Exchanging values of unsold seats with airline alliance partners in a competitive environment

Transcription

1 Exchanging values of unsold seas wih airline alliance parners in a compeiive environmen Alyona Michel 1,2 and Peer P. Belobaba 1 1 Massachuses Insiue of Technology Cambridge, Massachuses 2 oneworld Airline Alliance New York, New York s: almichel@mi.edu and belobaba@mi.edu Absrac Airlines in alliances expand he scope of heir neworks hrough code sharing, which allows an airline o sell seas on anoher airline s fligh as if he seas were is own. However, code sharing is problemaic for airline revenue managemen (RM) sysems. RM improves revenues by placing booking limis on he availabiliy of lower fares in a marke. Alliance parners do no work ogeher o deermine sea availabiliy for code share flighs, producing asymmeric booking limis. Thus, sea invenory is sold for less han is value o he nework, and revenues are lower. In his paper, we es wheher alliance parners facing compeiion from anoher alliance benefi from using informaion abou he esimaed value of unsold seas on parner flighs in heir RM sysems. I is found ha alliance nework characerisics and he behavior of he compeiion affec he performance of he wo mehods esed. Resuls from he simulaion show ha under favorable circumsances, benefis reach 0.43% if he alliance carries a large share of code share raffic a high fares and does no risk losing non-code share high fare raffic. Depending on nework characerisics, he sophisicaion of he compeiion s RM can decrease and even reverse revenue benefis. I. INTRODUCTION In he las 2 years, airline alliances have acquired a long lis of new members, and few of he world s remaining qualiy airlines large or small have no eiher joined an alliance or a leas parner wih oher airlines. In he year 2013, alliances earned 67% of worldwide passenger airline revenue, and earned a revenue premium on he raffic hey carried [14]. Wih alliances consanly searching for new members (or code share parners) o fill he gaps in heir worldwide neworks, hey will coninue o grow in size and imporance. The hree major alliances, oneworld, Sar and SkyTeam, complee inensely on he world s long-haul business roues. A major feaure of alliance cooperaion in providing good service in an imporan marke is code sharing. Code sharing allows an alliance parner o sell seas on anoher airline s fligh as if hey were flown by he selling parner. This allows parners o expand he scope of heir neworks o disan desinaions hey do no (or canno) fly o. Thus, parners virually connec disan poins, serving global markes each would oherwise be unable o serve wihou having flying righs and deploying is own planes and personnel. Airlines use revenue managemen (RM) o aemp o opimize he revenue earned from available seas on he flighs in heir nework. RM sysems use availabiliy conrol o place booking limis on he availabiliy of cerain fares in a marke, or origin-desinaion pair (O-D). Resricing he sale of low fares laer in he booking process (ime window during which ickes are sold) improves revenues because ickes are sold a higher fares when cusomers have a higher willingness o pay. In he absence of alliances, an airline s raffic consiss of local raffic (when he passenger flies on one fligh leg from his origin o his desinaion) and connecing (when he flies on muliple fligh legs). An airline s RM sysem only has access o informaion abou he sae of he airline s own nework. Thus, airlines canno accuraely evaluae booking limis for code share iineraries (which consis of fligh legs operaed by differen parners) due o he lack of informaion abou parners operaed fligh legs. Wheher hey are unable or unwilling, alliance parners do no work ogeher o deermine booking limis on code share flighs. This ofen produces asymmeric booking limis on iineraries ha involve code share flighs. Thus, sea invenory is sold for less han is poenial value o he nework, and revenues are lower for he alliance. In his paper, we es wheher i is beneficial for revenues if alliance parners facing compeiion from anoher alliance incorporae informaion ino heir RM sysem s decision-making process abou he esimaed value of unsold seas on heir parner s operaed flighs, called bid prices. Two mehods are examined. Bid price sharing (BPS) incorporaes he bid prices ino he decision process afer booking limis have been deermined for he airline s own local and connecing iineraries. Dynamic valuaion (DV) incorporaes bid prices ino he calculaion of booking limis a he same ime as oher booking limis are deermined, having a larger effec on booking limis on he airline s own local and connecing fares. The goal of his paper is o es which mehod performs beer for alliances wih a nework srucure resembling ha of he hree global alliances, which have srong parnerships across he Alanic. This paper addresses a poenial research direcion ha follows from a previous paper [17], which esed he effecs of BPS in a US-based alliance nework wih a very differen srucure. Also, his paper aims o gain insighs ino he performance of he wo mehods depending on he behavior of he compeing alliance, and he nework characerisics of he alliances.

2 II. BACKGROUND This secion provides background on alliance RM conceps imporan o he experimens in his paper. A. Three Types of Alliance Fligh Traffic: Local, Connecing, and Code Share Figure 1 below illusraes hree fligh legs. Alliance parner 1 operaes he flighs DEN-ORD and ORD-FRA, and also code shares on parner 2 s operaed fligh FRA-BUD. Parner 2 operaes FRA-BUD and also code shares on parner 1 s fligh ORD-FRA. Various combinaions of hese flighs comprise he hree componens of alliance fligh raffic: In an own local iinerary, he passenger flies on one fligh leg from his origin o his inended desinaion. The figure illusraes hree possible local iineraries, DEN-ORD, ORD- FRA, and FRA-BUD. In an own connecing iinerary, he passenger flies on muliple fligh legs of a single airline from his origin o his inended desinaion. In his illusraion, only one own connecing iinerary is possible: DEN-ORD-FRA on parner 1. In a code share iinerary, a passenger flies on muliple fligh legs operaed by differen parners, bu markeed under a single airline s code, i.e., sold by ha airline. In his illusraion, a passenger may purchase he code share iinerary ORD-FRA-BUD from eiher parner 1 or parner 2, because boh parners sell he iinerary under heir operaing codes. valuaion refers o he value used as inpu o he RM sea allocaor opimizaion model. The possibiliies include using he oal fare of he code share iinerary, he local fare of he parner s own raversed fligh leg (boh oal and local valuaion overvalue he code share iineraries), a proraion of he oal fare, or he novel mehod of DV. In his experimen, we use local valuaion of code share iineraries, as represenaive of curren airline pracices, for all ess ha do no involve dynamic valuaion. Figure 2 presens an example of he valuaion of ORD-FRA-BUD. In his example, local valuaion values he iinerary a $700, whereas DV values i a $650, boh higher han he O-D fare. Figure 2. Example of code share local/dynamic valuaion. Figure 1. Componens of Alliance Fligh Traffic. B. Alliance Revenue Managemen Alliance RM consiss of he following four componens, described below: recording and forecasing, code share valuaion, sea allocaion (or opimizaion), and code share availabiliy conrol. This paper s assumpions regarding he RM characerisics in he experimens are also saed. An airline s RM sysem draws on a hisorical daabase of prior bookings o forecas fuure demand by fare class and O-D marke. Recording and forecasing of bookings refers o how airlines record acceped bookings for laer use in forecasing. Full informaion of he code share booking is needed from he booking agen (such as a Global Disribuion Sysem), in order o know he full iinerary and disinguish i from a local booking when making fuure forecass. This paper assumes ha alliance parners record he full iinerary and forecas code shares separaely from locals. An airline ses fares (which range from low o high, depending on he booking class) for he local and connecing iineraries i offers for sale. These fares are provided as inpu o he RM sysem (along wih demand forecass) for calculaing booking limis. Less sraighforward, however, is he value an airline should aribue o a code share iinerary. Code share Airlines hen deermine how many seas on a fligh leg o make available for sale. Sea availabiliy is resriced hrough booking limis on low fares in an O-D marke. Booking limis are calculaed during sea allocaion (or opimizaion). The RM sysem may use opimizaion ha is leg-based (opimizing revenues on an individual fligh leg) or O-D/nework based (aiming o opimize revenues over he enire nework by considering fligh leg opporuniy coss). Expeced Marginal Sea Revenue (EMSRb) developed by Belobaba [1, 2, 3], is he leg-based mehod used in his paper. The heurisic calculaes he expeced revenue from an empy sea on a fligh leg and allows a booking o be made if he fare exceeds he EMSR. The wo nework-based mehods used in his paper are Displacemen Adjused Virual Nesing (DAVN) and Probabilisic Bid Price (ProBP). In DAVN, a nework linear program produces shadow prices ha represen he opporuniy coss of filling he las empy sea on a fligh leg (for more informaion, see [22]). Noe ha, if he airline fligh leg is no booked o near capaciy, mos shadow prices produced by he deerminisic linear program of DAVN are zero. However, as bookings on he fligh leg reach capaciy, hese shadow prices quickly approach an expensive fare on ha fligh leg. The RM opimizer uses he oal O-D fare less a fligh leg s shadow price o deermine which O-D fares involving a fligh leg o make available for sale. If a leg shadow price is small, he esimaed nework value of selling he sea will be higher. Thus, a sea on ha fligh leg is more likely o be available for booking a a low fare. Because i uses muliple fare level ranges (called buckes ) o deermine open/closed booking

3 availabiliy, DAVN is a robus nework RM mehod, wih resuls less sensiive o small variabiliy in hisorical inpu daa or demand forecasing mehods. However, i is sill a heurisic. ProBP opimally proraes he oal O-D fare o he affeced fligh legs using an ieraive algorihm. The values of he las and marginal seas on a fligh leg are calculaed. These EMSR values are proraed o he fligh legs comprising he iinerary unil convergence is reached (more informaion in [8]). The resuling proraed EMSR values are he bid prices of he affeced fligh legs. Theoreically more opimal han DAVN, his mehod is also more sensiive o inpus and forecass. I can perform worse han DAVN, especially a low demand levels or when here is high demand variabiliy. This paper ess using BPS or DV when he alliance parners boh use symmeric DAVN or ProBP, and when he compeing alliance uses EMSR, idenical RM, or idenical RM wih BPS or DV. RM mehods all produce some esimae of he value of an unsold sea, which we refer o, wihou loss of generaliy, as bid prices. Figure 3 below summarizes he characerisics of he hree RM mehods esed in his paper, and he ypes of bid prices generaed. Figure 3. Characerisics of he RM mehods EMSRb, DAVN, and ProBP. The RM sysem mus decide which code share iineraries are available a which fare, referred o as code share availabiliy conrol. These include bid price sharing conrol (proposed by [9]) and sandard availabiliy conrol, when code share booking limis are deermined inside he RM opimizer, during he sea allocaion sep. Imporanly, he code share valuaion inpus may differ. If local valuaion is used, code shares are worh as much o he nework as own local bookings. Wih sandard conrol using DV, code shares are reaed as disinc from locals because hey receive a differen valuaion, which may be lower or higher han he local valuaion. Wih BPS, sandard availabiliy conrol wih local valuaion is used o calculae bid prices. Bid prices are hen shared by parners o inform sea availabiliy decisions afer booking limis have been deermined for own local and connecing iineraries [15]. C. Descripion of he Passenger Origin-Desinaion Simulaor The Passenger Origin-Desinaion Simulaor (PODS) is a sofware simulaion developed a The Boeing Company in he mid-1990s by Hoppersad, Berge and Filipowski [13]. I is used in his research o es he effecs of sharing sea value informaion among alliance parners. Two underlying models wihin PODS represen he processes of passenger choice and airline RM sysems, which inerac a he levels of sea allocaion and passenger choice/decision [6]. The sofware simulaes he booking process ino 16 ime frames covering 63 days prior o fligh deparure. A he sar of each ime frame, sea allocaion is reopimized based on bookings o dae. Each simulaion rial consiss of 600 weeks, or samples. To eliminae bias from any iniial condiions, he firs 200 samples are discarded. An average of he remaining 400 samples gives he simulaion rial oupu, and ensures small sandard deviaion of he resuls. The PODS Research Consorium a MIT uses he sofware o es he effecs of differen revenue managemen echniques on airlines. Paricipaion from 9 airlines gives he Consorium a perspecive ha is oriened o he realiies and operaing consrains faced by airlines oday. D. Prior Work on Alliance Revenue Managemen Booking limis on code share iineraries are sub-opimal because he seas in he nework are no being allocaed as if hey were par of one nework, and each parner does no have full knowledge abou he sae of he oher s sysem. One simulaion resul esimaes ha nework RM should improve revenues by %, bu if code share raffic is presen, he gains drop o % [14]. This loss can be exacerbaed because code shares are ofen long-haul inernaional roues wih high revenue conribuion. Therefore, prior research in he field of operaions research has aemped o address he problem of alliance revenue managemen and propose opimal soluions [7, 12, 15, 17, 18, 20, 21, 23, 24]. Some soluion approaches involve modifying he valuaion of code share iineraries [15, 17, 18, 23]. Findings in [23] show ha he performance of saic valuaion schemes is nearly as good as he bes dynamic schemes (achieving abou 90% of revenues), bu ha saic schemes do no adjus well as nework parameers change, which could rapidly reduce revenues. The nework in [23] consiss of jus wo fligh legs wih en seas each, in which he proporion of code share iineraries was varied. In [15, 17, 18], i is found ha DV using daed parner bid prices performs slighly beer han BPS in a complex USbased nework wih consan nework parameers. An opion ha could achieve opimal join revenues, suggesed by boh [7] and [21], is o exchange seas among he alliance parners unil he relaive values of he seas o each airline's nework are equal. Afer he seas are exchanged and paid for among he carriers, each airline has individual conrol over he seas by is own RM sysem. In [7], his proposiion is formulaed in erms of marginal sea values, and in [21] in erms of bid prices, bu he idea is he same. In he proposed scenario, he resource allocaion is opimal and expeced revenue for he alliance can be maximized. However, in pracice, airlines do no calculae bid prices or expeced sea revenues for seas on flighs ha hey do no operae, as hey do no have access o he booking hisories and forecass for hose non-operaed flighs. Significan echnological changes o he airlines' RM sysems and anirus immuniy may be required before such a scheme could be implemened. Applying finance heory, [12] proposes an opions-based approach o capaciy conrol on a single fligh leg, where he markeing carrier can purchase opions from he operaing carrier for he righ o buy sea invenory in he fuure a a predeermined srike price. No heoreical basis or evidence is

4 presened ha his scheme performs beer han he oher mehods proposed in he lieraure. In [20], a deerminisic LP model inspired by ha in [22] is developed, similar o he approach of DAVN. The LP is solved over he join alliance nework, and hen he consrains linking he parners' decisions o he join alliance nework soluion are relaxed hrough he dual prices, hus decomposing he problem ino smaller problems by airline. Under BPS conrol, a code share iinerary is available for booking if he fare exceeds he sum of he bid prices on all he legs raversed. I was proposed in [9] ha airlines eiher exchange eiher acual bid prices (if hey have he legal abiliy o do so), or o infer bid prices from he lowes available fares, an idea re-ieraed in [21] in order o inform one anoher of he value of sea invenory on code share fligh legs. Bid price inference may be a feasible opion for airlines ha do no have he anirus immuniy o share such informaion. The smallscale simulaions in [23] show ha bid price conrol produces gains jus as good as hose of he dynamic ransfer price schemes in he cooperaive game of complee informaion. III. METHODS AND EXPERIMENT SETUP A. Characerisics of PODS Alliance Nework E Nework E represens wo compeing alliances, each wih parner hubs across he Alanic. A schemaic of nework E is illusraed in Figure 4. The wo compeing alliances have one se of parners (airlines 1 and 2) whose hubs are locaed in he cenral Unied Saes, while he oher wo parners (airlines 3 and 4) have hubs locaed in Europe. Airlines 1 and 3 form alliance 1, and airlines 2 and 4 make up alliance 2. The spoke ciies emanae from he coninenal hubs, wih 10 in he norhern par and 10 in he souhern par of Europe, as well as 10 in he wesern par and 10 in he easern par of he Unied Saes. In he baseline case for nework E, all airlines use EMSRb wih leg forecasing. Roughly speaking, nework E has a dumbbell srucure. The hub-o-hub runk roues (served by large aircraf) carry a large amoun of passengers across he Alanic ino he hubs and feed he smaller hub-o-spoke roues. Such runk flighs occur hree imes a day, providing connecing opporuniies during hree banks. The roues are served by each airline, resuling in a oal of six hub-o-hub flighs and 12 addiional rans-alanic flighs per bank. In addiion, some local and connecing raffic does no cross he Alanic, bu raverses spoke-o-hub and hub-ospoke roues, saying on he same coninen. Also, here are hub-bypass roues crossing he Alanic ha go from a coninenal hub o a major ciy on he oher coninen, wihou requiring a connecion a he parner hub. The differen ypes of roues ensure ha he nework is varied and represenaive of curren alliance neworks. B. Bid Price Sharing for Code share Availabiliy Conrol BPS refers o any process of parners exchanging nework displacemen cos informaion, or he esimaed value of selling one iem of sea invenory on a fligh leg. If he sea allocaion mehod used by he RM sysem is DAVN, hen he resuling bid prices are acually shadow prices from a deerminisic linear program for each leg on an airline s own nework. The oal code share fare (CSfare) mus exceed he sum of he minimum hreshold of he lowes open fare range bucke and he parner s shadow price (SP). This mus be saisfied for boh parners, known as dual conrol. The BPS availabiliy conrol equaion is: parner CSfare Bucke min SP. (1) Wih PROBP, ieraive opimal proraion of fares produces bid prices for each leg. The availabiliy conrol equaion upon BPS is: parner CSfare BP own BP. (2) Figure 5 shows an illusraion of he BPS conrols using DAVN or ProBP. Figure 5. DAVN and ProBP Code Share Conrol Comparison. Figure 4. Trans-Alanic Nework E. In he above equaions, he parner s bid price is he mos recen available. The experimens described in his paper use daily opimizaion for ProBP and daily calculaion of shadow prices for DAVN (hough booking limis on fare buckes are recalculaed a he sar of each of 16 ime frames). Prior experimens wih reopimizaion every ime frame and every 200 bookings are described in [17] and [18]. As par of he experimens described in his paper, less frequen opimizaion was also esed. Exacly as before, i was found ha more frequen opimizaion slighly improves resuls for ProBP, and largely for DAVN. The bid prices more accuraely reflec he sae of he nework, hus he RM sysems produce more opimal booking limis and more revenue is generaed. Because

5 he resuls were so similar, his paper only shows he resuls using daily reopimizaion. This paper hus assumes ha airlines have he echnological capabiliy o reopimize, exchange bid prices, and incorporae parner bid prices ino heir RM sysem (wih DV) on a daily basis. I is acknowledged ha a range of capabiliies exis among oday s airlines, some wih inegraed sysems and real ime communicaion among paren airlines and heir subsidiaries, and more rudimenary communicaion beween oher alliance parners. BPS and DV require sharing informaion so quickly and frequenly ha hey may require more advanced RM sysems and communicaion echnology on he par of he alliance members. C. BPS Availabiliy Conrol vs Dynamic Valuaion The key difference beween BPS and DV is ha BPS incorporaes he bid prices ino he decision process afer booking limis have been deermined for he parner s own local and connecing iineraries. DV incorporaes bid prices ino he calculaion of booking limis a he same ime as oher booking limis are deermined, having a larger effec on booking limis on own local and connecing fares. Wih DV, he valuaion of code shares is dynamic because i is changing wih every RM reopimizaion. Recall ha wih BPS, he value of a code share iinerary used as inpu o parner 1 s RM opimizer is he same as a local online fare of a passenger flying only parner 1 s fligh legs. In DV, he code share valuaion used as inpu o he opimizer (CSvaluaion) is he oal code share fare (CSfare) minus he mos recen bid price from parner 2 parner CSvaluion CSfare own BP. (3) This resuling valuaion can be lower or higher han local valuaion, depending on he bid prices and code share fare. Refer o figure 2 for an example of dynamic valuaion. The resuls using DV in [15] are very encouraging and show revenue gains near o hose of BPS for code share availabiliy conrol in DAVN sysems, and much beer resuls in ProBP (abou a 0.25% revenue gain for DAVN and ProBP). I is argued ha improved performance of ProBP under dynamic valuaion is a resul of he sensiiviy of he ProBP opimizaion o fare inpus. This paper coninues he examinaion of dynamic valuaion in a differen nework seing, wih he exchange of curren bid prices and more frequen opimizaion from ha used in [15], wih resuls indicaing more complicaed conclusions. D. Dimensions Tesed in Experimen The focus of his paper is alliance revenue managemen in a compeiive environmen, which previous research has no addressed. We also es he benefis of BPS or DV when he alliance uses he differing nework RM mehods of DAVN and ProBP. Alhough ess were performed o compare daily opimizaion wih opimizaion once per ime frame, he resuls are similar o prior findings ha show ha daily opimizaion largely benefis DAVN, and are no shown. The alliance may face a compeing alliance ha uses eiher EMSR, he same nework RM mehod, or he same nework RM mehod in combinaion wih BPS or DV. Two differen demand levels are esed o undersand how he demand level affecs he performance of each mehod. The aim is o gain insighs ino he performance of BPS compared wih DV in differen compeiive and nework environmens. Table 1 provides a summary of he dimensions esed in he experimen. TABLE I. SUMMARY OF EXPERIMENT DIMENSIONS Experimen: Alliance 1 DAVN ProBP EMSRb DAVN ProBP Alliance 2 EMSRb DAVN ProBP DAVN ProBP Code share valuaion Availabiliy Conrol BPS Frequency Demand Level Local and Dynamic Valuaion Sandard conrol wih local or DV, or Bid Price Sharing Daily exchange of bid prices High (86-88%) and medium (82-83%) load facor IV. RESULTS A. Alliance 1 Uses BPS or DV, Alliance 2 Does No Figure 6 shows he resuls obained when alliance 1 uses BPS or DV under differen demand levels. Under medium demand, DV leads o he highes revenue gains for boh DAVN and ProBP. However, ProBP wih BPS performs beer han DV a high demand. When he compeior also uses ProBP, he paern of gains over sandard ProBP is similar, bu smaller. Wih DAVN for alliance 1 and EMSRb for he compeior, BPS does no improve revenues a he medium demand level. A high demand, gains occur from boh BPS and DV. When he compeior uses DAVN as well, he order of gains from BPS and DV remains he same, bu all mehods produce revenue gains a boh medium and high demand levels because of a beneficial compeiive feedback effec. Figure 6. Alliance 1 uses BPS or DV, Alliance 2 does no. The revenue gains are sill presen for alliance 1 even when alliance 2 uses nework RM as is revenue managemen mehod raher han EMSRb. However, he resuling revenue gains are smaller when he compeior uses a more inelligen RM sysem.

6 B. Alliance 2 Uses BPS or DV, Alliance 1 Does No Figure 7 below shows ha alliance 2, using ProBP a any demand level, and DAVN a high demand, also benefis from BPS when he compeior uses EMSR, bu less han he benefi for alliance 1. When he compeiion uses nework RM, he revenue gains are generally smaller han hey were for alliance 1 as well. Also, noe ha alliance 2 benefis less from nework RM in general, compared wih alliance 1, due o he fac ha i carries a higher proporion of local raffic han alliance 2. Alliance 2 benefis less from DV han BPS, he opposie of he effec on alliance 1. Figure 9. Average componen fares for alliances 1 and 2. Figure 7. Alliance 2 uses BPS or DV, Alliance 1 does no. C. Boh Alliances Use BPS or DV, High Demand Only Figure 8 shows he resuls obained when boh alliances use BPS or DV, a high demand only. Alliance 1 reains a revenue gain from BPS and DV. The gain for alliance 2, in his siuaion, is very minor wih BPS. DV, however, causes losses for alliance 2 when is compeior also uses DV. Figure 8. Boh alliances use BPS or DV, high demand only. D. Discussion of Differences beween Alliances 1 and 2 We have seen ha BPS wih dual conrol is he only mehod ha consisenly helps boh alliances. BPS works bes for alliance 2 in all cases, while DV works bes for alliance 1. This is because he nework characerisics of he 2 alliances differ: code share average fares are lower for alliance 2 han for alliance 1, and average local fares are higher. Also, locals comprise he larges proporion of revenues for alliance 2, and code shares a smaller proporion, as compared wih alliance 1. The lower code share fares combined wih he higher local fares causes alliance 2 o experience spiral down (his occurs when he RM sysem undervalues seas and opens oo many low fares for booking) when using DV. Alliance 1 does no experience a revenue decline because of differen nework characerisics: higher code share fares, lower local and connecing fares, and higher revenue share from code shares. Figure 9 illusraes he average componen fares for alliances 1 and 2, and shows how he average fares change afer he applicaion of ProBP, and addiionally BPS and DV. Code share fares increase more for alliance 1 han for alliance 2 when applying nework RM and furher alliance RM echniques. Connecing fares increase much more for alliance 2 han alliance 1, making ha componen more imporan for revenues. When DV is used by alliance 2, subracing he already higher bid prices (resuling from he higher local fares of alliance 2) from he lower code share fares causes spiral down and harms revenues. Too many code share bookings are allowed a he expense of high-revenue local and connecing iineraries. This does no occur for alliance 1 because of is inherenly higher code share fares and lower local fares. V. DISCUSSION A. Conclusions In nework E, which differs from US-based nework A4 (used previously in his body of research in [15, 17, 18]) in erms of is physical srucure, he properies of he ypes of flighs (shor-haul local and domesic connecing, along wih long-haul inernaional code share flighs), and he benefis of BPS and DV, depend on he nework characerisics o a larger degree. If code share flighs are high-revenue relaive o local iineraries, hen BPS and DV improve revenues by raising bid prices (ProBP) or improving booking limis (DAVN). These benefis are more pronounced, and do no affec he revenues from he oher raffic componens, when code share iineraries comprise a larger proporion of revenues. If, however, he airlines in an alliance do no obain he majoriy of revenues from code share raffic and he average code share fare is relaively low compared wih he average local fare, hen he poenial danger of displacing a local passenger and causing revenue loss presens a problem for implemening BPS and DV. The benefis from BPS and DV are presen in mos cases for alliances of differing srucure, and hey are larger if he compeiion is using less sophisicaed RM mehods. Addiionally, if wo compeing alliances differ in heir nework characerisics such ha one carries higher-revenue code share passengers and obains a large amoun of revenues from ha componen, hen i could experience a larger benefi from BPS and DV o he derimen of he compeior. The resuls indicae

7 ha he frequen opimizaion of bid prices and booking limis may be paricularly imporan in neworks wih a large difference in he lenghs of heir ses of fligh legs, inense compeiion beween wo alliances, and fare srucures ha are prone o spiral down. B. Fuure Research Direcions The wo neworks, A4 [15, 17, 18] and E, used in his body of research had specific nework srucures ha resuled in somewha differen conclusions abou he performance of BPS and DV. Expanding his research o neworks of various oher srucures, and wih oher characerisics (e.g., he raios of he average code share fare o local and connecing fares, and he proporions of revenue derived from he various raffic componens) would help o compile a more holisic picure of he benefis of BPS and DV as a funcion of nework srucure. Some lieraure on he alliance RM opic [23] has modeled he effecs of parners choosing which bid prices o pos for heir own fligh legs o maximize heir own individual revenues, assuming ha ha he posed bid price is equal o he ransfer price, or he revenue ha will receive from he operaing parner if he iinerary is sold (revenues are no spli according o a prorae agreemen in his case). In addiion o his parner price scenario, i is possible ha some degree of bid price scaling (eiher up or down), in cerain cases, would produce beer availabiliy decisions. A opic for furher research is he effec on oal alliance revenues of modifying bid prices for eiher muual or individual benefi. This paper was concerned wih he combined alliance revenues and has assumed ha he revenue resoluion conracs are fixed and have been negoiaed such ha each parner views his share as a fair division. Anoher direcion for fuure research is o examine he effecs of differen revenue resoluion schemes or prorae agreemens beween alliance parners on he behavior and incenives of individual paricipans. Some airlines use fare adjusmen (also called marginal revenue opimizaion ), which decreases he fare inpus o he opimizer of lower class iineraries [11]. Some bid prices will be much lower if using fare adjusmen. The performance of BPS may be significanly differen if airlines are using his mehod. An idea suggesed by [9] is ha of bid price inference, where he lowes available fare on a fligh leg is used as a pseudo-bid price. If acual bid prices are skewed downwards because of echniques like fare adjusmen, hen using lowes available fares may be a feasible alernaive. Bid price inference is also applicable in he case when airlines do no have anirus immuniy o share bid prices, in which case communicaing o each oher he lowes available fares (which are public and would no require immuniy) is sill possible. Research on he success of BPS when using pseudo-bid prices, because i is unsuiable or infeasible o exchange acual bid prices, is anoher opic of ineres. We have seen ha DAVN performs well as a nework RM mehod because is composiion of muliple heurisics makes i robus, and alone (wihou BPS) i produces large revenue gains even wihou frequen opimizaion. However, in he nework E environmen, applying BPS or DV o DAVN wih ime frame opimizaion resuled in large revenue losses. The work of [15] showed ha a differen, sricer implemenaion of BPS in DAVN resuled in beer performance. Because DAVN comprises muliple heurisics, i may require airline-specific weaking o obain maximum performance. Furher research ino he adjusmens needed o improve he performance of BPS in DAVN, and in wha siuaions such adjusmens are appropriae, would be of pracical imporance. Also of pracical imporance o he indusry is o es BPS and DV wih differen RM combinaions ha are no limied o he O-D conrol mehods of DAVN and ProBP. We have esed various nework and leg RM combinaions in anoher nework (A4), bu i is imporan o coninue his research for neworks of varying srucures as well, and confirm wheher BPS wih dual airline conrol remains he only mehod ha consisenly delivers revenue gains, hough marginal in some cases. To he auhor's knowledge, mos of he airlines in alliances are only jus aking he firs seps owards sharing bid price informaion and incorporaing i ino heir code share availabiliy decisions. A major limiaion of he models and mehods proposed in he lieraure hus far is ha hey were esed on relaively small and simple hypoheical neworks. Oher han his research, he auhor is unaware of lieraure ha also examined he performance of alliance RM in a compeiive alliance environmen. Once such alliance RM sysems are in place as hose proposed here and in he lieraure, daa documening he code share availabiliy decisions and acual mechanisms used by airlines in real world compeiive and nework scenarios can be colleced. The success of he various mehods proposed here and elsewhere in he lieraure can hen be validaed according o real airline daa. C. Implicaions for Airlines The resuls show ha airlines in alliances who are considering sharing informaion abou he value of seas on heir fligh legs should carefully consider heir nework srucure and compeiive environmen before deciding o implemen such mehods ino heir availabiliy conrol. If he nework srucure and compeiive environmen are conducive o obaining large benefis from sharing bid price informaion, hen large benefis will compound over ime, ouweighing he one ime expense and echnological complexiy of implemenaion. Gains over $75 million per year can be obained by a large airline ha paricipaes in an alliance such as American Airlines (assuming an approximae percen revenue gain of +0.30%) [14]. Alhough DV can produce he larges revenue gains for an alliance, paricularly when code shares are a valuable and imporan par of he bookings and when he compeior uses less advanced RM, i is also more cosly or difficul o implemen because i requires modificaion of he RM sysem s valuaion procedure. BPS, on he oher hand, allows he RM sysem o funcion unmodified, and requires a real ime es by boh parners afer own booking limis have been deermined. Thus, BPS may be easier for airlines o implemen. Alhough BPS does no always provide he highes revenue gains, he resuls in his paper showed ha i is a more robus mehod for airlines whose nework srucures make hem more suscepible o spiral down when hey use DV (such as alliance 2). Wheher

8 BPS or DV is a worhwhile invesmen is ulimaely for airlines o decide. In an indusry wih iny profi margins and huge coss, small revenue benefis can make a big difference. This research has shown ha alliance parners can cooperae and share heir bid prices, wheher hey ake he form of proraed expeced values of empy seas on fligh legs, or nework displacemen coss approximaed as shadow prices from a deerminisic LP, o aemp o improve he oal alliance revenues. Through he use of bid price sharing for pos-fac code share iinerary availabiliy conrol, or by dynamically valuing he code share iinerary's revenue conribuion in he nework opimizer, airlines in alliances can improve heir revenues and affirm he benefis of enering ino code sharing agreemens. ACKNOWLEDGMENT This research was suppored by he PODS consorium s paricipan airlines. Thank you o Lufhansa, Unied, Swiss, SAS, Air Canada, Dela and KLM whose involvemen in he research process and vocal paricipaion in discussions abou alliance revenue managemen during PODS conferences were paricularly helpful. Wihou he PODS simulaor research ool (developed a The Boeing Company) and improvemens o is capabiliy hroughou his research by Craig Hoppersad, his research would no have been possible. REFERENCES [1] Belobaba, P. P Air Travel Demand and Airline Sea Invenory Managemen.Ph.D. hesis, Massachuses Insiue of Technology, Fligh Transporaion. [2] Belobaba, P. P Applicaion of probabilisic decision model o airline sea invenory conrol.operaions Research. 37(2) [3] Belobaba, P. P The Revenue Enhancemen Poenial of Airline Yield Managemen Sysems In: Proceedings of he 2nd Inernaional Conference on Advanced Sofware Technology for Air Transpor, November [4] Belobaba, P. P Airline Nework Revenue Managemen: Recen Developmens and Sae of he Pracice. The Handbook of Airline Economics.The Aviaion Weekly Group of he McGraw-Hill Companies, New York. Chaper 10, [5] Belobaba, P. P., A. R. Odoni, C. Barnhar The Global Airline Indusry. John Wiley & Sons, Ld. [6] Belobaba, P.P Passenger Origin-Desinaion Simulaor (PODS): Summary of Processes and Models. PODS Consorium. Sepember, [7] Boyd, A. E., Airline alliances. OR/MS Today 25(10), [8] Brau, S Nework Value Concep in Airline Revenue Managemen.Maser s hesis, Massachuses Insiue of Technology, Cambridge, MA. [9] Daro, J. F. J Revenue Managemen for Airline Alliances: Passenger Origin-Desinaion Simulaion Analysis. Maser s hesis, Massachuses Insiue of Technology, Cambridge, MA. [10] De la Torre, P. E. F Airline Alliances: The Airline Perspecive. Maser s hesis, Massachuses Insiue of Technology, Cambridge, MA. [11] Fiig, T., Isler, K., Hoppersad, C., Belobaba, P., Opimizaion of mixed fare srucures: Theory and applicaions. Journal of Revenue and Pricing Managemen 9, [12] Graf, M., Kimms, A., An opion-based revenue managemen procedure for sraegic airline alliances. European Journal of Operaional Research 215(2), 459{469. [13] Hoppersad, Craig Passenger O/D Simulaor PODS. Hoppersad Consuling, Seale.v.10 ediion, Sepember, [14] Inernaional Air Transpor Associaion, Airpor Inelligence Services Traffic Per Region Repor for he 12 Monhs Ended December February, airpor-is.com [15] Jain, H Alliance Revenue Managemen in Pracice: Techniques and Simulaion Analysis. Maser s hesis, Massachuses Insiue of Technology, Cambridge, MA. [16] Lufhansa The Lufhansa Group repors operaing profi of EUR 578m afer nine monhs. Available a hp://presse.lufhansa.com/ en/news-releases/singleview/archive/2011/ocober/27/aricle/2022.hml [17] Michel. A Airline alliances: Effecs on oal revenues of sharing esimaed nework value of unsold seas. Proceedings of Inernaional Conference on Reserch in Air Transporaion. Berkeley, California. May 22-25, [18] Michel, A Airline Alliance Revenue Managemen:Improving Join Revenues hrough Parner Sharing of Fligh Leg Opporuniy Coss Maser s hesis, Massachuses Insiue of Technology, Cambridge, MA. [19] Seer Davies Gleave, Compeiion impac of airline code-share agreemens, p. 7, hp://ec.europa.eu/compeiion/secors/ranspor/ repors/airlinecode share.pdf [20] Topaloglu, H., A dualiy based approach for nework revenue managemen in airline alliances. Journal of Revenue and Pricing Managemen (forhcoming wih advance online publicaion). [21] Vinod, B Alliance Revenue Managemen.Journal of Revenue and Pricing Managemen. 4(1) [22] Williamson, E. L Airline Nework Sea Invenory Conrol: Mehodologies and Revenue Impacs. Ph.D. hesis, Massachuses Insiue of Technology, Cambridge, MA. [23] Wrigh, C. P., H. Groenevel, R.A. Shumsky Dynamic Revenue Managemen in Airline Alliances. Transporaion Science. 44(1) 15-3 [24] Wrigh, C. P., Modeling and analysis of revenue managemen in airline alliances. Ph.D. hesis, William E. Simon Graduae School of Business Adminisraion a he Universiy of Rocheser.