EUROPEAN CONGRESS OF THE REGIONAL SCIENCE ASSOCIATION VOLOS- 2006 Ttle: Economes of Scale and Spatal Scope n the European Arlne Industry Ref. 905 Authors Manuel Romero-Hernández Unv. De Las Palmas de Gran Canara.Span ITS Unv. Calforna Berkeley mromero@daea.ulpgc.es http://www.personales.ulpgc.es/mromero.daea/ Phone: 34 928 458194 Fax 34 928 458183 C/ Saulo Torón, 4 35017 Las Palmas de Gran Canara. SPAIN Hugo Salgado Unversdad de Calforna-Berkeley Salgado@berkeley.edu 1. INTRODUCTION The man objectve of ths paper s to determne whether the market strateges followed by European carrers are smply a consequence of marketng polces, or f there are also Economes of Scale n costs assocated wth the expanson of producton. By modellng cost performance of European arlnes wth a translog cost functon, we are able to determne the exstence Economes of Densty, Economes of Network Sze and Economes of Spatal Scope for each company. mromero@daea.ulpgc.es 1
Wth these dfferent ndcators we are able to contrbute wth nformaton that can help to explan the behavor of frms, and to antcpate the possble evoluton of the market after the perod consdered n the data set. However, we do not beleve that ths nformaton s the only way to explan the behavor of companes n the market. In addton to cost structure, marketng strateges and demand response are mportant components of the observed behavour of frms. For example, by expandng the number of routes served, companes dversfy ther producton vector; ths, n turn, has cost mplcatons, but even more mportant s how demand responds to ths dversfcaton. When a carrer adds a new route to ts producton vector, t s able to capture customers from other routes who can use ths new route as a leg n ther trp. Users place a hgh value on the tme spent n layovers, and they are wllng to pay for a reducton n total travel tme. 2. THE EUROPEAN DEREGULATION POLICY The thrd package was mplemented between January 1993 and Aprl 1997. The market was completed opened to cabotage n Aprl 1997 for European arlnes. But ths package also gave companes complete freedom to establsh fares and opened doors to purchase ownershp of other European carrers. Carrers responded to the new market condtons wth three man strateges: frst, mergers and acqustons, ether n domestc or external markets; second, settng up low cost carrers; and thrd, arlne allances (for more detals see Chang and Wllams, 2002). mromero@daea.ulpgc.es 2
The man objectves of these three strateges were the consoldaton of domestc markets and the expanson of operaton n new external market. In ths case usng nfrastructure of exstng carrers was a drect way to enter new markets. On the other hand, enterng new markets was subject to the avalablty of a scarce resource slots owned by ncumbents. These strateges allowed companes to expand producton. By expandng the set of products n new markets, companes were able to explot Economes of Scale. Settng up new low cost carrers and acqurng establshed frms dd not always have the expected results; however, arlne allances have been establshed as a stable strategy for most companes. Wth ths polcy companes explot the advantages of denser networks. By addng new routes, companes become more attractve to customers. When customers are decdng whch carrer to fly, they do not only look at the fare, but also at total travel tme, whch s an mportant element n ther decsons. Currently, the arlne market s structured n a hub-and-spoke desgn, whch for many requre that users take more than one flght to arrve at ther fnal destnaton. By flyng wth the same company, users can reduce tme for connectons and avod mssng a connecton. Therefore, even f there are Constant Returns to Scale for carrers, the average socal cost functon declnes when output s rsng (see Mohrng, 1972). Allances allow companes to offer consumers denser routes, share cost and slots wth other carrers, and avod anttrust polces. Some of these allances have mromero@daea.ulpgc.es 3
converged n defntve mergers, as n the recent cases of Lufthansa and SwssAr, or Ar France and Lufthansa. Although there are mportant dfferences between US and European deregulaton processes, t seems that both tend towards the concentraton of producton n the long run. Ths agrees wth prevous results of Scale elastctes obtaned n the lterature. At the begnnng of the US deregulatory process, an ntensve entry of new companes was observed n the market. However, after that ntal perod, equlbrum processes started to work. The result was that some of new entrants companes ether began to leave the market, or to merge wth bgger carrers. The combnaton of two ssues defned ths process: frst, the hub and spoke structure, whch was strengthened by companes durng ths perod and second, the possesson of slots by the major carrers n the man hubs. After deregulaton, US concentraton decreased for longer routes and ncreased for shorter ones (Borensten, 1992). In 1977 the eght largest companes were responsble for 81% of producton; n 1991, over 90%. The hub-and-spoke structure allows companes to serve more arports, wth hgher loads. Companes not only compete n prce, but also through marketng. Hub and spoke networks provde an advantage for bgger companes by ncreasng the number of destnatons served and reducng connecton costs (compared wth a stuaton n whch the user has to change carrers). Other marketng factors also appears mromero@daea.ulpgc.es 4
relevant, ncludng frequent flyer programs and prorty access to reservatons, but overall, the man advantage s held by companes that have slots n mportant hubs. 3. THE MODEL To answer our queston we are estmatng a cost functon for the European arlne ndustry. In order to avod the effect of other ndustres and regulatons on our model, we only nclude European arlnes. It has been prevously reported that the European arlne market s dfferent from the Amercan market n several ways (see for example Ng and Seabrght, 2001). The orgn of these dfferences s, n part, due to the dfferent regulatory hstores of the two markets, as well as ther dfferent carrer szes. The soluton to the dual problem of mnmzng the expendture functon, subject to the transformaton functon, gves us the condtoned demand functon. The condtoned demand functon defnes the specfcaton of the cost functon (see Baumol et al, 1981). Furthermore, because of avalablty of nformaton we are forced to use aggregate data to model the cost performance of carrers. We are also assumng that frms mnmze a lnear expendture functon. Lnearty comes under the assumpton that operators are nput prces takers. The dual relaton between the cost and the transformaton functons allows us to study producton by estmatng the cost functon (McFadden, 1978). Measures of Cost Performance mromero@daea.ulpgc.es 5
How to measure cost performance n ar transport ndustres have been largely dscussed n transport lterature. Dfferent ndces have been proposed and used for ths purpose. The most commonly used have been dfferent measures of Economes of Scale. Nevertheless a number of dfferent ponts of vew and crtques of the more commonly used ndces have arsen. Caves et al (1984) use a translog cost functon wth measures of the aggregated outputs and the number of ponts served by arlnes as the ndcator of Network Sze. Wth ths estmaton the authors are able to estmate two dfferent measures of cost performance that they call Economes of Densty and Economes of Scale. Several others have replcated ths methodology n dfferent case studes, whle others have crtczed the real nterpretaton of the Economes of Scale ndcator because t does not hold the Densty of the network constant when t s expanded (see Xu et al 1994, Jara-Daz and Cortes, 1996, Oum and Zhang 1997). In a recent nnovatve work, Basso and Jara-Díaz (2005) propose the use of an ndcator that avods the crtcsms to the Economes of Scale measurement of Caves et al. (1984). They calculate a measure of Economes of Spatal Scope. In ther paper, these authors propose and use ths ndcator n a cost functon that uses the number of ponts served as an ndcator of Network Sze. By usng the number of routes that an arlne serves as an estmator of the Network Sze, we are able to renterpret the measure of Economes of Scale proposed by Caves et al. (1984). In our case ths ndcator shows how cost responds to a mromero@daea.ulpgc.es 6
proportonal change n the total Tonnes-klometers served by a frm, as well as the number of routes. We consder our measure an approprate ndcator of the effect of Network Sze ncrease on cost because t expands the network whle holdng the average tons-klometers served by each route constant. In ths artcle we use the methodology proposed by Basso and Jara-Díaz (2005), and apply t to a case n whch the number of routes s used as an ndcator of the Network Sze. In addton, by estmatng both total and varable cost functons, we are able to calculate an ndex of excess frm capacty. Ths ndex takes nto consderaton the level of the fxed nputs used by the frms and compares t wth the theoretcal optmum level that s obtaned by comparng the total and varable cost functons. The Economes of Scale and Densty Once we have obtaned cost elastctes for the vector of producton, we are able to obtan the Scale elastcty n order to characterze the technology for the European arlne market (Panzar and Wllg, 1977). In order to compare our results wth those obtaned n the lterature, we mantan the same defnton of Economes of Densty (ED) as n Caves et al (1984). We use the same defnton that these authors used for Economes of Scale, but because we nclude the number of routes, we call ths estmator Economes of Network Sze (ENS). These ndcators are calculated as follows: mromero@daea.ulpgc.es 7
ED CWY (, ) 1 1 = = = CWY (, ) CW (, Y) Y Y π Y Y C( W, Y) y (10) ENS = 1 π + π y N (11) where πy s the cost elastcty gven by the regressor of the estmated equaton, and πn s the regressor for the number of routes served by company. ED ndcates how producton ncreases when all nputs ncrease n a fxed proporton. Ths s under the assumpton of a radal analyss, and therefore holds the proporton of producton vector constant, ENS ndcates how producton ncreases proportonally wth respect to nputs when the number of routes served ncreases proportonally. Ths ndcator mantans the average use of the routes constant, because t holds the total ton-km by route of the dfferent outputs constant. As we are able to estmate the total and varable cost functons, we can also obtan ED and ENS by usng the results of the estmated varable cost functon. In order to do so, we need to make the followng changes: ED CV 1 π Z = (12) π y ENS CV = 1 π Z π + π y N (13) where πz s the cost elastcty of Z, the vector of fxed nputs. mromero@daea.ulpgc.es 8
The Economes of Spatal Scope Basso and Jara-Díaz (2005) proposed a new approxmaton to measure how the cost of an ar carrer changes when t decdes to add a new arport to ts network. They explan that the vector of producton ncluded n the specfcaton of cost functons Y s a vector of aggregate products; t hdes the real vector of produc,ts yj, whch would be the number of passengers (or n our case the weght) and weght of freght carred on each route (or the combnaton between the orgn and the destny j) served by one company. Therefore, when a company serves N P ponts, t s potentally able to serve N P (N P -1) dfferent combnatons between these ponts. Even though the authors do not dscuss the fact that actual use of ths network can be dfferent from the potental number of combnatons, ths fact does not have any effect n ther estmaton method. In our case, because we use the real number of routes served by the arlnes, whch dffer n an mportant way from the potental number of combnatons, we need to make use of an assumpton about how frms decde to use ther potental avalable networks. We solve ths problem by assumng that the number of new routes used when a new arport s added to the network s determned by mantanng the average use of the potental network durng the sample perod. For example, consder the case n whch a company that serves two arports has the followng real vector of producton: Y A =(y12,y21,0,0,0,0). When addng a new arport, mromero@daea.ulpgc.es 9
the vector of potental products would change to Y D =(y12, y21, y13, y31, y23, y32). We consder the queston of whether t s less expensve for the company to produce all the routes together, or to create a new company for the new routes wth the producton vector Y B =(0,0, y13, y31, y23, y32), comparng the cost of producng separately C(Y A )+C(Y B ) wth the cost of producng jontly C(Y D ). The authors apply the concept of Economes of scope to ths dfference and call t Economes of Spatal Scope. Snce the vectors A and B are orthogonal, we can answer ths queston by consderng whether the company has Economes of scope for that partton of the producton (Panzar and Wllg, 1981). In that case, there would Economes of scope f the cost of producng jontly s lower than the cost of producng separately n two frms. The ndcator for ths s as follows: 1 A B D ESS = C( Y ) + C( Y ) C( Y ) D CY ( ) (14) In our case f, ESS >0, then there are Economes of Spatal Scope n the frm wth respect to partton YA, YB of the total producton vector YD. However, the nformaton needed to calculate ESS s ncomplete. We know the aggregate vector of producton for the scenaro A, but not for scenaros B or D. In order to estmate the cost correspondng to these new ponts, we need to have an estmate of the number of routes and the total producton for ponts B and D. One alternatve proposed by Basso and Jara (2005) s to calculate the new aggregate level of producton YD requred to hold the Densty (d) of the actual routes served mromero@daea.ulpgc.es 10
constant. The Densty can be calculated by dvdng the total number of passengers carred on each route by the number of routes served (N R ). d j R y j = (15) N Basso and Jara-Díaz (2005) also obtan the average length of haul (Alh) n order to express the Densty as a functon of the aggregate product, whch s the dependent varable n the estmated cost functon. Y Alh = (16) j y j Substtung, we get d Y = (17) Alh N R Basso and Jara-Díaz (2005) propose two alternatves: smply hold Alh constant, or estmate Alh as a functon of the number of ponts served. They dd not fnd large dfferences n the results when comparng the two cases. In our case we assume that Alh s held constant. By dong so, we are able to calculate the aggregate level of producton for B and D, holdng the Densty of the network constant, as follows: A Y d = = Alh N A R D Y Alh N D R (18) whch mples, mromero@daea.ulpgc.es 11
N D D R A Y = Y (19) A N R Once we have calculated Y D, we can calculate Y B as the dfference between Y D and Y A. Basso and Jara-Daz (2005) develop ths expresson as a functon of the number of ponts served nstead of the number of routes, as we do. Y D A ( NP + 1) = Y A ( N 1) P A (20) 4. RESULTS REPORTED IN PREVIOUS LITERATURE The translog cost functon s the most popular specfcaton used to estmate cost performance of the arlne ndustry. Caves et al (1984), usng panel data from 1970 to 1984, found substantal Economes of Densty, and constant returns to Scale. They reported that both local and trunk carrers show Economes of Densty, even f trunk carrers have an advantage n average cost 1.Although the number of ponts served s smlar, trunk arlnes have hgher load factors and hgher average stage lengths. Caves et al 1984, agree wth other studes that also have found Economes of Scale for US trunk carrers (Keeler 1978 and Whte 1979). Gllen et al (1990) estmate the elastcty of Economes of Scale and Economes of Densty wth a sample of data from US and Canadan arlnes. Data were avalable for the perod from 1964 to 1980. They wanted to test f the Economes of Scale reported 1 In 1978 cost per passenger-mle per trunk arlnes was 7,7 cents, for local carrer was 11,2 cents. mromero@daea.ulpgc.es 12
for US carrers were also present for Canadan carrers, whch are generally smaller than US carrers. Wth fve nput prces and fve products ndexed n a hedonc producton functon, they fnd a Densty elastcty of 1.24. Scale elastcty, ncludng the number cost elastcty for the number of ponts served, was 1.07. Oum and Zhang (1991) 2, Wndle (1991) 3, Kumbhakar (1992) 4, Keeler and Formby (1994) 5 and Baltag Grffn and Rch (1995) 6 have also reported the exstence of Economes of Densty. Usng multvarate regresson and effcency fronter technques, Lu and Lynk (1999) questoned whether the results of the studes carred out for US market would be present, after the deregulaton process. Wth a small database of US carrers, but over a perod that permts them to model performance of carrers several years after US deregulaton, they fnd an average elastcty of Scale of 1.16. They obtaned a negatve, but not sgnfcant, parameter estmate for the number of ponts served. More recently Hansen et al (2001) compared dfferent specfcatons for US carrers and obtaned a consstently elastcty of Scale of 1.2. They use data from eleven quarters between 1995 and 1997 for ten domestc US carrers. Usng multvarate regresson and effcency fronter analyss, Ng and Seabrght (2001) use a very complete data base from 1982-1995 to estmate long run and short cost functons as Gllen et al (1990) dd. They nclude observatons of twelve European and seven US 2 Panel 64-81, Canadan Market. Translog. 3 Panel 70-83. Internatonal. Translog. 4 Panel 70-84, US market. Mcfadden model. 5 Panel 88-90, US market. Translog. 6 Panel 71-86, US market. Translog. mromero@daea.ulpgc.es 13
carrers. They obtaned an average Densty elastcty of 1.19 and an average elastcty of Scale of 1.09. Table 4 shows calculated elastctes of Economes of Densty and Network Sze for the ndustry evaluated wth respect to the mean of the sample, and as defned n secton three. We also provde estmates of these ndcators for each frm, usng the observatons from the last year of avalable data. We provde the probablty of ths value beng less than one, usng one of the methods proposed by Papke and Wooldrdge (2005). The result of Economes of Densty and Economes of Network Sze reported for the European arlne ndustry s comparable to results from prevous studes. The results show, on average, consderable Economes of Densty n the ndustry. By expandng all nputs n the same proporton, producton wll ncrease more than proportonally, so companes are able to reduce total unt costs of producton. By expandng producton and number of routes proportonally, companes total unt costs wll be only slghtly reduced, and usng the total cost functon, we cannot reject the null that Economes of Network Sze exst, on average. The translog cost functon also allows us to evaluate the elastcty of Densty and Scale for each company. Ths gves us the opportunty to explore more accurate nformaton regardng the producton process of each company. We have done so for the last observaton avalable for each company. The results are reported n the second half of Table 4. Almost all companes show ncreasng returns to Densty. The only company for whch elastcty s not sgnfcant greater than one s Vrgn. mromero@daea.ulpgc.es 14
Excludng ths case, the bggest company, Brtsh Arways, shows the smallest elastcty. Therefore t s the company that has most extensvely made use of ts returns to Scale n the last year of the sample. Table 4 also shows that several companes have ncreasng returns to Network Sze. Ths mples that they can reduce ther costs by expandng the Network Sze, and provdes statstcal evdence that these frms cost characterstcs are consstent wth the expanson, merger and allance strateges wdely used by arlnes to expand ther producton. Table 4: Economes of Scale and Network Sze ED p-value p-value ENS (ED<1) (ENS<1) Industry usng TC Functon 1.25 0.0000 1.07 0.0035 Industry usng VC Functon 1.27 0.0006 1.04 0.1269 By Frm (1) Ar France 1.28 0.0002 1.26 0.0005 Altala 1.31 0.0000 1.23 0.0000 Austran 1.49 0.0000 1.02 0.3799 Brtsh Arways 1.10 0.0455 1.18 0.0185 Brtsh Mdland 1.15 0.0326 1.65 0.0141 Fnnar 1.17 0.0009 1.21 0.0046 Ibera 1.50 0.0000 1.28 0.0000 Klm 1.17 0.0014 1.04 0.1134 Lufthansa 1.19 0.0024 1.19 0.0060 Olympc 1.35 0.0000 1.46 0.0012 Sas 1.17 0.0018 1.32 0.0000 Swssar 1.20 0.0000 1.01 0.4095 Tap 1.41 0.0000 1.08 0.1456 Vrgn 1.05 0.2294 0.84 0.9954 (1) Usng Total Cost Functon Table 5 shows the results of the spatal Economes of scope as proposed by Basso and Jara-Daz (2005). Our results show that some of the companes have Economes of Spatal Scope. We can check f our results depend on the assumpton that the proporton of potental routes effectvely used when a new arport s add to mromero@daea.ulpgc.es 15
the network, by smulatng changes n the R parameter. We fnd that the nterpretaton of the results does not change. Table 5: Economes of Spatal Scope ESS Ar France -0.0030 Altala 0.0000 Austran -0.0138 Brtsh Arways -0.0029 Brtsh Mdland 0.1496 Fnnar -0.0020 Ibera 0.0091 Klm -0.0038 Lufthansa -0.0027 Olympc 0.0157 Sas 0.0056 Swssar -0.0094 Tap -0.0087 Vrgn -0.0500 In bold Economes of Spatal Scope. The results n Table 5 show that not all companes would have Economes of scope wth the new vector of producton as a result of addng a new arport to ther network. The results are related to the actual number of routes, as Basso and Jara-Daz (2005) reported n ther paper. Even when the nterpretaton of these results seems to contradct the nterpretaton of Economes of Network Sze prevously dscussed, we thnk that we should consder the nterpretaton of Economes of Scope wth some caveats. Frst, we do not fnd any statstcal propertes of ths ndcator that allow us to nfer whether ths value s statstcally dfferent from zero7. Addtonally, calculatng ths ndcator requres 7 Basso and Jara-Daz (2005) do not dscuss ths ssue. Nevertheless, we are currently developng a way to calculate standard errors for ths ndcator. mromero@daea.ulpgc.es 16
predcton of the cost of an extremely small frm (the frm that represents producton at pont B). Ths nvolves makng predctons about a total cost that s outsde the range of values n the sample, where the predctve power of any econometrcally estmated functon s clearly reduced. 5.CONCLUSIONS Deregulaton mplemented by the European Commsson n the 1980 s and 1990 s radcally changed condtons under whch European arlnes compete n the market. Snce deregulaton, the market has been fully open to cabotage, companes are free to establsh fares, and most have changed from publc to prvate property. Some companes have responded to ths new stuaton by mergng wth other companes (as n the case of Ar France and KLM, or Lufthansa and Swss Ar). However, arlne allances have been the domnant strategy. Wth a database of European arlnes, we have modelled cost performance of companes n order to determne f cost structure has contrbuted to these strateges. Wth ths objectve we have modelled two translog cost functons, total and varable cost. By ntroducng nto the specfcaton of our models the number of routes served by each company, we are able to generate a more accurate measure of the Network Sze, and a renterpretaton of the ndcator of Economes of Network Sze. Ths estmaton also gves us the opportunty to study the exstence of Economes of Scope more precsely. mromero@daea.ulpgc.es 17
For most ar carrers we have found evdence that Economes of Densty and Economes of Network Sze exst n the European arlne ndustry. Our results also show the exstence of Economes of Spatal Scope for some companes n the sample. These results allows us to answer affrmatvely the queston that gudes ths research, and to provde evdence that expanson strateges of frms are related not only to marketng and demand behavor, but also to frms cost structures. Regulatory agences can expect frms to contnue developng strateges that help them to take advantage of the avalable Economes of Scale, whch wll lkely contnue ncreasng the concentraton n the arlne ndustry. mromero@daea.ulpgc.es 18