optimal pricing and seat allocation in airline industry...
TRANSCRIPT
Optimal Pricing and Seat Allocation in Airline Industry Under Market Competition
PhD Thesis Oral Exam PresentationSyed Asif Raza
Outline
Revenue Management (RM)Background and LiteratureModel and ResultsConclusion and Future WorksQuestions
Airline RM
Airline Schedule Design and Fleet Assignment
Airline Revenue Management
Forecasting
Pricing
OverbookingResearch
Seat Inventory
Static Seat Inventory Control
Dynamic SeatInventory Control
Static Pricing
Dynamic Pricing
Revenue Management is simply the practice of maximising Revenue from Perishable assets (seats) through a combination of Pricing & Inventory controls
Fares
Revenue Management
Pricing Reservation Control
Services/ Restrictions
Seat Inventory Control Overbooking
Itinerary Control Fare Class Mix
Segment OD Point-of-Sale
No Direct Connection
Market Competition
Ignored
Missing Connections in Airline RM Studies
Review Articles:McGill and Van Ryzin (1999)- Revenue Management: Research overview and prospects.Pak and Piersma (2002)- Overview of OR techniques in airline RM.Silver (2004)- An overview of heuristic solution methodsBarnhart et al. (2003) – Application of operations research in the air transport industry.Elmagraby and Keskinocak (2003) -Dynamic pricing in the presence of inventory considerations: Research overview, current practices, and future directionsBitran and Caldentey (2003)- An overview of pricing models for Revenue Management.
Cote et al. (2003)- Mathematical optimization approach for integrated seat allocation and pricing in airline industry.
Bi-level mathematical optimization model is presented for joint determination of fare pricing and seat allocationThe customer demand is considered price insensitive and the model is non dynamic
Dai et al. (2005)- Multiple firms competition in RM context.Competitive prices are determined for both the deterministic and stochastic price sensitive demand for a known capacity. Mostly the analysis consider duopoly competition
Chen et al.(2005) – Newsvendor pricing gameJointly determined the competitive pricing and capacity for the competing firms for a single commodity Unique Nash equilibrium is determined analytically
Competition in RM
Competition in Airline RM:Hotelling (1929)-Scheduling decision under competition Borenstein et al. (2003)- An empirical paper, focus on broad competition problem but ignore any seat allocation/ pricing modeling.Belobaba and Wilson (1997)- Simulation model to study scheduling strategy under market competition.Lippman and McCardle (1997)- two-firm pricing competition.Teodorovic and Krcmar-Nozic (1989)- Multi-criteria model to determine flight frequencies on an airline network under competitive conditions.
Objectives of thesis
Develop new models for joint determination of competitive fare pricing and seat allocation in airline IndustryDevelop new solution methodologies for joint determination of pricing and capacity in RM
Airline RM Competition Models
Airline RM game is developed mainly in duopoly marketTwo game theoretic models are presentedThe models are studied for both the cooperative and non-cooperative game settings.Numerical analysis with statistical DOE
Distribution Free Approach is RMAn alternative approach for joint determination of pricing and capacity under monopolyDetermines lower bound (worst possible) revenue estimateThe approach is addressed to:
Standard Newsvendor problem Extension to Shortage cost PenaltyExtension to Shortage and Holding costExtension to recourse caseExtension random yieldExtension to multiple items
],[
],[1]E[ξ]E[ξ
Where}2,1{,ξDS}2,1{,ξDS
ModeltiveMultiplica
Hi
Li
HiLi
HiHiHi
LiLiLi
HiHi
LiLi
ii
ξξξ
ξξξ
∈
∈
==
=∀==∀=
],[
],[0]E[ξ]E[ξ
Where}2,1{,ξDS
}2,1{,ξDSModel Additive
Hi
Li
HiLi
HiHiHi
LiLiLi
HiHi
LiLi
ii
ξξξ
ξξξ
∈
∈
==
=∀+==∀+=
assumedison Distributi A Uniform(IGFR) Rate Failure dGeneralize IncreasingfollowandBoth HiLi ξξ
Modeling Approaches
0D,0D,0D and0D,0D,0D that assumed also isIt
},2,1{,,,0,,
},2,1{,,,0,,
, functions demand ticdeterminsi sensitive pricelinear with modeled isDemandThe
21
H1
2
H1
1
H1
21
L1
2
L1
1
L1 ≥∂∂
∂≥
∂∂
≤∂∂
≥∂∂
∂≥
∂∂
≤∂∂
=≠>>
=≠>>
+−=+−=
HHHHLLLL
HijHiHijHi
LijLiLijLi
HjHijHiHiHiHiLjLijLiLiLiLi
PPPPPPPP
jiji
jiji
PPDPPD
θβθβ
θβθβ
θβαθβα
Assumptions
iHiLi
DB
LiLiii
Hi
y
HiHiLi
DB
LiLiiiHi
iLiiHiHiHi
BDdCy
ddBCP
BSCSPEE
Lii
Li
i
Hi
Lii
Li
HiLi
−−Φ+=
⎟⎟
⎠
⎞
⎜⎜
⎝
⎛Φ−Φ+−=
−=Π
∫
∫∫
−
−
ξξ
ξξξξ
ξ
ξξ
ξξ
)(
Where
)()(
}}],min{,min{[
Li
DB
LiLiLiiLi
iLiLiLiLi
iLiLiLi
dPBP
BSEPSEP
BSEP
Lii
Li
LiLi
Li
ξξξ
ξξ
ξ
∫−
+
Φ−=
−−=
=Π
)(
][][
}],[min{
( ) ( )
Hi
y
HiHiHi
Li
DB
LiLiLiHiiLiHiiHii
dP
dPPBPPCP
i
Hi
Lii
Li
ξξ
ξξ
ξ
ξ
∫
∫
Φ−
Φ−+−−=Π−
)(
)(
Additive Model
( ) ( )
Li
DB
LiLiLiiii
Hi
Dy
HiHiHiHi
Li
DB
LiLiLiLiHiiLiHiiHii
dDBCy
Where
dDP
dDPPBPPCP
Lii
Hii
Lii
ξξ
ξξ
ξξ
∫Φ+−=
Φ−
Φ−+−−=Π
∫
∫
/
0
/
0
/
0
)(
)(
)(
Multiplicative Model
Game Theory
Environment Certain
Dynamic
Static
1 DM
> 1 DM
1 DM
> 1 DMNon
Cooperative
Non Cooperative
Cooperative
Cooperative
Optimal Control
Optimization
No Side Payments
Side Payments
Side Payments
No Side Payments
Competition Analysis
Non-cooperative analysisUniqueness of Nash Equilibrium (NE) Numerical analysis with a Statistical Design Of Experiment (s) (DOE).
Cooperative analysisBargaining game with No Side Payment options (NSP)Bargaining game with Side Payment options (SP)
mequilibriuNash Unique
mequilibriuNash Unique
commitment-prelimit booking aFor ly.respective 2player
and 1player of setsstrategy therepresent X and X Where)x,(xf )x,(xf and )x,(xf )x,(xf
:X xallfor and X xallfor satisfied are esinequaliti ofpair following theif (NE) mEquilibriu
Nash a constitute tosaid is )x,(x strategies ofpair A
21
H12
21
H12
21
L12
21
L12
21
2N12
N2
N12
N211
N2
N11
2211
N2
N1
⇔∂∂Π∂
≥∂Π∂
⇔∂∂Π∂
≥∂Π∂
≥≥
∈∈
HHH
LLL
PPP
PPP
Nash Equilibrium
( )( )
NE
NE
NENE
22
11
2211
toSubjectMax
OptionPayment Side No
Π≥Π
Π≥Π
Π−ΠΠ−Π
( )FCNENEFC
NE
NE
2211
22
11
21
21SP
PaymentsSide
toSubjectMax
Option Payments SideWith
Π−Π+Π−Π=
Π≥Π
Π≥Π
Π+Π
Nash Bargain Solution
Numerical Analysis
}2,1{,]2,0[],[],[ModeltiveMultiplicaFor
}2,1{,]30,30[],[],[ModelAdditiveFor
=∀==
=∀−==
i
i
HiHiLiLi
HiHiLiLi
ξξξξ
ξξξξ
jijii
iPPPPC
HijLij
HiLiHiLi
HiHiLiLii
≠=∀===∀====
=∀=====
},2,1{,,10.0,15.0}2,1{,15.0,25.0,40,60
}2,1{,200,100,100,0,100
θθββαα
13355.11200.26171.5685.88Airline 2
13355.13200.26171.5685.91Airline 1Normal
13608.25208.32175.5084.90Airline 2
13608.25208.32175.5084.90Airline 1UniformMultiplicative
13349.00200.04171.4772.90Airline 2
13349.00200.04171.4772.89Airline 1Normal
13570.21205.18176.5372.35Airline 2
13570.21205.18176.5372.35Airline 1UniformAdditive
PayoffHigh Fare PriceLow Fare PriceBooking
LimitAirlineDistributionModel
Non-Cooperative Analysis (Symmetric market)
[Additive (-), Multiplicative (+)]Model Type (I)
[0.05 (-), 0.15 (+)]H21
[0.1(-), 0.2 (+)]L21
[0.1(-), 0.2 (+)]βH2
[0.2 (-), 0.3 (+)]βL2
[80 (-), 120 (+)]C2
LevelsParameters
Non-Cooperative Analysis (Asymmetric market)
21221222
21221221
21222
21212221
21221222
212221
2122121222
212212221
48.4074.3036.5485.1253.273ˆ05.18I99.779.1359.2383.663.235ˆ
83.1033.1863.13169ˆ882.4449.4256.6156.7050.6343.178ˆ
I7.605I946.7677.9033.10998.69ˆ858.0I172.5019.1329.1061.76ˆ
1724282111823080118215234ˆ922867133753759814861ˆ
HHHHH
HHHHH
LLL
HLHLL
HHHH
HHHL
HHHLHL
HHHHL
CP
CP
CP
CP
CB
B
C
θβθβ
θβθβ
θβ
θθββ
θβθβ
θβββ
θβθθββ
θβθββ
−+−−=
−++−−=
+−−=
++−−−=
+−++=
+++−=
−++−−=Π
−+−−−=Π
DOE-Non Cooperative Game
2162.22162.2Gain Of Cooperation
15732.42368.2820070.0015732.42368.2820070NBS with side payments
13570.21259.43156.0455.1113570.21255.81169.5754.05NBS no side payments
13570.21205.18176.5372.3513570.21205.18176.5372.35Non-cooperative
PayoffPH2PL2B2PayoffPH1PL1B1
Airline 2Airline 1
Cooperative Analysis (Bargaining Game)Additive
1604.001604.00Gain Of Cooperation
15212.24310.10200.0080.0015212.24310.10220080NBS with side payments
15212.24310.10200.0080.0015212.24310.10220080NBS no side payments
13608.25208.32175.5084.9013608.25208.32175.5084.90Non-cooperative
PayoffPH2PL2B2PayoffPH1PL1B1
Airline 2Airline 1
Cooperative Analysis (Bargaining Game)
Multiplicative
Comparing NSP with NE
Airline 1Revenue improves about 5.4%1.9% more seats for High fare customersLow and high fare prices increase by 4.8% and 35% respectively
Airline 2Revenue improves 6.5%5.3% more seats for high fare classLow and high fare prices increase is 9.6% and 28.4%
Comparing SP with NSP
Airline 1Revenue improves about 6%1.9% more seats for High fare customersLow and high fare prices increase by 6.7% and 8.6% respectively
Airline 2Revenue improves 5.2%1.6% more seat to high fare class but not significant.Low and high fare prices increase is 6.3% and 5.8% respectively
Newsvendor Problem and RMYao (2002)’ s thesis investigates newsvendor problem (NVP) with pricing in the context of SCM and reported that the elegant structure of the problem also finds applications in RM.In standard newsvendor problem optimal quantity is determined for a single perishable asset with stochastic demand having an objective to maximize the revenue
Proposed Extension to NVP
Investigate alternative solution approach to NVP with pricing using Distribution Free Approach.Using Distribution Free approach study several extensions to NVP with application in RM
Distribution Based Vs. Distribution Free Approach
Distribution BasedModeling requires precise information about demand distributionIt maximizes the revenue under a known distribution
Distribution FreeModeling requires only the mean and standard error estimate of the demand.It maximizes the revenue under the worst possible demand distribution.
Newsvendor problem with Pricing
( )( ) ( )
( ) ( )( )2
)(),(
2][
Where][)(
]},min{[),(
:ModelAdditive
2/122**
2/122
LBLBLBLBLBLBLBLB
QDQDPQCPQP
QDQDRDQE
RDQEPQCP
QCRDQPEQPξDRD
−−−+−−=Π
−−−+≤−
−−−=
−=Π+=
+
+
σ
σξ
ξ
ξ
Newsvendor problem with Pricing
( )( ) ( )
( ) ( )( ) ( )2
/1/1),(
2/1/1]/[
Where]/[)(
]},min{[),(
:Model tiveMultiplica
2/122**
2/122
DQDQDPQCPQP
DQDQDQE
ξDQEDPQCP
QCRDQPEQPξDRD
LBLBLBLBLBLB−−−+
−−=Π
−−−+≤−
−−−=
−=Π=
+
+
σ
σξξ
ξ
ξ
Proving concavity
)2('2),2);
modeltiveMultiplicaFor'4
),);
modelAdditiveFor
0ΠΠΠ)
0Π,0Π)
if,concavequasiisΠ
LB
LB
LB2
2LB
2
2LB
2
2LB
2
2LB
2LB
−≥≤≤−Π
−≥≥−Π
≥∂∂
∂−
∂∂
∂∂
=
≤∂∂
≤∂∂
−
σσ
σ
DDPiiDQDiifconcavequasiis
DPiiDQiifconcavequasiis
QPQPHii
QPi
LBLB
LBLB
LBLBLBLB
LBLB
Extension to Holding and Shortage cost case:
( )( )
( ) ( )( )2
)(
),(
][)(
]][][},min{[),(
2/122
**
LBLBLB
LBLBLBLBLB
QDQDSGP
DSQCSPQP
DSRDQEGSPQCSP
DRDSRDQGQCRDQPEQP
−−−+++−
−−+=Π
−−++−−+=
−−−−−=Π+
++
σ
ξ
ξ
Proving concavity
( )
( )SGDDPiiDQDiifconcavequasiis
DSGPiiDQiifconcavequasiis
LBLB
LBLB
+−−
≥≤≤−Π
−+≥≥−Π
)2('2),2);
modeltiveMultiplicaFor'4
),);
modelAdditiveFor
LB
LB
σσ
σ
Extension to Shortage cost case:
( )( )
( ) ( )( )
'4),);
2)(
),(
][)(
]][},min{[),(
:ModelAdditive
LB
2/122
**
DSPiiDQiifconcavequasiis
QDQDSP
DSQCSPQP
DSRDQESPQCSP
DRDSQCRDQPEQPξDRD
LBLB
LBLBLB
LBLBLBLBLB
σ
σ
ξ
ξ
−≥≥−Π
−−−++−
−−+=Π
−−+−−+=
−−−=Π
+=
+
+
Recourse case:
( )( )
( ) ( )( )
LBrLBLB
LBLBrLB
rLBrLBLBLBLB
rrr
PCiiiD
PiiDQiifconcavequasiis
QDQDCP
DCQCCPQP
DCRDQECPQCCPQPξDRD
<−≥≥−Π
−−−++−
−−+=Π
−−+−+−=Π
+=+
)'4
),);
2)(
),(
][)(),(
:ModelAdditive
LB
2/122
**
σ
σ
ξ
Random yield case:
( )( ) ( )( )
IPEis,;0Π)
2/1;0Π)
2)(1
),(
])([)(
]}),(min{[),(
:ModelAdditive
2LB
2
2LB
2
2/122
**
DDQifQ
ii
ifP
i
DQDQQP
QCPQP
RDQGEPQCP
QCRDQGPEQPξDRD
LBLB
LB
LBLBLBLB
LBLBLBLBLB
ρ
σ
ρρρρσ
ρ
ρ ξ
ξ
≥≤∂∂
≥≤∂∂
−+−+−+−
−=Π
−−−=
−=Π+=
+
Multiple Product case:
( ) ( )( ) ( )
∑
∑
∑
∑
=
=
=
=
≤
−−−+−−=Π
≤
−=Π
n
1ii
2/122
1
**
1
1
C
:toSubject2
),(
:toSubject
]},min{[),(
BQ
QDQDPQCPQP
BQC
QCRDQPEQP
i
ii
iii
i
LB
LBiLBiiLB
n
iLBiLBLBLBLB
n
iii
n
iiiiii
σ
ξ
Numerical Analysis
2] U(0,~,],[ model, tivemultiplicaFor
30] U(0,~,],[ model, additiveFor
P - D 15,000],U[10,000,~B 0.9], U[0.5,~C1.4] U[1.2,~CC,0.3] U[0.2,~GC,0.2] U[0.1,~S
0.3]U[0.05,~ 200],U[100,~ 100], U[20,~Cproblems generatedrandomly 100
:followsextension each ation withExperiment
r
ξξξξ
ξξξξ
βαρ
βα
=
=
=×××
Standard Newsvendor Problem
Additive Model
UniformEVAI Improves revenue 0.25%
13% over-stocking
0.85% under-pricing
NormalEVAI Improves revenue 0.13%
3% over-stocking
0.81% under-pricing
Multiplicative Model
UniformEVAI Improves revenue 0.75%
0.59% under-stocking
0.01% over-pricing
Normal
EVAI Improves revenue 0.60%
0.13% over-stocking
0.30% over-pricing
Extension to Shortage and Holding Cost
Additive Model
UniformEVAI Improves revenue 0.20%
0.48% under-stocking
0.89% under-pricing
NormalEVAI Improves revenue 0.09%
2.3% over-stocking
0.85% under-pricing
Multiplicative Model
UniformEVAI Improves revenue 0.48%
2.45% under-stocking
0.19% over-pricing
NormalEVAI Improves revenue 0.38%
1.68% over-stocking
0.30% over-pricing
Extension …
Recourse Case
UniformEVAI Improves revenue 0.29%
0.86% over-stocking
0.81% under-pricing
NormalEVAI Improves revenue 0.17%
3.5% over-stocking
0.77% under-pricing
Random Yield Case
UniformEVAI Improves revenue 0.42%
0.53% over-stocking
0.80% under-pricing
Normal
EVAI Improves revenue 0.22%
3.27% over-stocking
0.76% under-pricing
Conclusion
Competition models in Airline Revenue Management (RM) are proposed using game theoretic approach.The models enable us to jointly determine the booking limits and fare pricing for airlines in the game.Both non-cooperative and cooperative bargaining games are considered.
Conclusion (Contd.)
Numerical study shows that cooperation results superior revenue gain to airlinesCooperation with side payments is also
found superior to cooperation with no side payments option in the bargaining game.The use of distribution free approach for pricing in RM is explored
Conclusion (Contd.)
The approach establishes lower bound estimate on revenue and enables maximizing the revenue under worst possible demand distribution.The approach is applied to standard
newsvendor problem and many other extension
Conclusion (Contd.)
It is also reported that the use of the approach on extensions of standard newsvendor problem has direct implications on RM practice in aviation and other service industries
Future Works
Dynamic pricing with seat allocation Pricing with:
Re-saleable returnCustomer choice model
Research ArticlesA. S. Raza, M. U. Al-Turki, S. Z. Selim: 2007, “Early Tardy Minimization of Joint Scheduling of Jobs and Maintenance Operations on a Single Machine” special issue on scheduling in manufacturing, information and service industries in International Journal of Operations Research, 4, 1-10A. S. Raza, A. Akgunduz, M.Y. Chen: 2006, “A TabuSearch Algorithm for Solving Economic Lot Scheduling Problem” Journal of Heuristics, 12, 413-426A. S. Raza, A. Akgunduz: 2005, “The Use of Meta-Heuristics to Solve Economic Lot Scheduling Problem”, Lecture Notes in Computer Science, 3448, 190-201
Research Articles (Under review)
A. S. Raza, A. Akgunduz, “An Airline Revenue Management Fare Pricing Game with Seat Allocation”under second revision with International Journal of Revenue ManagementA. S. Raza, A. Akgunduz, “A Comparative Study of Heuristic Algorithms on Economic Lot Scheduling Problem” submitted to Computers and Industrial EngineeringA. S. Raza, A. Akgunduz, “A Distribution Free Approach for Jointly Controlling Price and Capacity in Revenue Management”, submitted to International Journal of Production EconomicsA. S. Raza, A. Akgunduz, “The Impact of Fare Pricing Cooperation in Airline Revenue Management”, submitted to INFOR