Low-Cost Entry, Inter-Firm Rivalry, and Welfare Implications in US Large Air Markets. Hideki Murakami

Transcription

1 009-5 ow-cost Entry, Inter-Frm Rvalry, and Welfare Implcatons n US arge Ar Markets Hdek Murakam

2 ow-cost Entry, Inter-Frm Rvalry, and Welfare Implcatons n US arge Ar Markets 1 Hdek MURAKAMI Abstract Ths paper emprcally analyses the patterns of nter-frm rvalry between low-cost and full servce carrers by carrer and arport bases, and demonstrate welfare mplcaton of CC, usng 1163 US cross-sectonal data of 1998 when CCs were purely no-frlled carrers. Our man fndngs are: (1) that both CC and full servce carrers keep hgher prce-cost margns when CCs enter n the secondary arport, whle especally full servce carrers suffer from low prce-cost margn when CCs enter the same markets, () that total gans of welfare are 5.5 mllon USD for our dataset, and 90% of welfare gans come from the gan n consumer s surplus. CCs cumulatve proft s 4.45 mllon USD, but full servce carrers lost 1.9 mllon USD n total due to the competton by CCs, (3) that CCs sometmes provde unreasonably small (.e, less-than-monopoly) capactes nstead of proft-maxmzng ones when they have no nformaton about own demand curves. Key Words: low-cost carrer, nter-frm rvalry, socal welfare 1 Please do not quote or cte wthout author s permsson. 1

3 1. Introducton There have been many studes on the economc mpact of the US low-cost carrer (CC) s entry nto the ar transportaton market. Morrson and Wnston (1995) emprcally showed that Southwest Arlnes reduces the fare of every carrer t competes wth. Dresner et al. (1996) and Morrson (001) showed the arfare-reducng effect of low-cost entry n the prmary and adjacent markets by ncorporatng CC dummy varables. Ptfeld (005, 008) studed the arfare change after low-cost entry by tme seres analyss. Goolsbee and Syverson (005) and Olvera and Huse (009) studed the entry effect of CCs on the respondng behavor of ncumbents. In studes on nter-frm rvalry among arlnes, Brander and Zhang (1990, 1993), Oum, Zhang, and Zhang (1993), Fscher and Kamerschen (003), and Fageda (006) emprcally estmated the conduct parameters of arlne ndustres n the Unted States (the frst three of four studes) and Span (the last study). Furthermore, Fu, jensen and Oum (006) brdged the studes of CCs vs. full-servce carrers (FSCs) and duopolstc nter-frm rvalry, and also ncorporated the effect of prcng behavor of unregulated-monopoly arports on the competton between CCs and FSCs. In ths study we attempt to brdge the ssues of the low-cost competton and nter-frm rvalry measured by conduct parameters, followng the lne of research ntated by Brander and Zhang (1993). We hghlght not only the duopoly but also the larger markets where more than two carrers operate, Morrson and Wnston (1995), pp

4 whereas prevous studes have dealt wth duopoly cases. Furthermore, we wll demonstrate, from several aspects, not only the consumer welfare but also the total welfare derved from smultaneous demand, arfare, and proft equatons, usng a carrer-specfc dataset wth 1163 data observatons. The next secton derves the conduct parameter and smultaneous equatons theoretcally and then converts them to the econometrc model. In Secton 3 we present the data and n Secton 4 we show the emprcal results and perform several analyses of nter-frm rvalry between CCs and FSCs. In Secton 5 we present welfare mplcatons and Secton 6 s the summary and concludsons.. The model Early studes on estmatng conjectural varaton by CPM (conduct parameter method) to analyze nter-carrer rvalry are Iwata (1979), Applebaum (198), and Bresnahan (1981)(1989) and they are followed by the studes on arlne ndustres lsted up at Secton I. Those studes on arlne ndustry use cross-sectonal data and focus on duopoly, n whch two symmetrc carrers, such as Unted Arlnes and Amercan Arlnes, compete. Our analyss uses, lke many studes do, the cross-sectonal data. Our choce of year s Now not a few CCs, such as ATA 3 and Jetblue, have entered long dstance markets and provde lmted frlls. However, around 1998, CCs stll perssted n ther orgnal busness domans such as provdng 3 ATA was fled n Chapter eleven and qutted ts operaton n Aprl

5 no frlls, servng for the markets of short or medum dstance, ssung no mleage servce and so on. Therefore, we expect economc mpacts such as the degree of arfare-dscountng may be stronger than pckng up the recent year for our analyss. Ths study has the followng dstngushng features: (1) We ncorporate tropoly and larger markets where multple carrers enter as well as duopoly markets, and we cover a wder range of the ndustry than was covered n prevous studes. We analyze the nter-frm rvalry of 1 carrers, and 9 out of the 1 carrers are CCs, () We compute arport-specfc conduct parameters as well as carrer-specfc ones. (3) We estmate not only CCs own capacty-expandng and consecutve arfare dscountng effects but also ther mpact on capactes and arfares of FSCs. (4) We nvestgate how the arfares change after an CC enters as the result of carrers behavors of capacty-expanson, though the number of cases for ths analyss s low. (5) We nvestgate how the output and arfare changed from the frst year to the second year of the new entry by CCs. (6) We derve the total welfare, whereas prevous studes focused on the consumer welfare only. In subsecton -1 we derve the conduct parameters, and n subsecton - we explan smultaneous systems to compute the total welfare effect. 4

6 -1 Conduct Parameter and route-specfc smultaneous equatons Our dataset conssts of 180 duopoly markets, 138 tropoly markets, 56 four-carrer-operatng markets, 19 fve-carrer-operatng markets, 7 sx-carrer-operatng markets, and 4 seven-carrer-operatng markets. Therefore, the route-specfc dataset conssts of 405 and the carrer-specfc dataset has 1163 sample observatons. As for the route-specfc data, the entre data are aggregate, so arfares of ths dataset mean the market-share-weghted average arfares. To estmate the conduct parameters, we use both carrer- and route-specfc dataset. In the carrer-specfc dataset, we derve the conduct parameter assumng the n-frm case wth one CC. 4 The market demand of route s denoted as follows: Q n1 n k j q q k 1 1 q ( j k, k 1,, n 1, 1,, n) (1) where both of the superscrpts k denotes a full servce carrer and j denote an CC. The proft functon of each carrer s denoted as follows: Q TC q k j q p () where TC TC Takng the frst-order condton of (), we have: q p Q q p Q Q Q q MC q 0 (3) We then defne the conduct parameter as (4) and (5): m d k 1 j q q ( m k) m j d k (4) q j dq (5) dq 4 We also have the case of n- full servce carrers and two CCs. Even for ths case we can derve the conduct parameter wthout losng generalty. 5

7 Substtutng (4) and (5) nto (3), respectvely, we obtan: q k 3 where MC MC. p Q q p Q 1 v MC q 0 Q (6) For example, the conduct parameter (4) means the margnal change n the output of other carrers (other FSCs plus carrer j) aganst the margnal change n the output of carrer k. If all of them move n the same drecton and have the same volume, the result s 1 and ths means colluson. If the conduct parameter s 0, (6) equals the frst-order condtons for Cournot competton. If t s 1, the arfare equals the margnal cost. In our model, f the arfare equals the margnal cost of an CC, FSCs would have to ext the k j market, snce MC MC as long as carrers operate at the mnmum effcent scale where average cost equals margnal cost. As n the prevous studes, the equatons (6) can be nverted to (7) by usng the route-specfc arfare elastcty of demand and that the market shares of each carrer p Q MC q s. v 1 (7) p Q s As for the varables and parameters n (7), we already have nformaton on p and s, but the route-specfc margnal cost for each carrer and the route-specfc arfare elastcty of demand are unknown. Therefore, we need to estmate these two unknown varables and parameters n advance to compute the conduct parameters. To obtan v, we use the followng proxy to approxmate route-specfc margnal cost for each carrer, as proposed by Brander and Zhang (1990),(1993) and Oum 6

8 et al. (1993) 5. MC Dst AC Dst (8) AF where AC s the aggregate average cost of carrer, Dst s the dstance of route, AF s the average dstance flown by arlne 6. Many studes on arlne costs, such as Caves, Chrstensen, and Tretheway (CCT, 1984), Gllen, Oum, Tretheway (1990), and Fscher and Kamerschen (003), show that economes of densty exst n the arlne ndustry. Ths means that the total cost functon s strctly concave. Therefore, n (8) ranges between 0 and 1. It s apparent that f s 0, the carrer s margnal cost s proportonal to dstance, whle f s 1, the margnal cost s ndfferent to dstance. Accordng to prevous emprcal research such as CCT (1984), Borensten (1990), Brander and Zhang (1990)(1993), Oum, Zhang, and Zhang (1993), may range between 0.15 and Armanter and Rchard (003) predct that the route-specfc margnal cost of an arlne s just equal to the product of cost per mle and dstance (ths means 0 ) 7. Oum, Zhang, and Zhang (1993) estmated the equaton (9) to obtan. Ths equaton (9) s derved from the frst order condton of carrer s proft functon, that s, a pseudo-nverse supply functon under olgopoly. p AC Dst AF 1 s Dst (9) 5 To estmate the route-specfc margnal cost for each carrer, Fscher and Kamerschen (003) jontly estmate a translog total cost functon and then approxmate the route-specfc margnal cost for each carrer. See Fscher and Kamerschen (003), pp See Brander and Zhang (1990), pp , Brander and Zhang (1993), pp , and Oum, Zhang, and Zhang (1993), pp Armanter and Rchard (003), pp

9 The system-wde conduct parameter can also be estmated n equaton (9). By substtutng estmated nto (8), we can approxmate the route specfc margnal cost. However, we have yet to know the (postve) route-specfc prce elastcty of demand. Therefore, we estmate the Marshallan demand functon. We mght as well smultaneously estmate demand equaton and pseudo-supply equaton (9) at the same tme, but what we need s the route-specfc prce elastcty of demand, not the carrer-specfc one. In other words, the dataset for estmatng (9) s dfferent from the one for estmatng the demand equaton. Consderng the demand and supply system, we detect the arfare on the rght hand sde of the demand equaton and the output on the rght hand sde of pseudo-supply equaton may be correlated wth the error term of each equaton. Therefore, we carry out Hausman specfcaton test to test the null hypotheses that nether arfare nor output s asymptotcally correlated wth the error term of demand and the pseudo-supply equatons, respectvely. p' u p' u Hausman test for demand equaton: H 0 : p lm 0, H 1 : p lm 0 n n q' q' Hausman test for pseudo-supply equaton: H 0 : p lm 0, H1 : p lm 0 n n where p s the average arfare n a market whch s weghted by the number of traffc carred by arlnes, q s the market aggregate traffc, e s the error term of demand equaton and u s also the error term of the pseudo-supply equaton. Under the null hypothess, OS estmator s unbased, consstent, and more effcent than SS estmator, whle under H 1 SS estmator s unbased, asymptotcally consstent, and more effcent than OS estmator. The test result s that the second null hypothess s rejected at 1% level ( ), and the frst null hypothess s rejected at 7.1% ( d. o. f. 1) ( 0.01/ d. o. f. 1) level ( ). Consderng these results, we may have to estmate the structural demand ( d. o. f. 1) 8

10 and pseudo-supply equaton together ether two stage or three stage least squares. In addton, Baley, Graham, and Kaplan (1985) suggest that the market concentraton s also an endogenous varable whch s determned by output, dstance and other exogenous factors such as the exstence of slot controls, and ths s followed by our analyss. Our route-specfc smultaneous equaton system s as follows. [Route-specfc demand equaton] 8 ln m Q ln p ln INC ln DIST ln POP MKT m m3 7 [Route-specfc pseudo-supply equaton] ln p ln Q 0 DCC 6 1 ln DIST ln HERF 3 ln MCD 4 DCC where p s the average arfare at route weghted by market share, INC s the arthmetc average of per-capta ncome of route, DIST s the dstance of route. Ths varable enters both n demand and pseudo-supply equaton, because t controls market demand, and also plays a role of proxy of the margnal cost 9. POP s the arthmetc average of O/D populaton, MKT m s a bnary varable that takes 1 for the market where m carrers compete. For example, MKT 3 s the dummy varable that takes 1 for tropoly markets and zero otherwse. HERF s the Herfndahl ndex, and hgher HERF means that the market s more concentrated. Snce hgh concentraton may lead to strong 8 Snce we wll compute the consumer welfare n Secton 5 by usng ths demand equaton, the functonal form must be carefully chosen. To determne whether the form of each equaton s a smple lnear, a log-lnear or others, we carred out the Box-Cox transformaton. Accordng to the result of R test, we cannot reject the hypothess that the best functonal form s the log-lnear functonal form at 5% level. ( d. o. f. 5) 0.1 ( 0.05 / d. o. f. 5) Snce our data s the route-specfc data, we have to use dstance as the proxy on behalf of the route margnal cost. 9

11 market power, the parameter wll be postve. DCC 1 and DCC are bnary varables that represent the presence of CC(s). The former takes 1 f at least one CC serves the route and 0 otherwse. The latter takes 1 f at least 1 CC serves the adjacent route and 0 otherwse. For example, DCC takes 1 for both the cases as follows: the case to connect two secondary arports such as Southwest s Houston/Hobby-Chcago/Mdway, and the case to connect the prmary and secondary arports such as Ar Tran s Atlanta/Hertsfeld-Chcago Mdway. If a qualty-dstngushed frm enters a market and tres to dstngush tself further, the market output decrease and arfare rses. Therefore, the sgn of DCC 1 and DCC would be postve f both carrers perfectly dstngush themselves from each other. If they are not completely dstngushed, both wll compete and the sgns of DCC 1 and DCC would be negatve. Mason et al. (199) dscuss that f the frm s structure s asymmetrc (e.g., hgh-cost and low-cost), t needs longer tme for the market to reach cooperatve equlbrum than f the frms are symmetrc. Ther dscussons are followed by our varable MCD, whch s the standard devaton of margnal cost. If ths s suffcently large, there may be at least one CC, and carrers n a market wll hardly agree on colludng. Therefore, the parameter wll have negatve effect on the market arfare, and, vse-versa, postve effect on market output. (Dscusson on the use of conjectural varaton to change a major frm s conduct) In a recent analyss of the estmaton of market power usng the conduct parameter method 10

12 (CPM), Fsher and Kamerschen (003) pont out that n a statc envronment, the notons of expectaton and conjectural varaton are not well defned. For example, f we start our analyss by modelng a one-shot Cournot competton and try to estmate the conjectural varaton by CPM, we face the problem that we cannot descrbe the frm s response or any dynamc change n the frm s behavor. As Bresnahan (1989) states, The estmated parameters tell us about arfare- and quantty-settng behavor; f the estmated conjectures are constant over tme, and f breakdowns n the collusve arrangements are nfrequent, we can safely nterpret the parameters as measurng the average collusveness of conduct. 10 Also, Corts (1999) ponted out, CPM estmates of market power can be serously msleadng. In fact, the conduct parameter need not even be postvely correlated wth the true measure of the elastcty-adjusted prce-cost margn, so that some markets are deemed more compettve than a Cournot equlbrum even though the prce-cost margn approxmates the fully collusve jont-proft maxmzng prce-cost margn. 11 Takng these crtcs statements nto account, Fsher and Kamerschen (003) nonetheless stressed the usefulness of conjectural varaton. They nssted, followng Brander and Zhang (1993) and Oum et al. (1993), that one can vew the conjectural varaton as a parameter of market conduct that can capture the whole range of market performance, from perfect competton to monopolstc behavor, 10 Bresnahan (1989), p Corts (1999), p

13 rather than takng t as an ndcator for the frm s expectaton. 1 Fageda (006) also computed the conduct parameters of Spansh ar markets by estmatng the demand and pseudo-supply equaton system usng sem-annual (summer and wnter) data of the years 000 and 001 by 3SS, not stressng the problem wth the dynamc features of conduct parameters but regardng conduct parameters as the set of statc equlbrum. 13 Our analyss n the present essay follows ths lne of research and does not nclude a dscusson of the dynamc aspect of competton. The problems ponted out by Corts (1999) arse when we model the dynamc competton and use the panel data, and Puller (008) suggested a prescrpton for addressng ths problem. We wll dscuss ths ssue further n the future. - Carrer-Specfc Smultaneous Equatons to Derve Total Welfare Next, we try to construct the carrer-specfc smultaneous demand and pseudo-supply equaton system. As we dd n -1, we assume that not only demand but also arfare s also an endogenous varable. As for the new entry and the response by ncumbent carrers, Joskow et al.(1994) fnd that new entrants enter wth low arfare, and ncumbents respond by cuttng arfares to keep ther traffc mantaned. Dresner et al. (1996) estmated the smultaneous and demand- and prce (pseudo-supply) equatons that ncorporate the drectly- and ndrectly-competng CC dummy varables. In order to ascertan the 1 Fsher and Kamerschen (003), p Fageda (006), pp

14 consumer welfare effect, we wll follow the method performed by Dresner et al. and need to estmate the carrer-specfc demand equaton as well as the pseudo-supply equaton n log-lnear forms. Our emprcal model to obtan the effects of low-cost entry on consumers welfare s as follows. For example, a duopoly market contans four equatons (demand and pseudo-supply equatons for two carrers), and the largest market does fourteen equatons, whch means we have seven carrers n the largest market. [Demand equaton] ln 7 N Q 0 1 ln p ln INC 3 ln DIST 4 ln POP 5 MKTm u 1 N 3 N 1 0, 0, 3 0, 4 0, 5 0 [Pseudo-Supply equaton] ln p ln Q D1CR 7 t1 t CRPE( WN) CRPE( HP) ln MC DCC1 DCC 8 D D t CCPE( WN) 14 (13) 9 ln HERF 3 T 1 D1CC1 D1CC CRPE( WN) DCR1 DCR T CCPE( HP) t t, 3 0, 4, 5, 6, 7, 8, 9, 10, 11, 1, 13, 14, 15, 16, t1 D1CR1 t CCPE( WN) 1 6 D 0 D t where u, and are error terms of carrer-specfc demand, and pseudo-supply equatons, respectvely. D 1CC1, D 1CC, D CC1, and CC D are bnary varables that represent the presence of CC(s). D 1CC1 takes 1 f an CC orgnates the prmary arport and 0 otherwse, and D 1CC takes 1 f two CCs exst n the prmary route, for example, the case to 13

15 connect two secondary arports such as Southwest s Houston/Hobby-Chcago/Mdway, and the case to connect the prmary and secondary arports such as Ar Tran s Atlanta/Hertsfeld-Chcago Mdway.. Smlarly, D CC1 takes 1 f an CC enters the adjacent route and 0 otherwse, and D CC takes 1 f two CCs enter. We assume the postve prce elastcty of demand s larger for CCs than FSCs, snce FSCs usually have tools to prevent passengers from swtchng from FSCs to CCs, such as mleage servces. As for the pseudo-supply equaton, the sgn of D 1CC1, 1CC D, D CC1, and CC D would be postve f both carrers perfectly dstngush themselves from each other. If they are not completely dstngushed, both wll compete and the sgns of these four bnary varables would be negatve. 1 can take ether negatve, postve, or zero. If a carrer supples at ts short run margnal cost curve, 1 wll be postve, and f t does on ts declnng average cost curve, t wll be negatve. In addton, f a carrer supples at mnmum effcent scale, t wll be zero. CCPE and CRPE are dummy varables that are ntended for measurng how the arfares of a newly entered CC and ts rval s arfares change over tme. In 1998, Southwest Arlnes entered the Chcago (Mdway) Manchester (New Hampshre), Chcago (Mdway) Brmngham (Alabama), and Chcago (Mdway) St. ous markets. CCPE 1 D ( WN) takes 1 for Southwest n Chcago-Manchester. The superscrpt and the subscrpt stand for the market type and the year from the new entry, respectvely. For example, the superscrpt D means that ths route s a duopoly market and the subscrpt 1 means the frst year of Southwest s entry. CCPE 1 T ( WN) takes 1 for Southwest 14

16 n Chcago Brmngham and Chcago St. ous markets. The superscrpt T means that these two routes are bgger than tropoly. CRPE 1 D ( WN) and CRPE 1 T ( WN) take 1 for all the other carrers n these three routes. The year 1998 s the second year of Southwest s entry n Chcago (Mdway) Jackson (Msssspp) and Amerca West s entry n Dallas/Fort Worth ong Beach. Therefore, CCPE D ( WN) and CRPE D ( WN) take 1 for Southwest and all the other carrers n Chcago Jackson, respectvely. CCPE D ( HP) and CRPE D ( HP) are created n the same way. The effect of these tme dummy varables s removed from the CC dummy varables such as D 1CC1 descrbed prevously. HERF s the Herfndahl ndex, and hgher HERF means that the market s more concentrated, snce hgh concentraton may lead to strong market power, the parameter wll be postve. In addton, HERF and the route-bass margnal cost of a carrer, k MC, are also endogenous varables. The margnal cost s the functon of output and also the ndependent varable of pseudo-supply equaton, so theoretcally we have to use the nstrument varable of margnal cost. To test the null hypothess that nether n HERF ) nor ( k ln MC s correlated wth the error term, we carred out Hausman test for each varable, and reject both of the null hypotheses at 1% level of sgnfcance ( , respectvely). In total, our structural equatons have fve ( 1) and endogenous varables, but we show the demand, and the pseudo-supply, because the estmated results of the rest equatons are out of the scope of ths paper. We also computed the carrer s average cost n 15

17 order to deduce producers profts and loss and derve the total welfare by addng producers profts and consumers surplus. 3 The Data We use the data of the scheduled operatons by cty-par route by frm: they are 1998 cross-sectonal data collected from DB1A, 10% samples of the US domestc flght data. Arfares are fared passenger fares and are descrbed n terms of the US dollar. Omtted are the carrers whch do not have 10% market share n duopoly markets, and 5% share at tropoly or markets more carrers servng. Carrers whose codes are not reported n DB1A (reported as XX) are also omtted, but, for example, a trpoly market wth one XX carrer are not regarded as a duopoly market, snce XX carrer s thought to have compettve effects on others. Flght data are outbound and non-connectng ones from the US sx large arports and ther regons: New York/Newark area, (JFK, aguarda, Newark), Washngton Ronald Reagan (Natonal), Atlanta Hartsfeld, Dallas/Fort Worth area (DFW, and ove Feld), and os Angeles. The source of cost data are from Ar Carrer Fnancal Reports,Form 41 Fnancal Data. Income and populaton data are from Regonal Accounts Data, Bureau of Economc Analyss. We use the Prmary Metropoltan Statstcal Area data (PMSA, an urbanzed county or set of countes that have strong socal and economc lnks to neghborng communtes) for each cty. The descrptve statstcs of contnuous varables are shown n Table 9 n Appendx Emprcal Results 4-1. Conduct Parameter Appendx shows the estmated result of route-specfc demand and pseudo-supply equatons 16

18 by Generalzed two stage least squares (GSS) and three stage least squares (3SS). We obtan wth t-statstcs = by GSS and wth t-statstcs = Both of them are acceptable value accordng to the survey study by Oum, Waters and Yong (199) whch surveys that the prce elastcty of demand of ar travel estmated by cross-sectonal data ranges from to To estmate the taperng effect of route margnal cost and system-wde conduct parameter, we use the result obtaned by GSS, snce by our smulaton we have found that conduct parameters of each carrers fall n the theoretcal nterval (that s, 0,1 ) better when usng than usng Usng the postve value of prce elastcty, we estmate equaton (9) by Non near east Squares. The result s shown n Table 1. Table 1 Estmated result of non-lnear pseudo-supply equaton (9) Parameter SE t-stat. P-Value og lkelhood= , n=1163, Statstcs Maxmum kelhood of estmated ˆ = Accordng to Table 1, the taperng effect of margnal cost s 0.71 whch falls between Oum, Zhang and Zhang (1993) and Borensten (1990). The system-wde conduct parameter s whch s not rejected the null hypothess that 0 at 5% level. Therefore, we conclude that Cournot competton s performed n the US ar markets that experenced low-cost entry. Fgure 1 Market share and Dstrbuton of Conduct Parameter 14 See Table 7 n Appendx 1. 17

19 Share and dstrbuton of Conduct Parameter Fgure 1 shows the dstrbuton of conduct parameter of each carrer whch has at least 0% market share. The horzontal axs s the conduct parameters and vertcal axs s carrer s market share. The fgure shows that carrers whch have large market share (about 65% or more) conduct n accordance wth the economc theory, but frnge carrers do not (ther arfares conduct parameters are ncredbly low or hgh). The low outlers of conduct parameters can be regarded as the outcome of ncreasng too much capacty to try to ncrease market share. Then as a result, equlbrum arfares go down. The hgh outlers may take place when a carrer does not know the demand curve for tself, and lmts ts capacty. In such a case t may happen that ts amount of capacty s smaller than the proft maxmzng level, so s ts arfare hgher than the monopoly level. Fgure demonstrates the average value of each carrer s conduct parameter and ts 95% confdence nterval. Accordng to Fgure, full servce carrers perform Cournot competton or more collusvely than Cournot competton. On the other hand, CC s behave more varously than full servce carrers. The most nterestng sample s Southwest Arlne (WN) s behavor. Its conduct parameters at 18

20 the secondary arports (Dallas ove and Chcago Mdway) are hgher than those at prmary arport (os Angeles), probably because WN can regonally buld more monopolstc stuaton at the secondary arports than at prmary arport, although t must be competng wth full servce carres at the prmary arport. Ths mples that Southwest earns ts proft manly at the secondary arport. Contrarly wth the case of Southwest, Amerca West Arlnes (HP) and Sprt Arlnes (NK),n most cases, competes wth full servce carrers drectly at the prmary arports and ths leads to the result that ts conduct parameters are lowest among CCs. Other CCs such as Ar Tran (F) and ATA (TZ), whch also enter and base secondary arports lke Southwest, have hgher conduct parameters, but ther conduct parameters vary more wdely than other carrers. In some cases n our dataset, ther market shares are very small, so they seems to create new demand that results n low arfares, and to counterbalance the losses generated by low arfares wth hgher arfares than ther average at other thrvng markets. Fgure Average conduct parameters wth 95% confdence nterval 1.5 Conduct Parameter AA CO D NW TW UA US F HP TZ WN1 WN B(95%) Ave. UB(95%) -1 Carrers Note: AA: Amercan Arlnes, CO: Contnental Arlnes, D: Delta Arlnes, NW: Northwest Arlnes, TW: TWA, UA: Unted Arlnes, US: US arways Fgure 3 Average Conduct Parameter at Orgn Arport 19

21 Conduct Parameter AT ORD MDW DFW DA AX NY WDC Arports B(95%) Ave. UB(95%) Note: B and UB means lower bound and upper bound, respectvely. Fgure 3 shows the average conduct parameter and ts 95% confdent nterval of all the arlnes whch orgnate n AT (Atlanta), ORD (Chcago O Hare), MDW (Chcago Mdway), DFW (Dallas=Fortworth), DA (Dallas ovefeld), AX (os Angeles), NY (average of JFK, aguarda, and Newark), and WAS (Washngton Dulles and Ronald Regan). It s apparent that the conduct parameters at os Angeles Internatonal Arport, where multple numbers of CCs enter, are lower than any other arport and those at Washngton D.C., where there are multple numbers of CCs enters, arfares are also low. On the contrary, these results are not followed by the followng two cases: one s the case n whch an CC entered but ts presence s weak, and the other s the case n whch CCs entered the adjacent secondary arport. In the latter case, CCs as well as full servce carrers must have regonal market power and can keep ther conduct parameters hgher than the level of Cournot competton. Ths mples that both full servce carrers and CCs beneft, despte the rvalry between carrers at the prmary arports and those at the secondary arport. 0

22 Fgure 4 Average Conduct Parameter of each arlne at os Angeles Conduct Parameter at AX AA CO D HP NW TW UA US WN Carrer B(95%) Ave UB(95%) Note: B and UB means lower bound and upper bound, respectvely. Fgure 4 shows the average conduct parameter of each arlne wth 95% confdence nterval. On average, conduct parameters of full servce carrers are lower than the level of Cournot competton except for TWA. Two CCs, Amerca West and Southwest, keep ther conduct parameters a lttle hgher than full servce carrers. Ths fact mples that these bg CCs are better off than full servce carrers. Table demonstrates the partal correlatons selected varables used n our analyss. Focusng on the correlaton between the conduct parameter and other varables, we fnd that the conduct parameter s negatvely correlated wth market share and dstance at 1% level of sgnfcance. These results are consstent wth Oum, Zhang, and Zhang (1993): a carrer wth hgh market share tres to expel frnge carrers by expandng ther supply, and then market arfare goes down. Goolsbee and Syverson (005), and Olvera and Huse (009) pont out these capacty-expandng behavors of carrers. We conclude that ths type of competton takes place n long haul markets than n short haul ones. 15 The postve 15 There are many dmensons to the dscusson about the relatonshp between dstance and degree of competton. One may nsst that the competton becomes softer n long-dstance markets because the 1

23 correlaton between conduct parameter and proft s also consstent wth the theory of economcs. Table Partal Correlatons between Conduct Parameter and Other Varables Conduct Parameter Market Share Dstance Herfndahl Index Proft Conduct Parameter - Market Share ** - Dstance ** * - Herfndahl Index ** -0.17** - Proft 0.7** 0.4** ** 0.157** - Note: ** and * means t s sgnfcant at 1% and 5% level, respectvely. 4-. Carrer-Specfc mpact of ow-cost Entry Ths subsecton nvestgates how the mpacts of low-cost entry on full servce carrer s behavors dffer when (an) CC(s) enter(s) n the prmary or secondary arport, and whether the number of CC s affects the full servce carrer s behavors. We smultaneously estmated equaton (1) and (13) by terated 3SS method and the results are shown n Table 8 n Appendx 4. Table 3 s the summary parameters of CC s and ther rval s dummy varables whch are pcked up from Table 8. Table 3 Summary of parameters of CC s and ther rval s dummy varables Parameters Standard error Dfference between one and mult carrer(s) 16 number of surface compettors declnes and because CCs tend to enter short- and medum-dstance markets. Another vewpont s that short-dstance ar markets are normally small markets. Therefore, n many cases we observe more carrers enterng long-dstance markets than short-dstance markets, and ths s supported by the fact that the partal correlaton between dstance and the Herfndahl ndex s -0.17, whch s statstcally sgnfcant at the 1% level (see Table ). Consderng these offsettng effects, t s an emprcal ssue whether the competton s softer n long-dstance markets. 16 The Wald tests that test the hypothess that two parameters are equal are not rejected at all at 5% level.

24 One CC at Prmary Multple CCs at Prmary FSC at Prmary competng wth one CC FSC at Prmary competng wth multple CCs One CC at Secondary Multple CCs at Secondary FSC at Prmary competng wth one CC at Secondary FSC at Prmary competng wth multple CCs at Secondary None at 5% level None at 5% level None at 5% level None at 5% level Table 3 tells us that the quantty competton between CCs and full servce carrers leads to sgnfcant dscount n ther arfares comparng wth the benchmarked full servce carrers. The number of CCs does not statstcally affect CC s and rval s arfares, though t appears t does at the secondary arports. In other words, the frst entry leads to sgnfcantly low market arfares, but the second or later entry does not. These results are almost consstent wth the results of Dresner et al.(1996) whch also ntroduce the dummy varables that reflect the number of compettve carrers. Whle they do not statstcally test the dfference of parameters of these dummy varables, our analyss reveals that the addtonal entres do not affect the rval s capacty-expandng behavors that may affect ther arfares. In the perfectly contestable markets, the number of frm does not affect the capacty or arfare. Snce the frst entry sgnfcantly affects the arfare, we can reject the hypothess of perfect contestablty. : One nterestng fndng s that capacty-expandng and consecutve arfare-reducng effects at the secondary arport look greater than those at the prmary arport, although they are not statstcally sgnfcant. For example, an CC s arfare at the secondary arport s whle that at the prmary 3

25 arport s To explan ths phenomenon, we replace all the carrer-related dummy varables (D1CC1, D1CC, DCC1, DCC, D1CR1, D1CR, DCR1, and DCR) n Table 6 wth each carrer s dummy varable, and re-estmate the carrer-specfc structural equatons usng the same data and the same estmaton method. The result s shown Table 9 n Appendx 5, and the summary of the parameters of carrer dummy varables s shown n Table 4. The methods to ntroduce each carrer s dummy varable are as follows: for example, speakng of Southwest (WN), we ntroduce WN1, WNR1, WN, and WNR. WN1 takes 1 for Southwest operatng n the prmary arport, and WNR1 takes 1 for the full servce carrers that are competng wth Southwest at the prmary arport. Smlarly, WN takes 1 for Southwest operatng n the secondary arport, and WNR takes 1 for the full servce carrers that are competng wth Southwest at the prmary arport. Ths method of ntroducng carrer dummy varables s followed by all the other carrers shown n Table 3, except for Sprts Arlnes and Tower Arlnes. They dd not operate the secondary arports n our dataset, so we have nether NK, NKR, FF, nor FFR. Table 4 Carrer-specfc mpacts on arfare at prmary and secondary arports CC s arfare FSC s arfare Entry by Prmary Secondary Prmary Secondary KP (Kw Int l) (0.179a) (0.5) (0.10) (0.183) TZ (ATA) (0.870b) (0.067) (0.05) 0.03 (0.047) HP (Amerca West) (0.04) (0.54) (0.09) (0.19) F (Ar Tran) (0.047a) (0.146a) (0.043a) (0.091) NJ (Vanguard) (0.090a) (0.147a) (0.069) 0.49 (0.099b) NK (Sprt Ar) (0.146a) (0.17b) 17 The Wald test cannot reject the hypothess that these two values are equal at 10% (Ch-square wth d.o.f.=1 s.01). 4

26 WN (Southwest) (0.050a) (0.04a) (0.033a) (0.09a) FF (Tower Ar) (0.11a) (0.080) Note: Values are the parameters of carrer dummy varables, and standard errors are n parentheses. a, b, and c mean that they are sgnfcant at 1% and 5%, respectvely. Table 4 tells us that Southwest s capacty-expandng behavor at the prmary arports leads to very low arfares, and so does n the secondary arports wth a lttle hgher arfares than n the prmary arports. Full servce carrers also expand ther capactes to cope wth Southwest, and subsequent arfares reducton s greater n the prmary arport than n the secondary arports. These results as for Southwest are qute comprehensve, but are not necessarly followed by other CCs. It s, ndeed, almost common for CCs to operate wth very low arfare at the prmary arports, but even ths prcng strategy s fallacy for Amerca West (HP). Compettons caused by Vanguard Arlnes (NJ), whch used to hub Chcago Mdway Arport, and Ar Tran (F) lead to low arfares both at the prmary and the secondary arports, but the mpact of ther presence on rval s arfares at the secondary arport are much weaker than that of Southwest s. Only does Southwest gve compettve pressure to the full servce carrers at remote arports. The results n Table 4 that shows the entry-mpact s stronger n the prmary arport than n the secondary arports for most CCs seem to contradct wth the results n Table 3. One reason s that the effect of Amerca West, the entry of whch exceptonally leads to lower arfares at the secondary arports than at the prmary arport, s strongly reflected for the estmaton of Table 3. Amerca West can not be regarded as a true CC, because ts average arfares are almost the same as those of typcal full servce 5

27 carrers. However, snce Amerca West bases Chcago Mdway whch s less convenent than O Hare nternatonal Arport n terms of access, t may have to dscount the arfare to offset the nconvenence of access. On the other hand, Southwest and Ar Tran, whch also base Chcago Mdway, are true CCs and ther entry consstently leads to low arfares anywhere, and dscount effects are much more at the prmary arports than at the secondary arports. About relaton between the conduct parameters and arfare level, we fnd that, as may be expected, full servce carrers are better off wthout any entry by CC, but they acheve hgher conduct parameter wth CC(s) at the secondary arport than for the benchmark case where full servce carrers are competng wth each other. (See Table 5). Table 5 Conduct parameters and arfares of FSCs and CC(s) at the prmary and secondary arports Conduct Parameter Full Servce Carrer Upper Bound of CP ower Bound of CP Arfare n Conduct Parameter ow Cost Carrer ower Bound Upper Bound of CP of CP No CC benchmark 413 One CC at Prmary Arport One CC at Seconday Arport Two CCs at Prmary Arport Two CCs at Secondary Arpo Arfare n In Table 5, leavng out the outlers of conduct parameters, we take the average of average of conduct parameters and 95% confdence ntervals for full servce and low-cost carrers for the cases n whch (an) CC(s) enter(s) n the prmary and secondary arports When an CC enters n the adjacent market, full servce carrers may qut competton wthn the prmary arport and try to wn the competton wth the CC. In addton, both full servce carrers 6

28 and the CC can keep, although not strong, the regonal monopolstc power comparng wth the case of head to head competton at the prmary arports. Ths s why conduct parameters at the prmary arport s comparatvely hgh for the case n whch one CC enters n the secondary arport. One nterestng fndng s that CCs at the secondary arport can keep a lttle large prce-cost margn despte ther low-arfares. One reason s that snce the arport charges at the secondary arport are not so expensve as those at the prmary arport, CC can beneft though average arfares are low. Our last fndngs from pseudo-supply equatons are that the degree of competton started by Southwest s sgnfcant n the entry year and s contnued to the second year of entry for some duopoly cases, whereas ths does not hold for the other cases (See the parameters of D CCPE( WN) 1, D CCPE( WN), D CRPE( WN) 1, and D CRPE( WN) ). The factor that causes these four results s the market share an CC gans n the entry year. If Southwest can gan a large market share n the entry year, the ncumbent pays close attenton to the compettor s entry and ncreases ts capacty, and as a result arfares are kept as low as, or much lower than the level n the entry year. On the other hand, f Southwest cannot gan much share n the entry year, the ncumbent expects the competton wll end soon and t actually ends n the second year. Another way of explanng the latter case may be the tact colluson due to mult-market contacts suggested by Bernhem and Whnston (1990), and emprcally tested by Evans and Kessdes (1994) and Morrson and Wnston (1995). Bernhem and Whnston suggested that collusve behavor due to mult-market contacts can take place even though the cost level 7

29 dffers between frms and the products are dfferentated n the repeated game. We wll nvestgate ths dynamc ssue further n a future study. 5. Welfare Effect Our fnal analyss s to compute the consumer s, producer s, and total welfare. Snce we do not have the supply curve under the mperfect competton, we do not compute the true producer s surplus. Instead, we compute the carrer s proft calculated by the carrer s route average cost, carrer s average yelds, and the number of passengers for a carrer. The route average cost s computed by the product of the route dstance and carrer s unt cost (total cost / aggregate RPM). The consumer s surplus s computed by the followng method: we compute the area of trapezods of our demand equaton (11) whch are surrounded by the benchmark arfare, lowered arfare computed from the carrer-related dummy varables, benchmark output and ncreased output due to low-cost competton. Fgure 5 llustrates the change n consumer s surplus n a smple way. The trapezod A s the gan n consumer s surplus due to CC s entry n the prmary arport, and trapezod C s also the gan n consumer s surplus due to full servce carrer s reacton to CCs at the prmary arport (FSC s arfare s hgher than CC s). Smlarly, the trapezods B and D are those for the cases of secondary arports. Snce the market demand s the sum of the demands for each carrer, the total welfare s the sum of the trapezods of CCs and 8

30 those of full servce carrers for the entry n the prmary and the secondary arports (that s, A+B+C+D). 18 Fgure 5 Gans n Consumer s Surplus P Benchmark Prce A B D C FSC s prce at the secondary arport FSC s prce at the prmary arport CC s prce at the secondary arport CC s prce at the prmary arport 0 Carrer specfc demand for CC at prmary arport Market demand q Table 6 demonstrates the gan n consumer s surplus, newly-entered CC s proft, the change n full servce carrer s proft, and the gan n total welfare. Snce we lmt the number of sample observatons by selectng only sx arport groups, the amounts themselves may not be mportant. However, the results mply, n overall, that the gan n consumer s surplus s very large, and CCs also benefts by entry. On the other hand, full servce carrers are losng ther profts due to the low-cost entry, and especally ther losses due to the competton from the adjacent arport are mostly caused by the entry by Southwest. However, snce the losses of full servce carrers are much smaller than the sum of the gan n consumer s surplus and CC s profts, the gan n total welfare s apparently large. Consderng the results n Fgure, full servce carrers are sufferng from the decrease n profts, although they keep the conduct parameters at more than Cournot level, whle CCs are beneftng at smaller conduct parameters than full 18 Snce we ntroduce CC dummy varables, the ntercepts of CCs have to be lower than those of full servce carrers, but for convenence we depct them as shown n Fgure 5. 9

31 servce carrers. Table 6 Summary of welfare effect of CC s entry Gan n Newly entered Change n Gan n Total Due to: Cons. Welf. CC's Proft FSC's Proft Welf. An CC's entry nto Prmary Arport An CC's entry nto Secondary Arport Two CCs' entry nto Prmary Arport Two CCs' entry nto Secondary Arport Sum of the Gan n Welfare Mllon USD (Cons, Welf. = Consumer's Welfare). Total Welf. = Total Welfare 6. Summary and concluson Our fndngs on the compettve behavors of full servce- and low-cost carrers and welfares are as follows: (1) Full servce carrers suffer from the competton wth CC(s) at the prmary arports: they expand ther capactes and ther arfares drop statstcally low and so are the conduct parameters, although the dfferences of conduct parameters are not statstcally sgnfcant. Ths fact mples that full servce carrers do not beneft when CC(s) enters n the same arports. () The conduct parameters of CCs are, on average, hgher than those of full servce carrers, and ths fact mples that CCs do not necessarly performng the cut-throat competton wth thn prce-cost margn but they make reasonable profts n spte of ther low arfares, especally when they stay ther domnant secondary arport such as Southwest s Dallas ove Feld. 30

32 (3) As for the mpact of the frst entry of an CC and the consecutve entry, the frst entry has great mpact on output and arfare, but the second and more addtonal entry have not much mpact than the frst one. (4) Full servce carrers are better off wthout any entry by CC, but they acheve hgher conduct parameter wth CC(s) at the secondary arport than for the benchmark case where full servce carrers are competng wth each other. (5) It s recognzed that an CC s low-arfare entry has capacty-expandng and arfare-reducng effects on ts rval at an ntal stage, but whether these effects are mantaned n the long run seems to be dependent on each carrer. (6) It sometmes happens that competton between CCs and full servce carrers ends n the second year of a new entry, especally when a newly entered CC cannot gan suffcent market share. We need to further nvestgate the ssues of dynamc competton n the future. (7) It s mpled that total gans of welfare due to CCs entres seems to be substantal, and 90% of welfare gans come from the gan n consumer s surplus and the rest comes from the proft of CCs. It s also mpled that full servce carrers do not earn profts although ther prce-cost margns are not necessarly low. Ths fact seems to be due to ther hgh average cost level. Furthermore, our mportant fndng s that CC s conduct parameters sometmes take unreasonably hgh value. In such a case, CCs may not know ther own demand curve and determne capactes and 31

33 arfares the left of proft-maxmzng pont. Ths fact may mply that the conduct parameters wll take wder range than the range that the economc theory assumes. [ ] References Armanter, O, and O. Rchard (003), Exchanges of Cost Informaton n the Arlne Industry, RAND Journal of Economcs Vol.34 No.3, pp Applebaum, E. (198), The Estmaton of the Degree of Olgopoly Power, Journal of Econometrcs Vol.19, pp Baley, E.E., D.R. Graham, and D.P. Kaplan (1985), Deregulatng the Arlnes, MIT Press, pp Bernhem,D. and M. Whnston (1990), Multmarket Contact and Collusve Behavor, RAND Journal of Economcs, 1, 1-6 Borensten, S. (1990), Hub and Hgh Prces: Arport Domnance and Market Power n the U.S. Arlne Industry, RAND Journal of Economcs Vol.0, pp Brander, J. A., and A. Zhang (1990), Market Conduct n the Arlne Industry: An Emprcal Investgaton, RAND Journal of Economcs Vol.1, pp Brander, J. A., and A. Zhang (1993), Dynamc Olgopoly Behavor n the Arlne Industry, Internatonal Journal of Industral Organzaton Vol.11, pp Bresnahan, T. (1981), Duopoly models wth consstent conjectures,. Amercan Economc Revew 71, pp Bresnahan, T. (1989), Emprcal studes of ndustres wth market power. In: Schmalansee R., and R. Wllg,(Eds.), The Handbook of Industral Organzaton, vol. II. Elsever, Amsterdam. Caves, D. W.,. R. Chstensen, and M. W. Tretheway (1984), Economes of Densty versus Economes 3

34 of Scale: Why Trunk and ocal Servce Arlne Costs Dffer, RAND Journal of Economcs Vol.15, pp Corts, K.S. (1999), Conduct parameters and the measurement of market power, Journal of Econometrcs, 88, pp Dresner, M., J. S. C. n, and R. Wndle (1996), The Impact of ow-cost Carrers on Arport and Route Competton, Journal of Transport Economcs and Polcy Vol.30, No.3, pp Evans, W.N. and I. N. Kessdes (1994), vng by the "Golden Rule": Mutmarket Contact n the U.S. Arlne Industry, Quarterly Journal of Economcs, Vol.109, No., pp Fageda, X. (006), Measurng Conduct and Cost Parameters n the Spansh Arlne Market, Revew of Industral Organzaton 8, pp Fscher T. and Kamerschen D. R. (003), Prce-Cost Margns n the US Arlne Industry Usng a Conjectural Varaton Approach, Journal of Transport Economcs and Polcy Vol.37 No., pp Fu, X., M. jesen, and T.H. Oum (006), An Analyss of Arport Prcng and Regulaton n the Presence of Competton between Full Servce Arlnes and ow Cost Carrers, Journal of Transport Economcs and Polcy, Vol.40, No.3, pp Goolsbee, A. and C. Syverson (005), How Do Incumbents Respond to the Threat of Entry? Evdence from the Major Arlnes, NBER Workng Paper Hofer, C., R. Wndle, and M. Dresner (008), Prce premums and low cost carrer competton, Transportaton Research Part E, 44, pp Iwata, G. (1974), Measurement of Conjectural Varaton n Olgopoly, Econometrca, Vol.4, pp Joskow, A, G. Werden and R. Johnson (1994), Entry, ext, and performance n arlne markets, Internatonal Journal of Industral Organzaton 1, pp

35 Mason, C. F., O.R. Phllps, Owen and C. Nowell (199), Duopoly Behavor n Asymmetrc Markets: An Expermental Evaluaton, Revew of Economcs and Statstcs, Vol. 74, No. 4, pp Morrson, S. A. (001), Actual, Adjacent, and Potental Competton: Estmatng the Full Effect of Southwest Arlnes, Journal of Transport Economcs and Polcy, Vol.35, No., pp Morrson, S. A. and C. Wnston (1995), The Evoluton of Arlne Industry, Brookngs Insttuton. Olvela, A.V.M.(008), An Emprcal Model of ow-cost Carrer Entry, Transportaton Research A, 4-4, Olvela, A.V.M., and C. Huse (009), ocalzed compettve advantage and prce reactons to entry: Full-servce vs. low-cost arlnes n recently lberalzed emergng markets, Transportaton Research Part E, 45, pp Oum, T. H., A. Zhang., and Y. Zhang (1993), Inter-Frm Rvalry and Frm-Specfc Prce Elastctes n the Deregulated Arlne Markets, Journal of Transport Economcs and Polcy Vol.7 No., pp Oum, T.H., W. Waters II, and Yong (199), Concepts of Prce Elastctes of transport Demand and Recent emprcal Estmates, Journal of Transport Economcs and Polcy, Vol. 6, No., pp Ptfeld, D.E.(005), Some Speculatons and Emprcal Evdence on the Olgopolstc Behavor of Competng ow-cost Arlnes, Journal of Transport Economcs and Polcy, Vol. 39, No.3, pp Ptfeld, D.E.(008), The Southwest effect: A tme-seres analyss on passengers carred by selected routes and a market share comparson, Journal of Ar Transport Management, Vol.14, No.3, pp Puller, S.. (008), Estmaton of compettve conduct when frms are effcently colludng: addressng the Corts crtque, Appled Economcs etters, February, pp