Dynamic Coordination of Revenue Management in Airline Alliances

Transcription

1 Dynamic Coordination of Reenue Management in Airline Alliances Christopher Wright Simon Graduate School of Business Uniersity of Rochester Harry Groeneelt Simon Graduate School of Business Uniersity of Rochester Robert Shumsy Tuc School of Business Dartmouth College

2 Presentation Outline Emergence of Alliances Related Literature Reenue Sharing in Free Sale Current Methods Alliance Model Results Benefits of Dynamic Trading Future Research 2

3 Emergence of Alliances First seen after deregulation of the U.S. airline industry in 980s Now, important part of airlines operations e.g., Star Alliance, SyTeam, oneworld U.S. domestic airline alliances in 2004: 20% of domestic passenger coupons with different ticeting airline than operating airline (code sharing in an alliance). 46% of domestic reenue collected from ticets including a code-shared flight leg. 3

4 Code sharing in U.S. ( ) fraction of all coupons Code-share flight operated by... national networ airline regional independent regional subsidiary Source: Analysis of DOT Database B (0% ticet sample) 4

5 Related Literature Alliance Reenue Management Boyd (998) First to analyze problem Established sharing mechanisms Static LP of soft bloc control Vinod (2005) Etensie discussion of implementation issues Some technical analysis of reenue sharing Model Talluri and an Ryzin (998) Single airline networ MDP 5

6 Sharing Mechanisms (Boyd 998) Hard blocs: airlines pre-buy a fied number of seats from each other. Free sale: airlines maintain own inentory, but hae access to each other s seats to fill an itinerary, paying a transfer price. Dynamic trading (soft blocs): airlines trade ownership of each other s seats throughout horizon, controlling those they own. Boundary ownership may be free or not. 6

7 Current Method in Free Sale Typically, static proration contracts e.g., special prorate agreements (SPA), mileage proration Fails to account for changing marginal alues Must renegotiate w/ long term system changes We as: Can we design practical and effectie schemes to dynamically coordinate alliance reenue management decisions? Is there an equilibrium in each scheme? How do reenues compare to central control? How are reenues split between partners? 7

8 Optimality and Transfer Prices Counter-eample to transfer price optimality Airline operates flight A, seat remains Airline 2 operates flight B, seat remains Period Itinerary Reenue Prob. Opp. Costs 2 ( 0) (0 ) $250 $ (400 p, p) ( ) to A $400.0 (0,0) To preent local sales: 250 < p 250< p < 50 CONTRADICTION! 250 < 400 p 8

9 Alliance Model Maro game K periods, = K, K-,, 2, 2 Airlines, i {, 2} Each operates m i flights (O/D, time) n itineraries, N n local itineraries with Airline s flights, n 2 local itineraries with Airline 2 s flights, n C connecting itineraries with both Airline s flights, N C N N 2 9

10 Alliance Model Subscripts and superscripts (typical) i i = airline = itinerary = period 0

11 Alliance Model Why a non-cooperatie game? At the highest leel, alliance partners hammer out the terms of the alliance (a cooperate process), but any agreement must be implemented within each airline's reenue management system. Each system (myopically) maimizes profits for that airline.

12 Alliance Model Arrial/reenue process: Airline i receies request for itinerary in period i with probability θ The reenue associate with the request is Eogenous with nown distribution ~ i Complementary CDF: F r F Decision process: i () () r PDF: i R Each (operating) airline posts transfer prices for required sub-itineraries (partner price): p i At most one (mareting) airline receies request i with net reenue r net ; accepts/reects: u r ; f i () r net 2

13 Alliance Model Value function: For each partner: For the alliance: Opportunity cost of selling inentory: Airline i s for itinerary : ΔJ i A J ( ) 2 = J + J ( ) J i ( ) i i A = J J A ; inentory used for itinerary ( note : A i sub - itinerary of operated by airline i) i i A = A + A 3

14 Alliance Model Possible sharing schemes: Partner price: transfer prices for each itinerary set dynamically by the operating airline. Bid price: mareting airline pays the marginal alue (bid price) of the operating airline's seats to the operating airline (Vinod 2005) Dynamic proration: reenues split proportionally based on their fractions of total opportunity cost i.e., In time period for itinerary, airline i receies: ΔJ i ΔJ A A J A r, 2, + Δ, 4

15 5 Sample Value Function (partner price) { } { } [ ] { } [ ] = R + J q R u A J q p R u A J p R pu q p R u A J p R u p R q R u A J R u R q p u J N N p N D u N D u C C ; E ; ; E ma E ma E ma, { } u A u D J r r r R = = : {0,} : and 0 ) ( 0

16 Free Sale Results Centralized control (first-best): Behae as one airline with combined networ. Accept if and only if reenue is more than alliance opportunity cost: r ΔJ, A, A u r; =, 0 otherwise N 6

17 Free Sale Results Decentralized control (second-best): For each scheme, there eists a unique, purestrategy Maro-perfect equilibrium. Mareting airline accepts request if and only if its net reenue eceeds its opportunity cost: u i i r Δ net J, A r ; = net where r 0 net = r p otherwise i r ( ), N N C i A 7

18 Free Sale Local Results Any transfer price scheme suffers from inefficient local decisions Local decision is similar to single networ: i i r ΔJ, A, A u r; =, N 0 otherwise i But, unlie single networ, alliance opp. cost is: ( ) i i ΔJ, A = ΔJ, A + ΔJ, A st order effect 2nd order effect So, critical alue may be too high or too low 2 nd order effect can be pos. or neg. 8

19 Free Sale Local Results Eample of pos/neg 2 nd order effects Airline operates flights A & B and sells itineraries B & AB Airline 2 operates flight C and sells itinerary AC 2 ΔJ, A ( AB ) 0 ( B, ) 0 2 ΔJ A A B C 9

20 Free Sale Connecting Results For Partner Price scheme: ~ i If () i F r / f r has a slope <, then the unique equilibrium price is gien by: ~ i F ( i J ( A ) i p ( ) ) i i Δ, + p = ΔJ, A + i i i f ΔJ, A + p ( ) Airline i s posted transfer price Airline i s opportunity cost Premium added by Airline i to capture some of the ecess reenue Mareting airline, thus, accepts if and only if: i i r ΔJ A + p = ΔJ, A, premium + Partner s opportunity cost Total alliance opportunity cost 20

21 Free Sale Connecting Results For Bid Price scheme: Transfer price is the opportunity cost: Full alue: Sub-itinerary: i i = ΔJ A i i i p = ΔJ, A p, Mareting airline, thus, accepts if and only if: i i r ΔJ A p ( ), + ( ) ΔJ, A = ( ) ( ) i i ΔJ, A ΔJ, A nd order effect of mareting airline' s sub-itinerary 2

22 Free Sale Connecting Results For Dynamic Proration scheme: Each airline will agree to sale if and only if: ΔJ ΔJ ( ), A ( ), A i i r ΔJ, A Request is, thus, accepted if and only if: r ΔJ, ( ) A 22

23 Free Sale Results Using strategic alue (partner price) s. marginal alue (bid price): Receiing only the marginal alue does not coney the partner's alue of inentory. Partner pricing allows the operating airline to capture some of the partner's reenue (lin). But, premium causes mareting airline to reect connection requests that would benefit the alliance. Dynamic proration proides some lin in alues w/o reecting beneficial connections. 23

24 Shift From Local To Connecting Flight A operated by airline (0 seats) Flight B operated by airline 2 (0 seats) Local demand for each flight goes to operating airline Connecting demand for itinerary AB split 50/50 Relatie demand aried from all connecting (AB) to all local (A, B) 24

25 Shift From Local To Connecting % of Optimal (Centralized Control) Reenue 00% 97% 94% Partner Price Bid Price (SI) Dynamic Proration 9% Fraction of Requests for Local Itineraries 25

26 Changing Connection Dominance Flight A operated by airline (0 seats) Flight B operated by airline 2 (0 seats) Local demand for itinerary A airline only Connecting demand for itinerary AB aried from airline 2 to airline : airline s portion airline 2 s portion 26

27 Changing Connection Dominance % of Optimal (Centralized Control) Reenue 00% 95% 90% Partner Price Bid Price (SI) Dynamic Proration 85% Fraction of Connecting Requests to Airline 27

28 Free Sale Results Distribution of net reenues Bid price: all net reenues are retained by the mareting airline. Partner price: decision power is shifted to operating airline; as are net reenues. Dynamic proration: distributes net reenues according to relatie contribution (opp. cost). 28

29 Value of Dynamic Trading Airlines own seats partner s flights. They buy and sell seats between themseles as time passes and demand is realized. Essentially trading options on each flight. Mechanisms Best case: 'oracle' finds alliance-optimal ector of echanges in each period. Gains allocated so both partners are willing to mae the trade. Myopic (greedy) case: partners buy/sell seats in order of decreasing marginal alue. 29

30 Shift From Local To Connecting % of Optimal (Centralized Control) Reenue 00% 99% No Trades (Dyn. Pro.) Optimal Trades 98% Myopic Trades 97% Fraction of Requests for Local Itineraries 30

31 Value of Dynamic Trading Preliminary results Best case: near-optimal performance (losses due only to integrality of inentory). Most (ecept boundary) decisions essentially local 2 nd order effects reduced to zero Myopic case: performance depends on alliance parameters (local maima); can perform well Only initial trade is liely to be large ; suggests initial best trade w/ periodic myopic updates Performance depends on choice of boundary trading scheme. 3

32 Conclusions Dynamic reenue management schemes can perform well, but best mechanism depends on alliance parameters. Distribution of reenues differs greatly; important for participation negotiations. Dynamic trading can enhance results, eeping inentory with airline that can best use it. 32

33 Future Research Approimation methods Asymptotic analysis Multiple partners Additional applications 33

34 References Boyd, E. A., Airline Alliances, OR/MS Today, October 998, Vinod, B., Alliance Reenue Management, Journal of Reenue and Pricing Management (Practice Papers), 4(), April 2005, Talluri, K. and G. an Ryzin, An Analysis of Bid-Price Controls for Networ Reenue Management, Management Science, 44 (), Noember 998,