Do Enhancemens o Loyaly Programs Affec Demand? The Impac of Inernaional Frequen Flyer Parnerships on Domesic Airline Demand Mara Lederman* Massachuses Insiue of Technology This Draf: March 3, 2003 ABSTRACT Consumer loyaly programs such as frequen flyer programs (FFPs) may aler he inensiy of price compeiion beween firms. The increasing marginal benefis ha are buil ino he reward schedules of FFPs give consumers incenives o concenrae heir flying wih a single carrier, raher han choose carriers on a fligh-by-fligh basis. When selecing he airline wih which o accumulae poins, consumers will prefer he dominan carrier a an airpor because i offers he bes opporuniies for earning poins and redeeming rewards. While prior research has no disenangled he marginal impac of FFPs from he oher advanages possessed by dominan airlines, his paper uses a novel empirical approach ha allows for he idenificaion of he marginal effecs of FFPs on airline demand and pricing. In he mid 1990s, domesic airlines increasingly enered ino FFP parnerships wih inernaional airlines. These parnerships allow members of a domesic airline s FFP o earn and redeem ha airline s FFP poins on flighs operaed by parner airlines. While hese agreemens had no direc impac on he qualiy of individual domesic flighs, hey did significanly change consumers earning and redempion opporuniies in domesic airlines FFPs. Using earning and redempion opporuniies as a measure of he value of earning a domesic airline s FFP poins, his paper explois ime-series variaion in he exen and scope of inernaional FFP parnerships o evaluae he economic impac of enhancemens o FFPs. Several findings sand ou. Firs, conrolling for he oher advanages of airpor dominance, enhancemens o an airline s FFP are associaed wih increases in an airline s marke share on individual roues. Second, enhancemens o FFPs are found o have a larger effec on roues ha depar from airpors a which an airline is more dominan and no effec on roues ha depar from airpors a which he airline is no a all dominan. Thus, improvemens o FFPs appear o enhance he advanages of airpor dominance, while conferring no benefi o airlines ha have only a small presence a he airpor. Finally, hese resuls canno be explained by a general ime rend increasing airlines hub advanage over he sample period and are robus o conrolling for he effecs of oher forms of alliance aciviy (such as codesharing) ha may accompany he formaion of a FFP parnership. *I am graeful o my advisors Susan Ahey, Nancy Rose, and Sco Sern. I also hank Rob Clark, Caherine Deri, Avi Goldfarb, Melissa Kearney, and especially Ig Horsmann, for helpful conversaions. Severin Borensein generously provided he DOT DBIA daa. I hank Michelle O Neill a Inside Flyer Magazine, Mary Kandel a OAG, Claire Fairfax a Reed Business Informaion, and Nina Rose a Norhwes Airlines for help in assembling daa. This paper was wrien while I was visiing a he Insiue for Policy Analysis a he Universiy of Torono, whose hospialiy is graefully acknowledged. Financial suppor from he Kellogg School of Managemen and he Social Sciences and Humaniies Research Council of Canada is acknowledged. Conac: mberman@mi.edu.
I. Inroducion Consumer loyaly programs such as frequen flyer programs (FFPs) may aler he inensiy of price compeiion beween firms. Indeed, boh economiss and policy analyss have suggesed ha FFPs migh enhance he marke power of airlines, in paricular, of dominan carriers a hub airpors. The increasing marginal benefis ha are buil ino he reward schedules of FFPs imply ha he marginal value of earning FFP poins increases wih he number of poins already accumulaed. This gives consumers an incenive o concenrae heir flying wih a single carrier, raher han choose carriers on a fligh-by-fligh basis. In selecing he airline wih which o concenrae heir poins, consumers will prefer he dominan carrier a an airpor because i offers he bes opporuniies for earning poins and redeeming rewards. While his relaionship beween airpor dominance and FFPs is well undersood, exising empirical research has no disenangled he impac of FFPs from he oher advanages ha dominan airlines may possess. 1 For example, airlines ha have a dominan posiion a a paricular airpor may boh offer a more aracive FFP and benefi from repuaion effecs or economies of scale in adverising. This paper develops and implemens a novel approach o idenifying he marginal effecs of enhancemens o FFPs on airline demand and pricing. The empirical sraegy is moivaed by he observaion ha a consumer s earning and redempion opporuniies in an airline s FFP include boh he airline s own flighs and he flighs of airlines ha are parners in he FFP. In he mid-o-lae 1990s, domesic airlines increasingly enered ino FFP parnerships wih inernaional carriers. These parnerships allow members of he domesic airline s program o earn and redeem he domesic airline s FFP poins on flighs operaed by parner airlines. While hese parnerships have no direc effec on he qualiy of airlines domesic flighs, hey do significanly increase consumers earning and redempion opporuniies in a domesic airline s program. Because of regulaory and financial barriers ha limi a domesic airline s abiliy o serve inernaional markes, FFP parnerships can expand a domesic airline s program o include many inernaional roues ha i does no serve on is own. The abiliy o earn and redeem on inernaional parners flighs should affec he value o consumers of earning an airline s FFP poins on domesic flighs hrough wo channels. Firs, parnerships expand consumers redempion opporuniies in a domesic airline s FFP by 1 See Borensein (1996), Cairns and Galbraih (1990), and Baneree and Summers (1987) for heoreical models of repea-buyer programs, wih applicaions o FFPs.
increasing he se of reward flighs for which he poins earned on domesic flighs can be redeemed. Second, parnerships increase earning opporuniies by expanding he oal se of flighs on which consumers can earn a paricular airline s poins. This should increase he value of earning ha airline s FFP poins on any given domesic fligh by increasing he likelihood ha consumers will be able o earn addiional poins ha can be combined wih hose poins for an evenual reward. Because he average inernaional fligh is significanly longer han he average domesic fligh, inernaional parnerships allow consumers o reach reward hresholds and earn elie saus afer only a small number of rips. This paper uses earning and redempion opporuniies in an airline s FFP as a measure of he value o consumers of earning ha airline s FFP poins. Changes in FFP parners generae ime-series variaion in earning and redempion opporuniies and, in urn, in consumers value of earning an airline s poins. The changes in parnerships ha occur over he sample period are mapped ino changes in he acual se of flighs on which consumers can earn and redeem an airline s FFP poins. These changes are used o idenify he effecs of FFPs on an airline s marke share and price on domesic roues. The empirical analysis focuses on esimaing he effecs of hese ypes of enhancemens o FFPs on marke share using a discree choice demand model ha includes a measure of earning and redempion opporuniies as an observable produc characerisic. Two such measures of earning and redempion opporuniies are consruced. The firs measures he number of parner flighs on which an airline s FFP poins can be earned and redeemed and hus capures he scale of an airline s FFP. The second measures he number of desinaions ha can be reached on flighs operaed by an airline s inernaional parners and capures he scope of an airline s FFP. In addiion o esimaing he impac on demand, I also briefly explore he impac of changes in earning and redempion opporuniies on he fares ha airlines receive using reduced form fare regressions. Previous work has idenified he advanages of airpor dominance using differences in he size of an airline s operaions across airpors. Borensein (1989) analyzes he effecs of airpor dominance on fares using daa from he hird quarer of 1987. He regresses an airline s fare on a roue on he airline s share of daily originaions a he origin airpor of he roue and he airline s share of passengers on ha roue, conrolling for he overall level of concenraion on he roue and a he airpor. He includes airline fixed effecs o conrol for unobserved differences across airlines and also esimaes a relaive fare equaion, which allows him o conrol for roue fixed
effecs. He finds ha, conrolling for he overall concenraion of a roue, increases in an airline s share of passengers on a roue and a he endpoin airpors allow an airline o charge higher prices. Borensein (1991) esimaes he effec of airpor dominance on an airline s marke share on roues deparing from ha airpor by relaing differences in an airline s marke share on flighs in opposie direcions of a ciy-pair o differences in an airline s dominance a he endpoin airpors. For example, he compares Dela s marke share on roundrips beween Alana and Boson o Dela s marke share on roundrips beween Boson and Alana and relaes his o he difference beween Dela s dominance a he Alana airpor and a he Boson airpor. This allows him o conrol for airline-ciy-pair fixed effecs. Using daa from he second quarer of 1986, he finds ha an airline wih a dominan presence a an airpor is able o arac a disproporionae share of consumers whose rips originae a ha airpor and ha his effec appears o be smaller on ouris oriened roues. 2 In boh of hese papers, however, he effecs of airpor dominance are being idenified from differences in airlines dominance across airpors. The inclusion of airline-ciy-pair fixed effecs only conrols for unobserved facors ha are common o an airline s flighs in boh direcions of a ciy-pair. The esimaes on he airpor dominance variables herefore capure boh he effecs of he airline s FFP and any oher advanages semming from he airline s dominance. As a resul, hese papers do no allow he effecs of FFPs o be isolaed. The empirical sraegy implemened here - using changes in airlines FFP parnerships o generae ime-series variaion in he value of an airline s FFP poins - has several advanages which allow for he idenificaion of he marginal impac of FFPs. Firs, changes in parnerships provide ime-series variaion in he value of earning a domesic airline s FFP poins bu do no affec he oher advanages ha an airline may possess as a resul of is dominance a an airpor. Thus, he marginal impac of enhancemens o FFPs can be idenified, while he oher advanages of airpor dominance can be carefully conrolled for using airline-roue fixed effecs. 3 Second, because he FFP parnerships are formed wih inernaional carriers, he changes in parnerships will affec he value of earning a domesic airline s FFP poins on domesic flighs, bu will have no direc impac on he qualiy of an airline s domesic flighs. Finally, he empirical sraegy 2 The finding ha he effec of airpor dominance is larger on business oriened roues is consisen wih FFPs being a leas one source of his advanage because business ravelers are generally hough o be mos heavily influenced by FFPs. 3 Throughou he paper, I use he erm roue o refer o a fligh beween an origin and desinaion airpor in a paricular direcion (i.e. a fligh from Boson o Alana) and use he erm ciy pair o refer o a fligh beween wo airpors, in eiher direcion (i.e. a fligh from Boson o Alana or from Alana o Boson).
allows for a regression framework ha can conrol for he poenial endogeneiy of he parnerships. Alhough he parnerships hemselves will have no direc effec on he qualiy of an airline s domesic flighs, airlines do choose when and wih whom o form hese parnerships. This raises he concern ha changes in parners may be correlaed wih oher unobserved changes in he qualiy of he airline s produc. To accoun for his possibiliy, I exploi he fac ha he effecs of changes in FFP parnerships should be larger for consumers who are more heavily invesed in an airline s program. For example, consumers in Alana, where Dela is dominan, should have high accumulaed balances in Dela s FFP. As a resul, hey will value Dela s addiion of Air France as a parner more han consumers in Philadelphia, who have primarily accumulaed US Airways poins. However, o he exen ha he formaion of FFP parnerships is correlaed wih oher changes in he qualiy of Dela s produc, hese should no differenially affec he qualiy of is flighs ou of differen airpors. Roues ha depar from airpors a which an airline is no dominan are herefore used o esimae an airline-specific ime rend ha conrols for changes in qualiy which may be correlaed wih changes in he airline s FFP parners, bu no expeced o differenially affec roues deparing from airpors a which he airline is dominan. The daa for his paper come from hree sources. Daa on fares and marke shares are aken from he Deparmen of Transporaion s Daabank 1A. Informaion on he FFP parners of he six larges domesic airlines was compiled using Inside Flyer magazine, a monhly publicaion direced a frequen flyers, and several ediions of he Official Frequen Flyer Guidebook. This informaion is combined wih daa from he Official Airlines Guide (OAG) on he complee direc fligh schedule of each parner airline. These daa are used o deermine he exac changes in earning and redempion opporuniies ha resul from each parnership and o consruc he variables ha measure earning and redempion opporuniies. The paper s main se of resuls esablishes ha, conrolling for he oher advanages of airpor dominance, enhancemens o an airline s FFP, in he form of improved parner earning and redempion opporuniies, are associaed wih increases in an airline s marke share on individual roues. Furhermore, increases in parner earning and redempion opporuniies are found o have a larger impac on an airline s demand on roues ha depar from airpors a which an airline is more dominan and no effec on an airline s demand on roues ha depar from airpors a which i is no a all dominan. Thus, he resuls sugges ha improvemens o FFPs
enhance he advanage of he dominan airline a an airpor, while conferring no benefi a all o airlines ha have only a small presence a he airpor. In addiion, he finding ha enhancemens o an airline s FFP differenially increase is marke share on roues ha depar from airpors a which i is dominan is no simply capuring a more general ime rend ha is increasing airlines hub advanage over he sample period. The resuls are robus o he inclusion of a hub-airpor ime rend, common across all airlines. The resuls are also robus o conrolling for changes over ime in he number of flighs offered by he domesic airlines hemselves, o conrolling for oher forms of alliance aciviy ha may coincide wih he formaion of a FFP parnership, o employing alernaive measures of earning and redempion opporuniies and airpor dominance, and o excluding he larges and mos dominaed airpors from he sample. Finally, esimaes from reduced form fare regressions, presened for descripive purposes and o moivae fuure research on supply-side effecs of loyaly programs, sugges ha one way ha airlines may exercise he marke power gained from enhancemens o heir FFPs is hrough fare increases. The resuls from he fare regressions indicae ha, even conrolling for any oher effecs of airpor dominance on fares, changes in earning and redempion opporuniies are associaed wih higher prices on roues ha depar from airpors a which an airline is dominan. Furhermore, he paern of esimaed coefficiens casually suggess ha he impac of enhancemens o FFPs is larges a he high end of he fare disribuion, consisen wih he commonly made claim ha FFPs are primarily argeed a less price-sensiive business ravelers who purchase higher priced, unresriced ickes. The remainder of he paper is organized as follows. The nex secion provides background informaion on FFPs and on he facors moivaing he formaion of FFP parnerships and oher forms of airline alliances. In Secion III, I presen he empirical model. Secion IV describes he daa and variable consrucion and presens he descripive fare regressions. The resuls of he demand esimaion and a series of exensions and robusness checks are presened and discussed in Secion V. A final secion concludes. II. FFPs, FFP Parnerships and Airline Alliances Since he deregulaion of he airline indusry, much aenion has been focused on he impac of airpor dominance on he compeiiveness of roues deparing from ha airpor. The
reorganizaion of airlines neworks from poin o poin o hub and spoke ha ook place afer deregulaion has resuled in a significan number of airpors ha are dominaed by a single domesic carrier. While here is evidence ha he formaion of hubs resuls in subsanial cosside benefis, here is also evidence ha airlines are able o charge higher prices and capure larger marke shares on roues ha depar from heir hubs (Brueckner, Dyer and Spiller, 1992, and Brueckner and Spiller, 1994 on coss; Borensein 1989 and 1991 on demand; Berry, Carnall, and Spiller, 1997 on boh). The abiliy o offer a more aracive frequen flyer program has repeaedly been cied as one of he mos imporan advanages ha a dominan airline may possess. The firs FFP was inroduced by American Airlines in May of 1981, hree years afer he deregulaion of he airline indusry. 4 The program s goal was o idenify and reward he airline s mos loyal cusomers. Using is compuer reservaion sysem, American compiled a daabase of is frequen flyers and awarded hem bonuses - including free round-rip ickes and free upgrades from economy o firs class - for flying a minimum number of miles per year. Wihin days of he inroducion of American s FFP, Unied Airlines inroduced an almos idenical program of is own. Laer ha year, Dela, Coninenal and TWA followed sui. By 1984, membership in American s FFP was already a wo million, and by 1992 i had reached 16 million. Today, he large U.S. FFPs (American, Unied and Dela) have over 20 million members each, hough many members are, of course, shared. Worldwide, here are now more han 70 differen FFPs which, combined, have over 100 million members. 5 FFPs award consumers frequen flyer poins for each fligh purchased from an airline. Accumulaed FFP poins can be redeemed for rewards, usually free ickes or free class upgrades wih he airline. FFP reward schedules are srucured such ha a minimum number of poins mus be earned before any reward can be redeemed (for mos domesic airlines, his is 25,000 poins), afer which he value of rewards increases non-linearly wih he number of poins required. In addiion o his non-lineariy in he reward schedule, he airlines have creaed elie programs ha award elie saus o consumers who fly a minimum number of miles wih he airline in a given calendar year. Mos elie programs have hree iers, wih qualificaion for each ier requiring an increasing number of miles flown (usually, 25,000, 50,000 and 100,000 4 See hp://www.frequenflier.com/ffp-005.hm and Mason and Barker (1996) for more on he hisory of FFPs. 5 hp://www.frequenflier.com/ffp-005.hm
respecively). 6 Each ier eniles a raveler o an increasing amoun of preferenial reamen, including airpor lounge access, firs-class check-in, free class upgrades, an exclusive phone number for bookings, expanded award availabiliy, and mileage bonuses on fuure flighs. Because he elie programs enile a consumer o preferenial reamen on all flighs aken wih he airline in he year of qualificaion (and, in some cases, flighs on he airline s parners as well), hey creae large, discree increases in he value of earning addiional FFP poins as one approaches he ier saus hresholds. These non-lineariies give consumers an incenive o accumulae all of heir poins in a single airline s FFP. If consumers fly o muliple desinaions, or if hey are uncerain abou where hey will need o fly o, hey will prefer an airline ha provides he opion of earning is FFP poins on flighs o many differen ciies. An airline wih a large roue nework, paricularly ou of a consumer s home airpor, will herefore offer he bes opporuniies for accumulaing poins. I will also offer a larger selecion of poenial reward desinaions. For his reason, airlines ha have a dominan posiion a a paricular airpor are hough o have an advanage in he use of FFPs. 7 Once consumers are invesed in an airline s FFP, any fligh no aken wih ha airline represens forgone FFP poins. As a resul, in order o induce consumers o purchase heir produc, carriers ha are no dominan a an airpor mus compensae consumers, wih a lower price or higher qualiy, for he forgone FFP poins. As described in he inroducion, parnerships wih inernaional carriers have become an imporan componen of domesic airlines FFPs. These parnerships allow members of a domesic airline s FFP o earn and redeem ha airline s poins on eligible flighs operaed by parner airlines. FFP parnerships wih inernaional airlines have been formed as par of a larger wave of alliance aciviy beween domesic airlines and inernaional carriers, which has aken place over he pas en years. In he pas decade, he demand for inernaional ravel has grown subsanially. In response, domesic airlines have sough o creae large inernaional hub-andspoke sysems, similar o hose ha hey have esablished domesically. However, regulaory and financial barriers preven domesic airlines from seing up inernaional neworks and, in paricular, from seing up inernaional hubs from which hey could serve a large number of 6 Only poins earned on purchased airline ickes coun owards elie-level qualificaion. Poins earned from non-airline parners (such as renal car companies) or hrough credi card purchases do no coun. In his way, he elie level programs ruly reward he airline s mos frequen flyers. 7 This does no imply ha airlines ha are no dominan canno use FFPs. Raher, he poin is ha airlines wih a dominan posiion a an airpor are, in general, able o offer much more aracive FFPs.
inernaional desinaions. 8 Furhermore, resricions due o naional ownership laws prohibi ourigh mergers beween domesic and inernaional airlines. Unable o esablish large inernaional neworks on heir own or hrough mergers, domesic and foreign airlines have urned o alliances and oin operaions as a way o mimic he behavior of a single airline and capure he benefis of large-scale inernaional neworks. 9 Allying domesic and foreign carriers coordinae heir aciviies in various ways, in an aemp o enhance he qualiy of inernaional iineraries ha involve ravel on boh he domesic and he foreign airline ( inerline service ). FFP parnerships have emerged as par of hese ypes of alliances, improving he qualiy of inerline service by allowing consumers o earn heir domesic airline s FFP poins on segmens ha are flown on inernaional carriers and o redeem heir domesic airline s FFP poins for reward ickes ha involve ravel on inernaional carriers. In addiion o reciprocal FFP privileges, many alliances also involve broader forms of cooperaion. Alliances ypically involve codesharing, whereby an airline will marke and sell ickes under is own wo-characer designaor code for a fligh ha is operaed by anoher carrier. Wih codesharing, a rip ha involves connecing flighs operaed by wo differen airlines will appear on ravel agens screens and on consumers ickes as being operaed by a single carrier. 10 Codesharing alliances may also involve shared erminal space, coordinaed baggage handling and oher ypes of oin operaions. In some cases, hey may also include anirus immuniy, whereby he airlines are allowed o collaborae on fares for roues on which hey codeshare, wihou being in violaion of U.S. laws. Mos recenly, airlines have begun o form large-scale mulilaeral alliances which involve codesharing, coordinaed service, reciprocal FFP privileges, and, in some cases, anirus immuniy beween all member airlines. 11 While FFP parnerships wih inernaional airlines have been formed in response o changes in inernaional ravel demand, hese parnerships have he addiional effec of increasing he value of earning a domesic airline s FFP poins on domesic flighs hrough heir expansion of consumers earning and redempion opporuniies. To see how FFP parnerships can change consumers earning and redempion opporuniies, consider he earning and 8 For informaion on inernaional aviaion regulaion, see U.S. General Accouning Office (1995) and (1998). 9 See Gellman Research Associaes, Inc. (1994) for a deailed discussion of inernaional airline alliances. 10 In indusry erms, codesharing urns inerline service, which involves ravel on wo differen airlines, ino online service, which involves ravel on only one airline. There is subsanial evidence ha consumers prefer online service. 11 There are currenly four such alliances ha involve a U.S. domesic carrier. Sar Alliance, which involves Unied, oneworld which involves American, SkyTeam, which involves Dela, and he Norhwes-Coninenal-KLM alliance.
redempion opporuniies wih and wihou parners, for members of Dela s FFP wishing o earn or redeem Dela poins on flighs o France. In he hird quarer of 1997, Dela iself offered direc flighs from hree U.S. ciies o Paris and from one U.S. ciy o Nice. Air France, which became a Dela parner in ha quarer, offered direc flighs beween eigh U.S. ciies and Paris and offered flighs ou of Paris o 14 oher ciies in France. To esimae he effecs of FFPs on marke shares and prices, his paper explois changes in airlines inernaional FFP parners over ime. The measures of earning and redempion opporuniies consruced do no accoun for wheher a FFP parnership is par of a more exensive airline alliance. For he purpose of deermining how a given parnership increases he se of flighs and desinaions ha are available o members of a domesic airline s FFP, all FFP parnerships can be reaed equally. However, he fac ha he formaion of some FFP parnerships may coincide wih he sar of oher alliance aciviy beween he parner airlines raises he possibiliy ha he laer could be affecing he resuls. Because his paper invesigaes he effecs of inernaional alliances on domesic airline demand, here should be no direc effec of codesharing or any oher ype of alliance aciviy beween he domesic airline and is inernaional parner on he qualiy of he domesic airline s flighs on domesic roues. There are, however, possible indirec effecs. I discuss his possibiliy in secion V and presen a specificaion ha addresses he coincidence of codesharing and FFP parnerships. Finally, while FFP parnerships have been formed primarily wih inernaional carriers, o a much lesser exen, domesic airlines have also enered ino FFP parnerships and codeshare alliances wih each oher. Unil 1998, codeshare agreemens beween domesic airlines were largely formed beween maor domesic carriers and regional commuer airlines ha ranspored passengers, usually from smaller communiies, o he ciies served by he maor carriers. In addiion, several maor carriers formed codeshare agreemens in a limied number of markes. FFP parnerships beween domesic airlines were formed eiher as par of hese codeshare agreemens or beween maor carriers and airlines ha specialized in a paricular domesic region, such as Alaska Airlines or Hawaiian Airlines. However, a he end of 1998 and beginning of 1999, he six maor domesic airlines enered ino hree large-scale markeing alliances. Coninenal and Norhwes, Unied and Dela, and American and US Airways each formed an alliance ha allowed members of heir FFP
programs o earn and redeem heir FFP poins on heir parner s flighs. 12,13 In conras o inernaional parnerships, which affec domesic demand and price only by expanding a domesic airline s FFP o include inernaional desinaions, parnerships wih domesic airlines may boh expand he domesic airline s nework domesic airlines have differen regional fligh paerns - and increase airline subsiuabiliy parners compeing on he same roue are now closer subsiues since each airline s poins can be earned and redeemed on he oher. Because domesic parnerships have his addiional subsiuion effec, an invesigaion of heir effecs is lef for fuure research. However, he fac ha he domesic parnerships are formed during he sample period raises he possibiliy ha hey may also be affecing he value of earning he airlines FFP poins and herefore may be affecing he resuls. This possibiliy is addressed in Secion V. III. FFPs and Airline Demand III. A. Demand To analyze he impac of enhancemens o FFPs airlines marke shares, I use a discree choice model of demand. A roue, r, is defined as an origin airpor and desinaion airpor, and a marke, m, as a roue-quarer combinaion. A produc,, wihin marke m, is defined as a carriericke ype combinaion for example, direc service on American Airlines. 14 The sample is resriced o direc or online ickes (connecing flighs where boh legs of he rip are raveled on he same carrier) on 12 domesic airlines. Connecing flighs by he same airline hrough differen connecing airpors are considered o be he same produc. Each airline herefore offers a mos wo producs in any marke. Following Berry (1994), consumer i s uiliy from purchasing produc, for roue r, in quarer is wrien as, (1) where u = x β + αp + ξ + v ir r r x r is a vecor of observed characerisics for produc, r ir p r is he price of produc, ξ r is a vecor of unobserved (o he economerician) characerisics for produc, and v ir is an 12 For more informaion on he domesic alliances, see U.S. General Accouning Office (1999). 13 Of he hree, he Coninenal-Norhwes alliance is he only one o involve codesharing and is sill in effec. The American-US Airways alliance has been severed and he Unied-Dela alliance seems likely o be severed as a Unied-US Airways alliance has recenly been approved by he governmen. 14 Alhough is used o refer o an airline-icke ype, I will someimes be somewha loose wih noaion and use o refer o us he airline.
idiosyncraic error erm. To allow increases in he prices of all airline producs o reduce aggregae demand, I inroduce an explici ouside good. Following Berry, Carnall and Spiller (1997), I assume ha he number of poenial rips for roue r in quarer is equal o he geomeric mean of he populaions of he origin and desinaion ciies of roue r. 15 Because he inside goods include direc or online service by 12 domesic airlines, he ouside good for roue r include consumers who fly roue r in quarer using direc or online service by an airline no included among hese welve, consumer who fly roue r in quarer using inerline service and consumers who choose no o fly roue r in quarer. 16 Each consumer is assumed o choose he produc from he ( J r + 1) producs available in marke r in quarer hese are J r airline producs and he ouside good (=0) - ha maximizes his uiliy. Le δ = x β + αp + ξ denoe he mean valuaion of produc across all consumers while v i represens he difference beween consumer i s valuaion of produc and he mean valuaion. 17 The disribuion assumed for v i deermines he choice probabiliies and subsiuion paerns. As in Berry, Carnall and Spiller (1997), I adop a nesed logi framework and group all inside goods ino one nes and he ouside good ino anoher. 18 This allows for correlaion in v i across he inside goods, allowing hem o be closer subsiues wih each oher han hey are wih he ouside good. 19 Formally, his can be represened as, (2) where vir = ζ + ( 1 σ) ε ir ir ζ ir is a random effec ha is common o all inside producs and has a disribuion funcion ha depends onσ, wih0 σ 1, and ε ir is an idenically and independenly disribued idiosyncraic ase for produc. As σ approaches one, he wihin group correlaion of uiliy levels goes o one and as σ approaches zero, he wihin group correlaion goes o zero. 15 Berry, Carnall and Spiller (1997) assume ha he size of he marke is proporional o he geomeric mean of he endpoin ciies populaions and esimae he facor of proporionaliy. I do no inroduce a facor of proporionaliy. 16 For he roues ha I consider (see secion IV.B below), hese 12 airlines, on average, carry more han 95% of roundrip passengers who fly direc or online. For hese roues, over he sample period, inerline ravel is rare. 17 For ease of noaion, I someimes omi he roue and ime subscrips in he descripion of he model. 18 Berry, Carnall and Spiller (1987) also include addiional consumer heerogeneiy by allowing wo ypes of consumers who have differen valuaions of he produc aribues. I do no include addiional modeling of consumer heerogeneiy in his paper because he focus here is on esimaing he average effecs of FFPs bu am pursuing i in separae research. 19 For an inuiive discussion of he nesed logi model, see Sern (1995).
Assuming a generalized exreme value disribuion for producs in group g, he marke share for produc can be shown o be, ε ir and leing J g denoe he se of (3) s δ (1 σ ) e ( δ, σ ) = σ (1 D σ ) g [ Dg g ] where, for a produc in group g, (4) D g = e J g δ ( 1 σ ). Berry (1994) shows ha he sysem of marke share expressions in he nesed logi can be analyically solved for he mean uiliy levels, δ ( s, σ ). Subsiuing δ = x β + αp + ξ and seing he mean uiliy of he ouside good o zero, he following esimaing equaion for he nesed logi can be derived, (5) ln( s ) ln( s0 ) = x β αp + σ ln( s / g) + ξ where s / is produc s wihin-group share (his is produc s share of all passengers who fly g roue r in quarer on one of he 12 inside airlines) and produc ) is he economeric error erm. ξ (he level of unobserved qualiy for III. B. Incorporaing FFPs characerisics, I now urn o he issue of how o incorporae FFPs ino he vecor of observable produc x r, in order o esimae he impac of enhancemens o an airline s FFP on is demand. The FFP poins available on a fligh are clearly a characerisic ha consumers consider in deciding which airline produc o purchase. However, here remains a quesion as o wha is he appropriae way o measure frequen flyer poins as a produc aribue. Simply couning he number of poins available on a fligh would no be adequae because his is generally equal o he disance of he fligh and would herefore generally be he same for all airlines offering service on a roue. However, he value of a given number of poins may be very differen depending on he airline from which hey are earned and he airline s presence a he consumer s home airpor. Moreover, frequen flyer poins do no affec consumers uiliy from consuming produc in he same way ha, for example, having a direc fligh does. Raher, consumers value
frequen flyer poins because when a large enough number of poins have been accumulaed, hey can be redeemed for rewards (and may enile he consumer o he benefis of ier saus). Thus, he value of frequen flyer poins is derived from he value of he rewards for which hey can be redeemed and he probabiliy ha a consumer will be able o accumulae enough poins o redeem a reward (or achieve ier saus). While his value is no direcly observable, variables ha affec his value can be idenified and measures of hese can be included in he vecor x r. 20 Specifically, consumer i s value of earning airline s frequen flyer poins in quarer is assumed o depend on hree facors: (1) r i : Consumer i s redempion opporuniies in airline s FFP in quarer. (2) e i : Consumer i s earning opporuniies in airline s FFP in quarer. (3) b i : Consumer i s accumulaed balances in airline s FFP in quarer. Consumer i s value of earning airline frequen flyer poins in quarer is denoed, v ( r, e, b ). Redempion opporuniies, r i, measure he rewards ha can be redeemed wih airline s FFP poins. Because mos poins are redeemed for free ravel, he erm redempion opporuniies is used here o refer o he se of flighs on which an airline s FFP poins can be redeemed. Improved redempion opporuniies hrough he addiion of flighs and desinaions should increase he value of earning airline s FFP poins. Earning opporuniies, e i, measure he ways in which addiional poins in airline s FFP can be earned. FFP poins can be earned on flighs and from he airline s non-airline parners, such as renal car companies and affiniy credi cards. The focus here is on poins earned on flighs. 21 Improved earning opporuniies should increase he value of poins ha can be earned on a domesic fligh oday by increasing he probabiliy ha, regardless of where a consumer will need or wan o fly, he will be able o earn addiional FFP poins wih ha airline which could be combined wih he poins earned oday for an evenual reward (or o achieve ier saus). Because of he increasing marginal value of rewards, on average, having higher accumulaed balances, b i, in an airline s FFP should increase he value of earning addiional poins. Furhermore, he higher are a consumer s curren balances wih airline, he larger i i i i 20 Morrison and Winson (1989) include a measure of FFPs in a mulinomial logi demand model. They use he number of FFP poins available ineraced wih he number of desinaions served by he airline. This capures he idea ha he value of FFP poins increases wih he number of ciies served by he airline. In heir sample period, no all he airlines ha hey considered operaed FFPs. 21 Because only poins earned on flighs coun owards qualificaion for elie saus, business ravelers, in paricular, may care he mos abou fligh earning opporuniies.
should be he impac of an increase in earning or redempion opporuniies in airline s FFP, because he closer is he consumer o reaching he nex reward hreshold. 22 By changing r i and e i, changes in FFP parners provide ime-series variaion inv i. 23 Secion IV describes in deail he variables used o measure earning and redempion opporuniies. To allow enhancemens o an airline s FFP o be valued more highly by consumers wih higher balances in he airline s FFP (higher b i ), he effecs of earning and redempion opporuniies on demand are allowed o vary based on he airline s presence a an airpor. Alhough individual balances are no observable, I assume ha b i is, on average, higher for consumers a airpors a which airline has a larger presence. A measure of earning and redempion opporuniies and an ineracion beween his variable and a measure of he airline s local dominance are herefore included in he vecor x r as a way o capure FFP poins as a produc aribue. III. C. The Empirical Sraegy Though embedded in a discree choice model of demand, he empirical sraegy being used o idenify he impac of changes in earning and redempion opporuniies on airlines demand is essenially a differences- in-differences approach. The sandard differences- indifferences framework uses a reaed and a no n-reaed group and a pre-reamen and posreamen period o esimae he effec of a reamen on a reaed group. The pre-reamen period is used o conrol for unobservables ha differenially affec he oucomes of he reamen group and he non-reaed group is used o conrol for unobservables ha differenially affec oucomes in he pos-reamen period. Applying his erminology o he sraegy used here, I am esimaing he effec of changes in earning and redempion opporuniies ha resul from changes in FFP parners (he reamen) on an airline s marke share on roues ha depar from airpors a 22 vi can also be inerpreed as he swiching cos ha consumer i incurs if he chooses an airline oher han airline. The swiching cos is equal o he value of he FFP poins ha would have been earned had airline been chosen. The higher are consumer i s balances in airline s FFP or he beer are consumer i s earning and redempion opporuniies in airline s FFP, he higher is he value of he forgone FFP poins and he larger he swiching cos. The swiching cos lieraure ofen uses FFPs as an example of arificial or endogenous swiching coss. See, for example, Klemperer (1987a), (1987b). 23 Changes in earning and redempion opporuniies on he domesic airline iself would also affec v. A casual look a he daa i suggess ha here are no maor changes in he domesic airline s own fligh schedules over his period. Noneheless, in Secion V, I presen a specificaion ha explicily conrols for changes in earning and redempion opporuniies on he airline iself, in addiion o changes in parner earning and redempion opporuniies. Changes in earning and redempion opporuniies on he airline iself are found o have no effec on demand in his period and he impac of changes in parner earning and redempion opporuniies is unchanged.
which i is dominan (he reaed group). This sraegy requires: (1) conrolling for unobservable produc characerisics ha differenially affec he qualiy of flighs ha depar from airpors a which an airline is dominan, and (2) conrolling for unobservable facors ha differenially affec he qualiy of producs in he reamen period. Wih changes in parners occurring hroughou he sample period, (2) amouns o esimaing he ime rend ha he reaed producs would follow, were i no for he effec of he reamen. Airline-roue fixed effecs (i.e. a dummy variable ha equals one for Dela on Alana o Boson in each period) are used o conrol for differences in he qualiy of reamen and conrol roues ha are consan over ime. Because reamen roues are, by definiion, roues ha depar from airpors a which an airline is dominan, hese fixed effecs conrol for non-ime-varying unobservable facors ha increase he qualiy of an airline s flighs ou of airpors a which i is dominan. Moreover, by including a separae fixed effec for each airline-roue, as opposed o a fixed effec ha is common across all reamen roues, all of an airline s reamen roues, or even all of an airline s roues ou of a given airpor, he effec of airpor dominance is conrolled for in a very flexible way. The effec of airpor dominance is allowed o be differen, for example, for Dela s flighs from Alana o Boson and Dela s flighs from Dallas o Boson, even hough boh are flighs ha depar from a Dela hub airpor and arrive in Boson. The airline-roue fixed effecs also capure oher unobservable characerisics of an airline on a paricular roue (ha may or may no be relaed o airpor dominance) ha do no vary over ime. For example, airlines may differ in he frequency of heir flighs on a given roue. Higher fligh frequency may increase he qualiy of he airline s produc on ha roue. Alhough fligh frequency is no observable, he airline-roue fixed effecs conrol for differences across airlines in heir average fligh frequency on a given roue. Airline-quarer fixed effecs (i.e. a dummy variable ha equals one for Dela flighs on all roues in quarer ) are used o esimae an airline-specific ime rend. The airline-specific ime rend conrols for changes over ime in he qualiy of an airline s produc which may be correlaed wih he formaion of FFP parnerships bu which affec he airline s flighs on all roues equally. For example, suppose ha he addiion of FFP parners was par of a service improvemen package by an airline ha also included increased leg room and beer food. Alernaively, suppose ha domesic airlines were beer able o form inernaional parnerships when inernaional airlines perceived hem o be of higher qualiy, based on facors unobserved
o he economerician. In boh of hese cases, changes in FFP parners would be correlaed wih unobservable increases in he qualiy of he airline s produc, bu hese increases would be capured by he airline-quarer fixed effecs. The basic esimaing equaion can herefore be wrien as, (6) ln( s r) ln( s0 r ) = r + ϕ + x rβ αp r + σ ln( s / gr γ ) + ξ r where x r includes a dummy variable for wheher produc is a direc fligh and an ineracion erm beween a measure of consumers earning and redempion opporuniies in airline s FFP in quarer and a measure of airline s dominance a he origin airpor of roue r. 24 effec for airline on roue r, ϕ is a fixed effec for airline in quarer, and γ r is a fixed ξ r capures he unobserved qualiy of airline s produc on roue r in quarer afer conrolling for he average qualiy of airline on roue r and he average qualiy of airline in quarer. The key idenifying assumpion of he model is ha here are no unobserved changes in he qualiy of airline s producs ha are correlaed wih changes in airline s FFP parners and ha differenially affec he qualiy of airline s producs on roues ha depar from airpors a which i is dominan, relaive o roues ha depar from he airpors a which i is no dominan. ξ r, he unobserved qualiy of produc on roue r in quarer afer conrolling for airlineroue and airline-quarer fixed effecs, may include unobserved ime-varying changes in qualiy ha are common o all carriers on a paricular roue or all roues ou of a paricular airpor. For example, here may be seasonaliy in demand ha affecs all carriers on a given roue or all roues ou of a paricular ciy. This migh sugges he inclusion of roue-quarer or roue-season fixed effecs. However, noe ha he esimaing equaion under he assumed nesing srucure (equaion 6 above) includes ln( s / gr ) as a righ-hand-side variable. Recall ha his erm is produc s share of all passengers ha fly roue r in quarer on one of he 12 inside airlines. Noe ha ln( s / gr ) and he lef-hand-side variable, ln( sr) ln( s0 r), will have he same numeraor (he number of passengers flying produc on roue r in quarer ), bu differen denominaors. 24 The effec of earning and redempion opporuuniies on roues ha depar from airpors a which airline is no dominan is no separaely idenified from he airline-quarer fixed effecs.
Roue-quarer fixed effecs canno be included because, for boh of hese erms, he denominaors do no vary across producs in a given roue-quarer and would herefore be eliminaed by he fixed effecs, leaving he wo variables idenical. Roue-season or roue-year fixed effecs also canno be included because, wihin a roue-season or roue-year, here is lile or no variaion in he denominaor of he lef-hand-side variable and very lile variaion in he denominaor of he righ-hand-side variable. Insead, airpor-quarer fixed effecs are included o conrol for he possibiliy of ime-varying unobservables ha are common o all carriers and all roues ou of a paricular airpor. Wih airline-roue, airline-quarer, and airpor-quarer fixed effecs, ξ r capures he unobserved qualiy of produc on roue r in quarer as a deviaion from he average qualiy of airline s producs on roue r, he average qualiy of airline s producs in quarer, and he average qualiy in quarer of flighs ha depar from he origin airpor of roue r. ξ r will hen capure characerisics such as improved scheduling by airline on roue r in a paricular quarer (i.e. more convenien deparure imes) or he presence of consrucion a airline s erminal a he origin airpor of roue r in a paricular quarer ha reduces he qualiy of flying airline ou of ha airpor in ha quarer (his would no be capured by he airpor-quarer fixed effecs). Because unobservables such as hese may be considered by firm when seing is prices and considered by consumers when choosing among airlines, hey may induce correlaion beween ξ r and boh p r and ln( s / gr). To accoun for his, I insrumen for hese erms wih wo ypes of variables ha are assumed o be uncorrelaed wih a produc s unobserved level of qualiy bu correlaed wih is price and wihin-group share. 25 The firs se of insrumens I use includes measures of he level of compeiion in he marke. Following common pracice in he discree choice lieraure, I include as insrumens he number of compeior producs in he marke and he sum over rival producs in he marke of he characerisics ha are included in x r. These insrumens are moivaed by supply-side heory ha predics ha as he number of compeing producs in he marke changes or as he degree of closeness beween produc and compeing producs changes, produc s price and wihin-group-share will be affeced. The second ype of insrumen used explois he fac ha a firm ha offers muliple producs in a marke will oinly se he prices of 25 The moivaion for he ypes of insrumens used is based on a differeniaed producs model of firm compeiion.
is producs. For example, an airline ha offers muliple producs will recognize ha increasing he price for is direc fligh on a roue may cause consumers o subsiue o is own connecing fligh, raher han o a rival produc. Given he consrucion of he daase, an airline offers a mos wo producs in a marke a direc fligh and a connecing fligh. Wih he excepion of price and oal disance, all oher observable variables are idenical for boh of a firm s producs. As a resul, I include as an insrumen he number of producs offered by an airline, bu no he characerisics of he airline s oher producs. IV. Daa and Descripive Resuls This secion begins by describing he daa, he sample selecion and he variable consrucion. I hen presen descripive saisics and he resuls of reduced form fare regressions, inended o provide a sense of he basic relaionship beween enhancemens o FFPs and fares ha appears in he daa. IV. A. Sources of Daa Three sources of daa are used. The daa on prices and marke shares are aken from Daabank 1A (DB1A) of he Deparmen of Transporaion s Origin and Desinaion Survey (O&D). This daabank is a random 10% sample of all domesic ickes ha originae in he U.S. on domesic carriers each quarer. Each observaion in he daabank conains informaion on he roue (he origin, desinaion and any connecing airpors), he carrier(s), he ype of icke (oneway, round-rip or open-aw; direc or connecing), he dollar fare, he disance of he rip, and he number of passengers observed flying ha iinerary a he given fare in he quarer. The daa also conain a variable indicaing he poin of purchase of he icke, which allow deparing and reurning legs of roundrip ickes o be disinguished. The daa unforunaely do no conain any informaion on icke resricions, such as wheher he icke required a Saurday nigh say-over or wheher i was purchased more han 7 or 14 days in advance. The daa also do no idenify he dae or ime of ravel (only he quarer is idenified), so observaions canno be mached o paricular days of he week or o specific fligh numbers.
This daase is combined wih daa on he FFP parners of six airlines: American Airlines, Coninenal Airlines, Dela Airlines, Norhwes Airlines, Unied Airlines, and US Airways. 26 The informaion on FFP parners was compiled using Inside Flyer magazine, a monhly publicaion argeed owards frequen flyers. 27 In each ediion, Inside Flyer publishes a feaure called Programs & Parners which provides deailed informaion, for every airline operaing an FFP, on all of he parners paricipaing in ha airline s program. I use his informaion o idenify he FFP parners for each of hese six airlines in each monh and o rack addiions and losses of parners over ime. 28 Parners are considered o oin an airline s FFP in he firs monh in which hey appear wih ha airline in he Programs & Parners secion. In order o combine he parner daa wih he DB1A daa, I assume ha an airline is a parner in a given quarer if i appears as a parner in he firs monh of ha quarer. To deermine he exac changes in earning and redempion opporuniies resuling from he addiion or loss of a FFP parner, changes in parner airlines are mapped ino changes in he acual se of flighs ha are eligible for poins accrual and redempion. This requires examinaion of each individual parnership because, in many parnerships, only a subse of he parner airline s flighs is acually made available o members of he hos airline s FFP. This informaion was obained using five ediions of he Official Frequen Flyer Guidebook (OFFG). The OFFGs provide very deailed informaion on each airline s FFP, is parners and he exac rewards ha can be redeemed wih each parner. 29 For example, OFFG sixh ediion liss Aer Lingus as a Dela Airlines parner and liss wo ypes of rewards ha are available o Dela Airlines FFP members on Aer Lingus. The firs is, [b]eween he 48 coniguous saes, Alaska, Bermuda, Canada, he Caribbean or Mexico and geographical Europe. The second is, [w]ihin geographical Europe. Based on his informaion, all Aer Lingus flighs ha fall ino eiher of 26 I focus he FFP analysis on hese six airlines because hey have he greaes amoun of FFP parnership aciviy. TWA, Midwes Express, America Wes and Alaska Airlines, which are included among he 12 inside airlines, have a limied amoun of parnership aciviy over my sample period. A fuure draf will ideally include FFP daa for hese airlines as well. 27 Inside Flyer and he Official Frequen Flyer Guidebooks are published by a company called Frequen Flyer Services. This company develops and disribues producs and services ha focus on frequen raveler services. For more informaion on Frequen Flyer Services, see hp://www.webflyer.com/. 28 Occasionally, I quesioned he accuracy of specific pieces of informaion aken from he Programs & Parners secion. For example, an airline would appear as a parner in a program for a number of consecuive monhs, hen no appear for a shor period and hen reappear. In siuaions like his, I aemped o confirm he saus of he parnership using oher sources such as press releases appearing on he airlines websies, frequen flyer oriened websies and he OFFGs. 29 I compared he informaion in he OFFGs o he Program Guides, which are produced by he airlines for disribuion o members of heir FFP. The OFFGs appear o ake he parnership informaion almos direcly from he Program Guides. I could no use he Program Guides hemselves for his informaion because pas years Program Guides for all airlines in all years could no be obained.