs among the airports it chooses to enter, then similar price cuts from incumbent carriers would occur before Southwest enters (Goolsbee and Syverson, 2008). If this would be true, then the dummy variables would indeed detect pre-emptive fare reductions, which would not be caused by the increased probability of Southwest entering that route but by the lower operating costs. To account for such a spurious regression, a proxy for the operating costs is created. Goolsbee and Syverson (2008) argue that if the operating costs were falling at a particular airport, then this would also decrease incumbents’ fares on routes to airports that Southwest does not provide service on. To control for this, two endpoint variables are generated for each route-carrier-quarter which denotes the mean fares to the 100 largest airports where Southwest does not provide service on in that quarter. For example, an American Airlines flight from IAD to LAX in the second quarter of 2005 has control variables operating costs endpoint1 and operating costs endpoint2. Here operating costs endpoint1 is the mean fare of all American Airline routes in that quarter from IAD to any of the 100 largest airports (based on the number of

(18)

18 flights) where Southwest did not provide service on during the studied time period. Fares are normalized by miles between endpoint to create a comparable statistic. Operating costs

endpoint2 is similarly calculated but for all American Airline routes that quarter from LAX to

any non-Southwest airports. This data is available for most of the observations in this dataset, yet some carriers did not fly to any of the 100 other largest airports. This reduced the total observations on threatened routes from 67,998 to 64,926. Although the

observations are decreased, by accounting for the operating costs the quality of the analysis improves.

4. Results

Table II provides the regression results of model (I) on all threatened routes, from this one can verify whether the findings of Goolsbee and Syverson (2008) hold in this extended dataset. Since the dependent variable has been logarithmically transformed, the coefficients of the dummy variables can be interpreted as percental changes after calculating them using (1-exp(coefficient)). The reference period of the model is the 9th_{to 12}th_{quarter before}

Southwest started service in both endpoints. Coefficients of SW_both_endpoint and

SW_flying can therefore be interpreted as percental changes in comparison to the excluded

reference period.

In table II, the Southwest effect on actual antry is clearly portraited. On te the average

incumbents’ fares are decreased with 3.83 percent relative to three years before Southwest threatened entry. For all SW_flying dummies, negative and significant results are shown up to a decrease in fares of 8.47 percent for the 7th_{quarter after Southwest started flying the}

route.

Pre-emptive price cuts on threatened routes are also exhibited from table II, all

SW_both_endpoints dummies have negative coefficients. For the periods after t0 these

coefficients are all significant at the 5 percent level, and show even larger decreases than for the period in which Southwest flies the route. Fare decreases are reported ranging from

(19)

19 2.62 percent for t0 to 12.23 percent for t0+7. Before Southwest started operating at both

endpoints, incumbents were aware of the date on which this would occur for several months. Table II reports significant decreases at the 5 percent level for the 3rd_{and 4}th

quarter before t0 of 1-2 percent and a significant decrease at the 1 percent level for the 1st

quarter before t0 of 3.15 percent. This is in line with Goolsbee and Syverson (2008), who

state that although Southwest announced entry publicly on average 6 months before starting service, industry insiders could find out about this in advance due to Southwest negotiating with the concerning airport over gate leases.

The operating costs proxies for both endpoints show significant and positive results, as could be expected. An increase in operating costs on an airport would increase overall fares from that airport. This would also have a positive effect on the fares on the threatened route, indicating a positive relationship.

Table III reports the regression results of specification (II), which consists of specification (I) with the addition of interaction terms of a tourist route dummy with both

SW_both_endpoints and SW_flying variables. This regression is done on a subset of the data

used in table I, where only the tourist and business routes are included. This enables one to determine whether the pre-emptive price decreases differ between groups. Although table III only reports on one regression, it is presented in four columns. Out of these, the first and third column represent all explanatory variables, while the second and fourth column show their corresponding coefficients. This is done in order to compare each dummy to its’ interaction variable more easily. Since only business and tourist routes are included, the coefficients on the left side of the table, without interaction effects, can be interpreted as the effect of threatened entry on logged fares on business routes. The effect of threatened entry on logged fares of tourist routes can be inferred by adding the coefficient of the interaction term to the SW_both_endpoints or SW_flying coefficient.

The coefficients of SW_both_endpoint, the effect of threat of entry on business routes fares, show predominantly negative coefficients. Most coefficients are significant, albeit not as

(20)

20 significant as for the complete dataset. These results are found before and after Southwest becomes active in both endpoints, similar as for the entire dataset. This suggests that pre-emptive price cutting also occurs when only business routes are taken into consideration. The size of the coefficients are similar to those in table II, whilst the largest significant decrease is smaller in table III than it is in table II. Here significant decreases in fares range from 2.70 percent for t0-6 to 8.74 percent for t0+3, compared to the fares on that route, by

carrier, to three years before t0. On these business routes, there is also evidence for the

Southwest effect, all coefficients of SW_flying are negative. Similar to the

SW_both_endpoints coefficients, fewer significant results are found in this subset in

comparison to the regression of the entire dataset.

Looking at the interaction effects of the tourist dummy with the SW_both_endpoints dummies, only few significant results can be found. These significant coefficients are positive, like the majority of the insignificant coefficients. Since the difference in the effect of threat of entry on tourist routes in comparison to business routes can be interpreted by these interaction terms, positive coefficients are as hypothesized. The estimated effect of threat of entry in t0+3 for business routes is a decrease in fares of 8.74 percent, in

comparison to the reference period. The estimated effect of threat of entry in t0+3 for tourist

routes is an increase of 1.87 percent, compared to the reference period. Here both the interaction term as well as the SW_both_endpoints coefficient is significant at the 1 percent level. Therefore, the threat of entry in t0+3 can be interpreted as having a significant negative

effect on fares on business routes and, a relatively small, significant positive effect on fares on tourist routes. Similar findings at the 5 percent level are only exhibited for t0+4 and t0+9,

providing an arguably low number of significant estimates out of 21 pre-entry dummy variables.

The Southwest Effect on actual entry is hypothesized to be greater for tourist routes than for business routes. This is based on the assumption that the pre-entry effect on fares is greater on business routes than on tourist routes. This would translate into negative effects for the interaction terms with the SW_flying variables. However, table III displays

coefficients with altering sings. The only coefficients significant at the 5 percent level are positive though, for te+2 and te+6. For these quarters, it can be interpreted that both business

(21)

21 and tourist routes display a negative effect of Southwest starting to fly a route, albeit a more negative effect for business routes. For te+2 and te+6, the decreases on tourist routes

were 1.92 and 1.28 percent, respectively, compared to decreases of 6.76 and 6.89 percent on business routes. The control variables for operating costs once more show positive and significant coefficients.

Thus far, the lack of sufficient significant coefficients in table III has failed to explain the rationale behind pre-emptive price cutting. Goolsbee and Syverson (2008) propose another methodology to study a different reasoning behind pre-emptive price cutting: incumbents attempting to deter entry. In line with Dafny (2005), pre-emptive action to deter entry should only be undertaken when there is a threat of entry, and not when entry is known to happen. Goolsbee and Syverson (2008) propose a methodology to test for this, using the occasions on which Southwest started flying immediately upon entering both endpoints of a route. Southwest announces beforehand which endpoint airport it will start service on and thereby also the routes on which they would immediately start flying upon entry. This would have been necessary in order to sell tickets for those routes. Incumbents would not regard this as a threat of entry but as an announcement of actual entry. Goolsbee and Syverson (2008) state that there was no evidence of Southwest not entering a route after announcing to do so. Because of this, the authors were able to test whether pre-emptive price cutting also took place on pre-announced routes. When this behaviour is explained with the theory of locking-in valuable customers, this would occur on both announced and not pre-announced routes (Goolsbee and Syverson, 2008).

Goolsbee and Syverson (2008) recreated their dataset with the same methodology using 223 pre-announced routes and only found significant decreases post-entry. This would suggest that deterring entry is the rationale behind the pre-emptive fare decreases. However, when testing this effect using a model that included interaction effects, they lacked the significance to attribute the incumbents’ motives to deterrence. Since this research makes use of an extended dataset, this methodology will be replicated in order to test whether significant results can be found.

(22)

22 The results of regression model (I) using all pre-announced 523 routes in the dataset are presented in the column (II) of table IV. Table IV also shows the output of Table II in column (I) in order to compare to two. In column (II), no results are shown for t0+1 to t0+12, since

Southwest started flying in t0 for all observations. Significant results are found indicating

that entry deterrence could be the motive behind pre-emptive price cutting. All but one of the SW_both_endpoints coefficients exhibit positive signs, while 3 out of 8 prove to be significant. The results for the Southwest effect of actual entry appears to hold for the pre-announced routes. All but two of the SW_flying coefficients are negative and significant. The size of the coefficients is similar as well. The operating cost control variables, an addition to Goolsbee and Syverson’s (2008) methodology, are once more positive and significant.

To verify whether the difference between pre-announced and threatened routes is significant, the results of a fully-interacted model are portraited in table V. Here the

SW_both_endpoints and SW_flying variables are interacted with the Pre dummy variable,

yielding 1 when entry was pre-announced. From the output in this table we can statistically reject the hypothesis that the coefficients on SW_both_endpoints for pre-announced and threatened routes is equal. For all interacted SW_both_endpoints variables, table V shows positive and significant results. For all but t0-8, these coefficients were higher than the

corresponding negative and significant SW_both_endpoints coefficients. This suggests that fares on routes where entry of Southwest is pre-announced, experience significant increases pre-entry. Fares on threatened routes experience pre-emptive decreases in fares. This finding is in line with the view of price cutting in order to deter entrance. There does not seem to be a significant difference between the Southwest effect of actual entry between threatened and pre-announced routes. Coefficients show alternating positive and negative signs, while only two report significance at the 5 percent level.

To determine whether the results correctly depict the economics behind the data, a critical view needs to be taken on the validity of this research. Some concerns arise from

measurement error. Although the data used is a random sample of 10 percent of all air

(23)

23 travel within the U.S., several adjustments were made to the data. One of these is using only the legacy carriers as used by Goolsbee and Syverson (2008): American Airlines, Continental Airlines, Delta Airlines, Southwest Airlines, Northwest Airlines, TWA, United Airlines and US Airways. These were the most important airlines in the 1993-2004 period. However, since this research extends upon that period, changes in the composition of these legacy carriers could have arisen. Since fares of legacy carriers function as reference, selecting an incorrect reference group could bias the results. Although TWA bankrupted in 2002, it would be incorrect to leave that carrier out for the whole period due to their large market share pre-bankruptcy. Also, after bankruptcy, the company was purchased by American Airlines which is also included in the sample. Since the other incumbent legacy carriers still have relatively high market shares, this should not have affected the results too detrimentally.

Another potential issue is that low-cost carriers have increased their relative market share in comparison to legacy carriers. Southwest now endures competition from other low-cost carriers such as JetBlue and AirTran (Kwoka & Alapin, 2016). Due to time limitations, only Southwest is taken into consideration since it was the dominant low-cost carrier in the studied dataset of Goolsbee and Syverson (2008). However, not taking other low-cost carriers into account can cause a bias in the results. If other low-cost carriers would already have entered a certain route, thereby decreasing the price alike the Southwest effect, then Southwest threatening entry would not have the same effect as on a route where only legacy carriers operate, which is indirectly assumed for all threatened routes. This could have affected the results and will need to be taken into account in future research.

Also, the distinction made between business and leisure customers has its limitations. Not all travellers on tourist routes are leisure travellers and vice versa, due to the unavailability of traveller-specific data, complete separation was not feasible for this research. However, more precise criteria for business and tourist routes are implemented, based on relative treats of both groups. Therefore, these measures aim at providing a more accurate identification strategy.

Another possible problem with regard to internal validity is omitted variable bias. This research partially reduces omitted variable bias with the use of an event study including

(24)

24 carrier-route and carrier-quarter fixed effects. By doing so, prices of a specific carrier on a specific route are compared to prices of that specific carrier on that specific route several quarters ago. Only the three years before t0 and three years after te are taken into account.

By doing so, the regressions only need to explain the differences in prices over the years, eliminating the need to add control variables to the regression that are assumed to stay equal over that time period.

One set of control variables is implemented however, namely a proxy for the operating costs of both endpoints. This is done to mitigate issues concerning reverse causality, similar to Goolsbee and Syverson (2008). If the expansion strategy of Southwest was targeted at airports where operating costs were decreasing, thus prices were falling, price decreases upon entry of Southwest would also emerge from this methodology. The proxy for operating costs allows one to disentangle the effect of decreasing prices due to falling operating costs and due to Southwest threatening entry.

In order to use the correct functional form, the dependent variable has had a logarithmic transformation. As mentioned in the methodology, the mean fare was rejected using a test for normality, using the logarithm of the mean fare mitigates this. To account for

heteroskedasticity and autocorrelation, standard errors are clustered by route-carrier.

5. Conclusion

This research is aimed at determining the rationale behind pre-emptive price cuts upon threat of entry of a low-cost carrier in the airline industry. First it was tested whether evidence for pre-emptive price cutting would hold after the dataset was expanded. Then it was tested whether the price reduction was dependent on the proportion of

price-insensitive business customers. This would comply with the theoretical view of prices being lowered in advance to lock in loyal customers, as suggested in previous research. In

addition, it was tested whether deterrence of entry could provide an explanation. In order to do so, prices of incumbents were compared between routes that were confronted threat

(25)

25 of entry and routes that were confronted with the announcement of entry. Incumbent on these latter routes had no incentive to act if entry deterrence was their motive.

The results show that both the actual entry and threat of entry of Southwest had a negative impact on incumbents’ fares. However, no clear significant difference between routes with a higher proportion of business customers and routes with a lower proportion is found.

Therefore, no statistical evidence is provided for locking in loyal customers as motivation to lower price pre-entry. Results comparing pre-announced routes to threatened routes do exhibit significant results. On threatened routes, where Southwest do not yet fly, and therefore entry could be deterred, prices are significantly decreased. On pre-announced routes, where Southwest is determined to fly, making deterrence of entry unfeasible, prices are not decreasing.

This result builds upon previous research of Goolsbee and Syverson (2008) that found pre-emptive price fare cuts in the airline industry, yet lacked the statistical power to determine the rationale. With an increased dataset, this study has the statistical power to reject the hypothesis that post-entry prices on pre-announced routes are equal to post-entry prices on threatened routes. This suggests that pre-emptive price cuts arise from an incumbents’ intent to deter entry.

When interpreting these results, limitations of this research need to be taken into account. Only the effect of Southwest, as one of several low-cost carriers in the airline industry, is considered. If a route was previously entered by one of these carriers, then prices could have been lowered in advance of Southwest entering a route. Future research may provide a more complete model of the airline industry, including additional low-cost carriers to account for this. The insignificant findings for the difference between business and tourist routes can also be attributed to this being an imperfect method of separating business and leisure travelers. If this distinction is improved upon, for instance using passenger-specific airline data, future research would be more able to verify whether incumbents pre-emptive price cutting depends on customer loyalty.

(26)

26

References

Bennett, R. D., & Craun, J. M. (1993).The Airline Deregulation Evolution Continues: The Southwest Continues. Department Of Transportation Washington Dc.

Borenstein, S. (1989). Hubs and high fares: dominance and market power in the US airline industry. The RAND Journal of Economics, 344-365.

Borenstein, S., & Rose, N. L. (1994). Competition and price dispersion in the US airline industry. Journal of Political Economy, 102(4), 653-683.

Borenstein, S. (2010.) DESCRIPTION OF "MARKET DATA" FILES CREATED BY SEVERIN BORENSTEIN, inferred from http://faculty.haas.berkeley.edu/borenste/mktdata.htm

Brons, M., Pels, E., Nijkamp, P., & Rietveld, P. (2002). Price elasticities of demand for passenger air travel: a meta-analysis.Journal of Air Transport Management,8(3), 165-175.

Cairns, R. D., & Galbraith, J. W. (1990). Artificial compatibility, barriers to entry, and frequent-flyer programs.Canadian Journal of Economics, 807-816.

Dixit, A. (1979). A model of duopoly suggesting a theory of entry barriers.J. Reprints Antitrust L. & Econ.,10, 399.

Dresner, M., Lin, J. S. C., & Windle, R. (1996). The impact of low-cost carriers on airport and route competition.Journal of Transport Economics and Policy, 309-328.

Federal Register (2010), inferred from:

https://www.federalregister.gov/documents/2010/06/28/2010-15605/2010-standards-for-delineating-metropolitan-and-micropolitan-statistical-areas

Gerardi, K. S., & Shapiro, A. H. (2009). Does competition reduce price dispersion? New evidence from the airline industry.Journal of Political Economy,117(1), 1-37.

Goolsbee, A., & Syverson, C. (2008). How do incumbents respond to the threat of entry? Evidence from the major airlines. The Quarterly journal of economics,123(4), 1611-1633.

Lederman, M. (2007). Do enhancements to loyalty programs affect demand? The impact of international frequent flyer partnerships on domestic airline demand.The RAND Journal of Economics,38(4), 1134-1158.

Klemperer, P. (1987). Entry deterrence in markets with consumer switching costs.The Economic Journal,97, 99-117.

(27)

27 Milgrom, P., & Roberts, J. (1982). Limit pricing and entry under incomplete

information: An equilibrium analysis. Econometrica: Journal of the Econometric Society, 443-459.

Morrison, S. A. (2001). Actual, adjacent, and potential competition estimating the full effect of Southwest Airlines.Journal of Transport Economics and Policy (JTEP),35(2),

239-256.

Pitfield, D. E. (2008). The Southwest effect: A time-series analysis on passengers carried by selected routes and a market share comparison.Journal of Air Transport Management,14(3), 113-122.

Selten, R. (1978). The chain store paradox.Theory and decision,9(2), 127-159.

Southwest Media (2017). When did we arrive? Inferred from https://www.swamedia.com/pages/when-did-we-arrive

Vowles, T. M. (2001). The “Southwest Effect” in multi-airport regions. Journal of Air

Transport Management, 7(4), 251-258.

Windle, R. J., & Dresner, M. E. (1995). The short and long run effects of entry on US domestic air routes. Transportation Journal, 14-25.

Dafny, L. S. (2005). Games hospitals play: Entry deterrence in hospital procedure markets. Journal of Economics & Management Strategy, 14(3), 513-542.

Ellison, G., & Ellison, S. F. (2011). Strategic entry deterrence and the behavior of pharmaceutical incumbents prior to patent expiration. American Economic Journal:

Microeconomics, 3(1), 1-36.

Kwoka, J., Hearle, K., & Alepin, P. (2016). From the fringe to the forefront: Low cost carriers and airline price determination. Review of Industrial Organization, 48(3), 247-268.

(28)

28

Appendix

2006 2007 2008 2009 2010 2011 2012 2013 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 t0 te Reference SW_both_endpoints SW_flying

Figure 1: Identification of threatened routes

(29)

29

(1) (2) (3) (4)

Statistic All routes Threatened routes Tourist routes Business routes

Average fare 220.73 216.64 208.90 234.61

Standard deviation (120.29) (103.95) (99.50) (108.58)

Average log(Fare) 5.27 5.28 5.25 5.36

Standard deviation (0.49) (0.45) (0.42) (0.45)

Different routes 2362 968 229 303

Routes immediate flying 523 0 0 0

Threatened routes 968 968 229 303

N 612,369 66,926 17,344 24,671

This table reports summary statistics of the dataset containing airline information of the Bureau of Transportation Statistics from 1993 to 2016. Observations are on route-carrier-quarter level and are divided into four categories. Column (1) provides information on all incumbents’ routes to or from airports where Southwest offered service from 1993 to 2016. Column (2), (3) and (4) only report information regarding the threatened routes, where a route is considered threatened from the moment that both endpoints of a route are serviced by Southwest. Column (2), (3) and (4) only report observations for three years before a route is threatened, up to three years after Southwest starts flying the route. Column (2) provides information on all threatened routes. Columns (3) and (4) provide information on disjunct subsets of column (2). Column (3) entails information regarding threatened routes to or from airports within a Metropolitan Statistical Area above the 85th_{percentile of the ratio of accommodation to non-farming}

income. Column (4) provides information on threatened routes between airports in the 30 most populated Metropolitan Statistical Areas. Different routes represents the number of distinct routes within each category. Routes immediate flying represents routes in which Southwest starts flying immediately once it starts service on both endpoints. Threatened routes represent the routes in which Southwest does not start flying immediately.

(30)

30

Table II: Overall effect of threat of entry on incumbents’ prices

Specification (1) All routes Log(fare) Dependent variable SW_both_endpoints t0-8 -0.0138* (0.00831) SW_both_endpoints t0-7 -0.0138 (0.00841) SW_both_endpoints t0-6 -0.0096 (0.00877) SW_both_endpoints t0-5 -0.0152* (0.00869) SW_both_endpoints t0-4 -0.0212** (0.00914) SW_both_endpoints t0-3 -0.0101** (0.00940) SW_both_endpoints t0-2 -0.0036 (0.00943) SW_both_endpoints t0-1 -0.0320*** (0.00952) SW_both_endpoints t0 -0.0265** (0.01046) SW_both_endpoints t0+1 -0.0762*** (0.01164) SW_both_endpoints t0+2 -0.0559*** (0.01308) SW_both_endpoints t0+3 -0.0708*** (0.01351) SW_both_endpoints t0+4 -0.0496*** (0.01480) SW_both_endpoints t0+5 -0.0583*** (0.01483) SW_both_endpoints t0+6 -0.0781*** (0.01679) SW_both_endpoints t0+7 -0.1305*** (0.01767) SW_both_endpoints t0+8 -0.1026*** (0.01811) SW_both_endpoints t0+9 -0.0784*** (0.01988) SW_both_endpoints t0+10 -0.0608*** (0.02067) SW_both_endpoints t0+11 -0.0983*** (0.02488) SW_both_endpoints t0+12 -0.0478** (0.02138) SW_flyingte -0.0390*** (0.01376) SW_flying te+1 -0.0691*** (0.01263) SW_flying te+2 -0.0868*** (0.01202) SW_flying te+3 -0.0514*** (0.01192) SW_flying te+4 -0.0651*** (0.01233) SW_flying te+5 -0.0606** (0.01263) SW_flying te+6 -0.0824*** (0.01274 SW_flying te+7 -0.0668*** (0.01336) SW_flying te+8 -0.0915*** (0.01347) SW_flying te+9 -0.0705*** (0.01339) SW_flying te+10 -0.0677*** (0.01379) SW_flying te+11 SW_flying te+12 -0.0764*** -0.0504** (0.01412) (0.02221) Operating Costs Endpoint1 0.170*** (0.03122) Operating Costs Endpoint2 0.182*** (0.03235)

Constant 5.384*** (0.05324)

Observations 64,998

0.1018 R2

This table reports the regression results on the mean logged fares of all threatened routes. The SW_both_endpointst0 dummy represent the quarter in which

Southwest offers service in both endpoints but does not start flying. The SW_flyingte dummy represents the quarter in which Southwest starts flying the

(31)

31

Table III: Interaction effects results for tourist and business routes

Specification (2)

Log(fare) Dependent variable (continued)

(2) Log(fare) (continued) Dependent variable SW_both_endpoints t0-8 -0.0289** (0.0126) Tourist*SW_both_endpoints t0-8 0.00587 (0.0179) SW_both_endpoints t0-7 -0.0236* (0.0127) Tourist*SW_both_endpoints t0-7 Tourist*SW_both_endpoints t0-6 Tourist*SW_both_endpoints t0-5 Tourist*SW_both_endpoints t0-4 Tourist*SW_both_endpoints t0-3 Tourist*SW_both_endpoints t0-2 Tourist*SW_both_endpoints t0-1 Tourist*SW_both_endpoints t0 Tourist*SW_both_endpoints t0+1 Tourist*SW_both_endpoints t0+2 Tourist*SW_both_endpoints t0+3 Tourist*SW_both_endpoints t0+4 Tourist*SW_both_endpoints t0+5 Tourist*SW_both_endpoints t0+6 Tourist*SW_both_endpoints t0+7 Tourist*SW_both_endpoints t0+8 Tourist*SW_both_endpoints t0+9 Tourist*SW_both_endpoints t0+10 Tourist*SW_both_endpoints t0+11 Tourist*SW_both_endpoints t0+12 Tourist*SW_flyingte Tourist*SW_flying te+1 Tourist*SW_flying te+2 Tourist*SW_flying te+3 Tourist*SW_flying te+4 Tourist*SW_flying te+5 Tourist*SW_flying te+6 Tourist*SW_flying te+7 Tourist*SW_flying te+8 Tourist*SW_flying te+9 Tourist*SW_flying te+10 Tourist*SW_flying te+11 Tourist*SW_flying te+12

Operating Costs Endpoint1 Operating Costs Endpoint2 Constant -0.00687 (0.0188) SW_both_endpoints t0-6 -0.0274** (0.0139) 0.0176 (0.0192) SW_both_endpoints t0-5 -0.0199 (0.0136) -0.0159 (0.0179) SW_both_endpoints t0-4 -0.0375*** (0.0141) 0.0339* (0.0178) SW_both_endpoints t0-3 -0.0217 (0.0143) 0.0214 (0.0192) SW_both_endpoints t0-2 0.00904 (0.0149) 0.00437 (0.0188) SW_both_endpoints t0-1 -0.0241 (0.0157) 0.0118 (0.0182) SW_both_endpoints t0 -0.0198 (0.0161) 0.0239 (0.0196) SW_both_endpoints t0+1 -0.0792*** (0.0188) 0.0313 (0.0230) SW_both_endpoints t0+2 -0.0690*** (0.0201) 0.0442 (0.0273) SW_both_endpoints t0+3 -0.0915*** (0.0218) 0.110*** (0.0266) SW_both_endpoints t0+4 -0.0529** (0.0244) 0.0772** (0.0312) SW_both_endpoints t0+5 -0.0604*** (0.0234) 0.0420 (0.0330) SW_both_endpoints t0+6 -0.0212 (0.0268) -0.0499 (0.0373) SW_both_endpoints t0+7 -0.0872*** (0.0296) -0.00217 (0.0402) SW_both_endpoints t0+8 -0.0720** (0.0286) 0.0192 (0.0402) SW_both_endpoints t0+9 -0.0812** (0.0329) 0.0899** (0.0451) SW_both_endpoints t0+10 -0.0359 (0.0362) 0.0355 (0.0472) SW_both_endpoints t0+11 -0.0734* (0.0398) 0.0902 (0.0621) SW_both_endpoints t0+12 -0.0156 (0.0370) 0.0642 (0.0493) SW_flyingte -0.0256 (0.0195) -0.0188 (0.0301) SW_flying te+1 -0.0541*** (0.0202) 0.0121 (0.0241) SW_flying te+2 -0.0700*** (0.0195) 0.0506** (0.0231) SW_flying te+3 -0.0376* (0.0208) 0.0508** (0.0226) SW_flying te+4 -0.0223 (0.0210) -0.00965 (0.0225) SW_flying te+5 -0.0276 (0.0213) 0.00370 (0.0223) SW_flying te+6 -0.0714*** (0.0216) 0.0585*** (0.0216) SW_flying te+7 -0.0434* (0.0226) 0.0288 (0.0211) SW_flying te+8 -0.0958*** (0.0229) 0.0250 (0.0211) SW_flying te+9 -0.0495** (0.0223) 0.0252 (0.0206) SW_flying te+10 -0.0234 (0.0247) 0.0272 (0.0221) SW_flying te+11 -0.0357 (0.0250) 0.0283 (0.0217) SW_flying te+12 -0.0629* (0.0366) -0.0362 (0.0540) 0.209*** (0.0476) 0.219*** 5.390*** (0.0497) (0.0902) Observations 41,421 0.1191 R2

This table reports a single regression result on mean logged fares, independent variables are depicted in the first and third column, their coefficients are shown in the second and fourth column. The SW_both_endpointt0 dummy

represent the quarter in which Southwest offers service in both endpoints but does not start flying. The SW_flyingt0 dummy represents the quarter in which Southwest starts flying the route. For this regression, only

tourist and business routes are included. Tourist and business routes are a subset of the total dataset, their classification is explained in the text. The third column shows all interaction effects of the tourist dummy (which equals one for tourist routes) with the SW_both_endpoints- and SW_flying-dummies. Carrier-quarter and carrier-route fixed effects are included in the regression. Clustered standard errors (by carrier-carrier-route) in parentheses *** p<0.01, ** p<0.05, * p<0.1

(32)

32

Table IV: Price response on threatened routes and pre-announced routes

Specification (1) Threatened routes Log(fare) (2) Pre-announced routes Log(fare) Dependent variable SW_both_endpoints t0-8 -0.0138* (0.00831) -0.0185 (0.0113) SW_both_endpoints t0-7 -0.0138 (0.00841) 0.00657 (0.0111) SW_both_endpoints t0-6 -0.0096 (0.00877) 0.0103 (0.0124) SW_both_endpoints t0-5 -0.0152* (0.00869) 0.0286** (0.0121) SW_both_endpoints t0-4 -0.0212 ** (0.00914) 0.0180 (0.0127) SW_both_endpoints t0-3 -0.0101** (0.00940) 0.0205 (0.0130) SW_both_endpoints t0-2 -0.0036 (0.00943) 0.0335*** (0.0127) SW_both_endpoints t0-1 -0.0320*** (0.00952) 0.0278** (0.0136) SW_both_endpoints t0 -0.0265** (0.01046) SW_both_endpoints t0+1 -0.0762*** (0.01164) SW_both_endpoints t0+2 -0.0559*** (0.01308) SW_both_endpoints t0+3 -0.0708*** (0.01351) SW_both_endpoints t0+4 -0.0496*** (0.01480) SW_both_endpoints t0+5 -0.0583*** (0.01483) SW_both_endpoints t0+6 -0.0781*** (0.01679) SW_both_endpoints t0+7 -0.1305*** (0.01767) SW_both_endpoints t0+8 -0.1026*** (0.01811) SW_both_endpoints t0+9 -0.0784*** (0.01988) SW_both_endpoints t0+10 -0.0608*** (0.02067) SW_both_endpoints t0+11 -0.0983*** (0.02488) SW_both_endpoints t0+12 -0.0478** (0.02138) SW_flyingte -0.0390*** (0.01376) 0.00879 (0.0130) SW_flying te+1 -0.0691*** (0.01263) -0.0803*** (0.0136) SW_flying te+2 -0.0868*** (0.01202) -0.0919*** (0.0142) SW_flying te+3 -0.0514*** (0.01192) -0.0845*** (0.0144) SW_flying te+4 -0.0651*** (0.01233) -0.0891*** (0.0149) SW_flying te+5 -0.0606** (0.01263) -0.0980*** (0.0150) SW_flying te+6 -0.0824*** (0.01274 -0.0844*** (0.0154) SW_flying te+7 -0.0668*** (0.01336) -0.0960*** (0.0153) SW_flying te+8 -0.0915*** (0.01347) -0.0578*** (0.0155) SW_flying te+9 -0.0705*** (0.01339) -0.0585*** (0.0160) SW_flying te+10 -0.0677*** (0.01379) -0.0566*** (0.0160) SW_flying te+11 SW_flying te+12 -0.0764*** -0.0504** (0.01412) (0.02221) -0.0697*** 0.2063 (0.0169) (0.129) Operating Costs Endpoint1 0.170*** (0.03122) 0.200*** (0.0436) Operating Costs Endpoint2 0.182*** (0.03235) 0.173*** (0.0437)

Constant 5.384*** (0.05324) 5.109*** (0.0414) Observations 64,998 0.1018 31,609 0.144 R2

This table reports regressions results on the mean logged fares of all threatened routes (1) and of all pre-announced routes (2). The SW_both_endpointt0 dummy represent the quarter in which

Southwest offers service in both endpoints but does not start flying. The SW_flyingte dummy

represents the quarter in which Southwest starts flying the route. For the pre-announced routes t0

equals te, as explained in the text. Clustered standard errors in parentheses *** p<0.01, ** p<0.05,

(33)

33

Table V: Interaction effect results for threatened and pre-announced routes

Specification Dependent Variable

(1)

Log(fare) Dependent variable (continued) Log(fare) (continued) SW_both_endpoints t0-8 -0.0728*** (0.0148) Pre* SW_both_endpoints t0-8 0.0631*** (0.0183)

SW_both_endpoints t0-7 -0.104*** (0.0155) Pre* SW_both_endpoints t0-7 0.123*** (0.0190)

SW_both_endpoints t0-4 -0.0299** (0.0123) Pre* SW_both_endpoints t0-4 0.0711*** (0.0167)

SW_flying te -0.00683 (0.0075) Pre* SW_flying te 0.0274* (0.0133)

SW_flying te+1 -0.0499*** (0.0102) Pre* SW_flying te+1 -0.0116 (0.0152)

SW_flying te+3 -0.0304*** (0.0090) Pre* SW_flying te+3 -0.0368*** (0.0141)

SW_flying te+4 -0.0427*** (0.0092) Pre* SW_flying te+4 -0.0260* (0.0146)

SW_flying te+5 -0.0427*** (0.0093) Pre* SW_flying te+5 -0.0353** (0.0149)

SW_flying te+7 -0.0438*** (0.0096) Pre* SW_flying te+7 -0.0280* (0.0144)

SW_flying te+8 -0.0649*** (0.0093) Pre* SW_flying te+8 0.0215 (0.0139)

Operating costs Endpoint1 0.191*** (0.0256) Operating costs Endpoint 2 0.186*** (0.0263)

Constant 5.167*** (0.0274)

Observations 96,524

R2 _0.098

This table reports a single regression result on mean logged fares, independent variables are depicted in the first and third column, their coefficients are shown in the second and fourth column. The SW_both_endpointt0 dummy

represent the quarter in which Southwest offers service in both endpoints but does not start flying. The SW_flyingt0

dummy represents the quarter in which Southwest starts flying the route. For this regression, all threatened and pre-anounced routes are included. The third column shows all interaction effects of the Pre dummy (which equals one for pre-announced routes) with the SW_both_endpoints- and SW_flying-dummies. Carrier-quarter and carrier-route fixed effects are included in the regression. Clustered standard errors (by carrier-route) in parentheses *** p<0.01, ** p<0.05, * p<0.1

Analyzing pre-emptive price cuts upon threat of entry : Comparing the effects for business and leisure travelers within the airline industry C

Master’s Thesis