OPTIMAL AIRLINE FLEET PLANNING AND MANAGEMENT STRATEGIES UNDER STOCHASTIC DEMAND TEOH LAY ENG DOCTOR OF PHILOSOPHY IN ENGINEERING

Transcription

1 OPTIMAL AIRLINE FLEET PLANNING AND MANAGEMENT STRATEGIES UNDER STOCHASTIC DEMAND TEOH LAY ENG DOCTOR OF PHILOSOPHY IN ENGINEERING LEE KONG CHIAN FACULTY OF ENGINEERING AND SCIENCE UNIVERSITI TUNKU ABDUL RAHMAN MAY 2015 ii

2 OPTIMAL AIRLINE FLEET PLANNING AND MANAGEMENT STRATEGIES UNDER STOCHASTIC DEMAND By TEOH LAY ENG A hesis submied o he Deparmen of Civil Engineering, Lee Kong Chian Faculy of Engineering and Science, Universii Tunku Abdul Rahman, in parial fulfillmen of he requiremens for he degree of Docor of Philosophy in Engineering May 2015 iii

3 ABSTRACT OPTIMAL AIRLINE FLEET PLANNING AND MANAGEMENT STRATEGIES UNDER STOCHASTIC DEMAND Teoh Lay Eng The sochasic naure of he world has posed significan challenges o such a compeiive airline indusry. There are many unexpeced evens, e.g. fuel price volailiy and naural disaser ha could affec airline s ravel demand and profi margin. As such, how airlines make a sraegic flee planning decision o mee sochasic demand profiably is imporan. To properly capure supplydemand ineracion, raveler's response and subjecive percepion of airline's managemen are significan o assure an adequae flee supply. Besides, i is imporan o noe ha aircraf operaions are sricly conrolled under regulaed limis a some airpors and hence airlines cerainly require a proper flee planning (by incorporaing opimal slo purchase) o mee increasing demand wih addiional service frequency. In addiion, he environmen should no be compromised in flee planning. By having a green flee in operaions, a winwin siuaion beween airlines and he environmen could be achieved. Wih he aim o solve he flee planning problem sraegically, a novel mehodology is developed o formulae long-erm flee planning model, in he form of probabilisic dynamic programming model, o deermine he opimal quaniy of he respecive aircraf ype (wih corresponding service frequency) ii

4 o be acquired/leased under uncerainy. By developing a modeling framework of sochasic demand, he level of demand could be deermined realisically. Besides, mode choice modeling and Analyic Hierarchy Process are adoped o comprehend supply-demand ineracions in greaer deail so ha airline's flee supply is sufficienly adequae o mee sochasic demand. To consider muliple crieria in making flee planning decision, bi-objecive and wo-sage flee planning models are formulaed mahemaically o opimize he flee planning problem. By examining numerous case sudies, i was found ha he resuls are comparable wih airline s acual performance and he findings showed ha he developed mehodologies are pracically viable o assure airline's susainabiliy in erms of economy, social and environmen. iii

5 ACKNOWLEDGEMENTS Firs of all, I would like o express my sincere appreciaion o my supervisor, Associae Professor Ir Dr Khoo Hooi Ling for her kind guidance, precious advices, consrucive commens and ideas hroughou he research. This hesis would no be possible wihou her. I would also like o convey my graiude o Universii Tunku Abdul Rahman for graning research scholarship as well as research gran (IPSR/RMC/UTARRF/C1-10/T3). Besides, I would like o hank he Minisry of Educaion, Malaysia for supporing he research hrough he Fundamenal Research Gran Scheme (FRGS/1/2012/TK08/UTAR/03/3). Las bu no he leas, I would like o express my deep appreciaion o my parens, family members and friends for heir suppors, paience and undersanding hroughou my candidaure. iv

6 APPROVAL SHEET This hesis eniled OPTIMAL AIRLINE FLEET PLANNING AND MANAGEMENT STRATEGIES UNDER STOCHASTIC DEMAND was prepared by TEOH LAY ENG and submied as parial fulfillmen of he requiremens for he degree of Docor of Philosophy in Engineering a Universii Tunku Abdul Rahman. Approved by: (Assoc. Prof. Ir. Dr. KHOO HOOI LING) Dae:... Supervisor Deparmen of Civil Engineering Lee Kong Chian Faculy of Engineering and Science Universii Tunku Abdul Rahman v

7 LEE KONG CHIAN FACULTY OF ENGINEERING AND SCIENCE UNIVERSITI TUNKU ABDUL RAHMAN Dae: SUBMISSION OF THESIS I is hereby cerified ha Teoh Lay Eng (ID No: 09UED09073) has compleed his hesis eniled Opimal Airline Flee Planning and Managemen Sraegies under Sochasic Demand under he supervision of Assoc. Prof. Ir. Dr. Khoo Hooi Ling (Supervisor) from he Deparmen of Civil Engineering, Lee Kong Chian Faculy of Engineering and Science. I undersand ha Universiy will upload sofcopy of my hesis in pdf forma ino UTAR Insiuional Reposiory, which may be made accessible o UTAR communiy and public. Yours ruly, (Teoh Lay Eng) vi

8 DECLARATION I TEOH LAY ENG hereby declare ha he disseraion is based on my original work excep for quoaions and ciaions which have been duly acknowledged. I also declare ha i has no been previously or concurrenly submied for any oher degree a UTAR or oher insiuions. (TEOH LAY ENG) Dae: vii

9 TABLE OF CONTENTS Page ABSTRACT ACKNOWLEDGEMENTS APPROVAL SHEET PERMISSION SHEET DECLARATION TABLE OF CONTENTS LIST OF TABLES LIST OF FIGURES LIST OF ABBREVIATIONS ii iv v vi vii viii xii xv xvi CHAPTER 1.0 INTRODUCTION Background Research Objecives Research Scope Thesis Overview LITERATURE REVIEW Travel Demand Forecasing: Deerminisic vs Sochasic Airline Flee Planning Approach Sraegic Flee Planning Modeling Framework Mode Choice Analysis: Air and Ground Transpor Analyic Hierarchy Process: A Tool o Quanify he 35 Key Aspecs of Flee Planning Decision-Making 2.4 Service Frequency and Slo Purchase in Flee Planning Service Frequency Deerminaion in Flee Planning Slo Purchase Green Flee Planning Environmenal Issue of Air Transpor Sysem Miigaion Sraegies Environmenal Assessmen Approaches Summary FLEET PLANNING DECISION MODEL UNDER 64 STOCHASTIC DEMAND 3.1 Making Opimal Aircraf Acquisiion and Leasing 64 Decision under Sochasic Demand 3.2 Modeling Sochasic Demand under Uncerainy An Illusraive Example (To Deermine he 70 Probabiliy of he Occurrence of Unexpeced Evens) viii

10 3.3 Aircraf Acquisiion Decision Model Consrains Objecive Funcion Probable Phenomena in Flee Planning Problem Formulaion Soluion Mehod An Illusraive Case Sudy: Linear Programming 86 Model Resuls and Discussions Summary Aircraf Acquisiion and Leasing Decision Model Consrains Objecive Funcion Problem Formulaion Lower Bound and Opimal Soluions Soluion Mehod An Illusraive Case Sudy: Nonlinear 105 Programming Model Inpus for Sochasic Demand Modeling Inpus for Aircraf Acquisiion and 107 Leasing Decision Model Resuls and Discussions Summary STRATEGIC FLEET PLANNING MODELING 123 FRAMEWORK 4.1 Supply-demand Ineracion In Flee Planning Mode Choice Analysis Local and Trans-border Trip Saed Preference Survey Experimenal Design Traveling Aribues Quesionnaire and Respondens Modeling Approach: Mulinomial Logi Models Findings: Mode Share of Trips Analysis of LCC's Impacs on Mode Choice 138 Decision Impac of LCC on FSC Impac of LCC on Ground Transpor Effec of Socioeconomic Background Implicaions for Managerial Pracices Summary Analyic Hierarchy Process (AHP) Modeling Framework The Role of Analyic Hierarchy Process (AHP) 147 In Flee Planning Decision-Making Modeling Framework Numerical Example An Applicaion in Solving Flee Planning Problem Flee Planning Decision Model 166 ix

11 4.3.6 Daa Descripion Resuls and Discussion Benchmark Problem versus Scenario P Benchmark Problem versus Scenario Q Summary OPTIMAL FLEET PLANNING WITH SLOT PURCHASE Slo Purchase and Flee Planning Decision-Making Sage 1: Slo Purchase Decision Model (SPDM) Consrains Problem Formulaion Sage 2: Flee Planning Decision Model (FPDM) Consrains Objecive Funcion Problem Formulaion Soluion Mehod Sage 1: Slo Purchase Decision Model (SPDM) Sage 2: Flee Planning Decision Model (FPDM) An Illusraive Case Sudy Daa Descripion Resuls and Discussions Furher Applicaion: New Nework Expansion Resuls Verificaion Summary ENVIRONMENTAL PERFORMANCE ASSESSMENT 214 FOR FLEET PLANNING 6.1 The Role of Environmenal Performance Assessmen Quanify Green Index: Gini Coefficien Green Emission Index Green Noise Index Green Fuel Efficiency Index Quanify Green Flee Index: Analyic Hierarchy Process An Illusraive Case Sudy Daa Descripion Resuls and Discussions Sraegy A: Increase Load Facor Sraegy B: Operae New Aircraf Sraegy C: Reduce Service Frequency Sraegy D: Reduce Fuel Consumpion Advanages of he Proposed Framework Summary GREEN FLEET PLANNING DECISION MODEL Bi-objecive Green Flee Planning Problem Formulaion Consrains 244 x

12 7.2.2 Objecive Funcion Green Flee Planning Decision Model Soluion Mehod An Illusraive Case Sudy Resuls and Discussions The Resuls of Benchmark Scenario Impac of Objecive Ranking Impac of Green Consideraion Impac of Increasing Load Facor Impac of Reducing Service Frequency Poenial Cos Savings for Greener Flee Summary CONCLUSIONS Summary Fuure Works Research Accomplishemen 289 REFERENCES 291 APPENDIX A 309 Convoluion Algorihm APPENDIX B 311 The Relevan Sources of he Mode Choice Modeling Variables APPENDIX C 312 Model Modificaion for New Nework Expansion xi

13 LIST OF TABLES Table 3.1 The Informaion of he Respecive Unexpeced Even 3.2 The Expeced Value of Fligh Fare and Fligh Cos per Passenger 3.3 Aircraf Resale Price, Depreciaion Value and Purchase Price ($ millions) 3.4 The Resuls of Benchmark Scenario (Aircraf Acquisiion Decision Model) Page The Oupu of Sochasic Demand Aircraf Resale Price, Depreciaion Value, Purchase Cos, Lease Cos and Residual Value ($ millions) 3.7 The Resuls of Benchmark Scenario (Aircraf Acquisiion and Leasing Decision Model) 3.8 The Summary of Flee Planning Decision (Aircraf Acquisiion and Leasing Decision Model) The Aribues of KL-Penang Trip (Local Trip) The Aribues of KL-Singapore Trip (Trans-border Trip) The Characerisics of Respondens The Descripion of he Ineresed Variables The Choice Probabiliy of Local and Trans-border Trips (%) 4.6 The Modeling Resuls of KL-Penang Trip (Local Trip) 4.7 The Modeling Resuls of KL-Singapore Trip (Trans-border Trip) 4.8 The Sensiiviy Analysis of KL-Singapore Trip (Trans-border Leisure Trip) xii

14 4.9 The Sensiiviy Analysis of KL-Penang Trip (Local Business Trip) The Evaluaion of Relaive Comparison The Modeling Resuls of Travel Survey The Evaluaion of he Raio of Key Aspec (for Travelers Response) The Evaluaion of Key Aspec in Flee Planning Furher Analysis in Solving Flee Planning Problem The Resuls of Flee Planning Model The Resuls of Flee Size Aircraf Specificaions The Travel Demand of Airline The Operaional Daa of Inernaional Roues The Expeced Value of Fligh Fare and Fligh Cos per Fligh 5.5 The Purchase Cos, Lease Cos and Depreciaion Cos of Aircraf ($ million) 5.6 The Resale Price and Residual Value of Aircraf ($ million) 5.7 The Sandard Operaions Hour of Aircraf a Airpor 5.8 The Esimaed Demand Level and Average Fare for New Nework Expansion The Compuaional Resuls of Respecive Scenario The Summary of Flee Planning Decision The Summary of Service Frequency of Airline The Operaing Informaion of Inernaional Roues The Specificaion of Aircraf 233 xiii

15 6.3 The Emission Rae, Noise Level and Fuel Consumpion of Aircraf The Sraegy for Improvemen Acions The Resuls of Sraegy A-D The Summary of he Possible Soluion Mehods The Travel Demand and Service Frequency of Airline 7.3 The Expeced Value of Fligh Fare and Fligh Cos per Passenger Addiional Scenario for Furher Analysis The Green Performance of Airline (Benchmark Scenario) 7.6 The Flee Planning Decision of Airline (Benchmark Scenario) The Green Flee Index (GFI) for All Scenarios The Flee Planning Decision (In Average) for Various Scenarios The Environmenal Cos Flee Planning Decision Model 286 xiv

16 LIST OF FIGURES Figure 1.1 The Overall Framework of Flee Planning Page Modeling Framework of Sochasic Demand The Resuls of Scenarios A and B The Resuls of Scenarios C and D The Resuls of Scenarios E and F The Locaion of Klang Valley, Penang and Singapore 4.2 The Modeling Framework o Quanify he Probabiliy of Key Aspec The Evaluaion of he Raio of Key Aspec The Graphical Resuls of he Probabiliy of Key Aspecs Two-sage Flee Planning Decision Model The Graphical Resuls of Two-Sage Flee Planning Decision Model 6.1 The Modeling Framework o Quanify Green Flee Index The Graphical Resuls of Green Index The Flow Char of he Opimizaion Approach 257 xv

17 LIST OF ABBREVIATIONS c Airfare of business class biz, Fi c Discouned fare of economy class dec, Fi c Full fare of economy class fec, Fi close End of working hours a airpor S i f D, A Funcion of he number of flighs in erms of D and A f i, m S n D A Service frequency of operaing roue n, F A f M Aircraf noise level a approach sage L f M Aircraf noise level a laeral sage f M, E Aircraf noise level a flyover sage F S n gf D, A Funcion of he raveled mileage in erms S n f D, A S i of he number of flighs, S n hg D, A Funcion of he mainenance cos in erms of he raveled mileage, g m n open Aircraf saus (1:new aircraf, 2:aging aircraf) Aircraf ype Sar of working hours a airpor P p s Probabiliy o have I (corresponds o phenomenon S) and L I r v biz, F i Discoun rae for which he discoun facor is 1 r Operaing period Operaing cos of business class v, Discouned cos of economy class dec F i xvi

18 v fec, F i Full cos of economy class w y y1 Environmenal facor Aircraf age Local leisure rip y 2 Local business rip y 3 Trans-border leisure rip y 4 Trans-border business rip Number of passenger in business class * p biz, F i * p dec, F Number of passenger ha pay discouned i fare for economy class * p fec, F Number of passenger ha pay full fare for i economy class biz, fec, dec Se of classificaion of passengers Classificaion of passengers (biz: business class, fec: economy class (full fare), dec: economy class (discouned fare)) Parameer of environmenal susainabiliy Significance level of demand consrain Significance level of lead ime consrain Significance level of selling ime consrain The larges eigenvalue max n A Toal operaed aircraf Af n, F i Addiional service frequency resuled from slo purchase decision AVT Aircraf availabiliy (number of days) nf, i xvii

19 Biz % Porion of passenger in business class BLK Block ime nf, i C fuel n Funcion of fuel expenses Ci Decisional crieria CF i Slo price CE CI CR Consulancy of expers Consisency index Consisency raio D Forecased demand wih mean, f f and sandard deviaion, f D Possible incremen of forecased demand f ( inc) D The demand level of he operaing period S D, Sochasic demand of operaing roue Fi (corresponds o phenomenon S) Dec % L L L L 1y 2y ny Porion of passenger in economy class (discouned fare) DEP dep, dep,, dep Depreciaion value of leased aircraf P P P P 1y 2y ny DEP dep, dep,, dep Depreciaion value of purchased aircraf DIS F i Disance of a paricular operaing roue 1 2 n DL dl, dl,, dl Payable deposi for aircraf leasing 1 2 n DLT DLT, DLT,..., DLT Desired lead ime of aircraf acquisiion DP Decision policy of airline 1 2 n DP dp, dp,, dp Payable deposi for aircraf acquisiion 1 2 n DST DST, DST,..., DST Desired selling ime of aging aircraf xviii

20 E S n Number of engines E cos Expeced value of fligh cos per passenger S E Expeced value of fligh fare per passenger fare n S E Expeced number of seas (capaciy) of sea n aircraf EC Environmenal cos GFI EFF n ER EX EXN F ex F i F nw Nework efficiency facor Emission rae of aircraf Toal aircraf emission Cumulaive noise level Exising operaing nework Operaing roue (fligh) New operaing nework F s k w The relevan componen of key aspec, s k FC, Fuel consumpion m n F i Fec % FEL Porion of passenger in economy class (full fare) Fuel efficiency index FVi Exising variey of flee composiion GFI GI E Green flee index Green emission index GI Green fuel efficiency index FE GI N Green noise index xix

21 1y 2y ny I I, I,..., I Iniial quaniy of purchased/leased aircraf L L L L 1y 2y ny I I, I,..., I Iniial quaniy of leased aircraf P P P P 1y 2y ny I I, I,..., I Iniial quaniy of purchased aircraf Index Sochasic demand index (SDI) LEASE lease, lease,, lease Lease cos of aircraf 1 2 n LF, Load facor n F i M Aircraf weigh MAX Allocaed budge for aircraf leasing and budge () acquisiion MXU Maximum uilizaion of aircraf (in erms nf, i of service frequency) NA Ned Ne1 Ne 2 NEW Funcion of he number of aircraf Operaing nework Shor-haul nework Medium/long-haul nework Proporion of new aircraf NP Number of passengers m nf, i 1 2 n O O, O,..., O Quaniy of aircraf o be ordered OLD Proporion of aging aircraf ORDER Quaniy of aircraf ha could be purchased in he marke P ij Choice probabiliy P hc P I P L Probabiliy of unexpeced even c happens wih a probable occurrence of h I Discouned profi funcion xx

22 PARK PP PP Area of parking space Pas performance of airline Produc of he probabiliy of unexpeced even for operaing period PURC purc, purc,, purc Purchase cos of aircraf 1 2 n R The revenue of operaing roue (fligh) F, i R r Random number 1 2 n R R, R,..., R Quaniy of aircraf o be released for sale RESALE,, resale1 y resale Resale price of aircraf ny RGn RI Aircraf range (maximum disance flown) Random consisency index 1 2 n RLT RLT, RLT,..., RLT Real lead ime of aircraf acquisiion 1 2 n RST RST, RST,..., RST Real selling ime of aging aircraf S s1, s2,, s Probable phenomena k SEAT Sea (capaciy) of aircraf nf, i SIZE size1, size2,, size Aircraf size n 1y 2y ny SOLD sold, sold,, sold Quaniy of aircraf sold T Planning horizon TC Toal cos of airline P L I I TEC TNC Toal emission cos Toal noise charges TR Toal revenue of airline P L I I xxi

23 TUN Turn round ime a airpor n, Fi, k U ij Uiliy funcion of an alernaive j for an individual i 1 2 n U u, u,, u Seup cos for aircraf acquisiion UEC Uni cos of emission nf, i UFS Uni fuel cos UNC Uni noise charges nf, i W F i Willingness o pay for slo purchase L L L L 1 2 n X x, x,, x Quaniy of aircraf o be leased P P P P 1 2 n X x, x,, x Quaniy of aircraf o be purchased xxii

24 CHAPTER 1 INTRODUCTION 1.1 Background Flee planning deermines he opimal quaniy of he respecive aircraf ype ha is needed by an airline o mainain a argeed level of service while maximizing is profi margin. In flee planning, here are wo major decisions o be made, i.e. o deermine he opimal quaniy and he ype of aircraf o be purchased and leased hroughou he long-erm planning horizon in order o mee sochasic demand profiably. A proper flee planning is imporan as i would affec he economic efficiency of airline and i has an influenial impac on cusomer saisfacion (Zak e al., 2008). An oversized flee implicaes an increased cos while an undersized flee implicaes an unsaisfied demand and consequenly resuling o a decrease in revenue and profi (Czyzak and Zak, 1995; Crainic and Lapore, 1997; Crainic, 2000). In order o mainain a good level of service for an airline, here is a need o balance he supply and demand when opimizing flee planning. By incorporaing he supply and demand in making opimal flee planning decision, airlines would obain umos profi while providing a desired service 1

25 level. Consequenly, an airline s susainabiliy in erms of economy, social and environmen could be assured effecively under sochasic demand. The demand is defined as he number of passengers asking for service while he supply refers o services (aircraf, service frequency, service slos, ec.) ha could be provided by airlines o fulfill he demand. As such, hese aspecs become he mos criical componens ha need o be considered in flee planning models. Travel demand forecasing is an imporan componen as i could influen he resuls' robusness. There are wo ypes of ravel demand, i.e. deerminisic and sochasic demand, ha are involved in he modeling. The deerminisic demand associaes iself wih he level of ravel demand ha could be deermined wih cerainy. I is inelasic and known as a priori. Conversely, sochasic demand, as a random variable, refers o demand flucuaion which is uncerain a varying degrees primarily due o he occurrence of unexpeced evens which could ake place unexpecedly. Insead of deerminisic demand, sochasic demand should be considered because airline's operaing environmen is sochasic in naure due o he presence of uncerainy (Barnhar e al., 2003). Pas sudies revealed ha by considering sochasic demand, he soluion obained is more robus and closer o realisic implemenaion (Lises and Dekker, 2005; Yan e al., 2008; Hsu e al., 2011a, 2011b). 2

26 According o he airlines (Malaysia Airlines, 2010a; AirAsia Berhad, 2010a), some possible unexpeced evens include fuel price volailiy, poliical insabiliy (e.g. erroris aacks), global economic downurns, naural disasers, and ohers. When hese evens occur, he demand level would decrease remendously. Neverheless, sochasic demand is always being negleced by pas sudies in solving he flee planning problem. In oher words, he incorporaion of sochasic demand in long-erm flee planning is under research. As such, exising approaches and models for airline flee planning which are formulaed by pas sudies migh no be funcional for real pracice. This has moivaed he developmen of a well-defined long erm flee planning decision model in order o assure ha airlines can achieve heir argeed profi a a susainable manner. In view of he fac ha air ravelers (passengers) are he main users of airline's services which consiues he main income o airlines, he needs and expecaion of passengers are imporan o airlines in order o gain a larger marke share under such a compeiive airline indusry. As such, how airlines make an opimal flee planning decision, i.e. a muliple crieria decisionmaking, for each operaing period hroughou he planning horizon is imporan no only o ensure profiable reurns bu also o mee he ravel demand a a desired service level. Therefore, flee planning decision-making which is governed by muliple crieria (wih numerous key aspecs) should be handled wih care. 3

27 Among he key aspecs ha received grea concern from airlines are he operaional and economy aspecs (AirAsia Berhad, 2010a; Malaysia Airlines, 2010a). These aspecs are crucial for airlines no only o susain profiably bu also o assure he feasibiliy of aircraf operaions in supporing he operaing neworks. If he relevan key aspecs are no aken ino consideraion properly in flee planning, he resulan decision-making may no be viable o suppor he operaing sysem. Undeniably, his would consequenly resul in a subsanial loss o airlines no only in erms of moneary aspec bu also he ineres or loyaly of air ravelers. In he pas, here are some sudies ha adoped various approaches o solve flee planning problems. However, hey did no show how opimal flee planning decision is made wih regard o he influenial key aspecs of flee planning decision-making ha may vary differenly among airlines. While providing an adequae flee supply, i is imporan o capure he mode choice analysis (raveler s response) in view of heir needs and expecaion which would affec airline s service and profi margin o a grea exen. Furhermore, raveler s behavior changes wih he exensive growh of mulimode ransporaion neworks. Therefore i is necessary o frame his scenario in a beer manner. In he pas, some sudies, including Mason (2000, 2001), Evangelho e al. (2005), O Connell and Williams (2005, 2006), Pels e al. (2009) and Abda e al. (2011), had been conduced and conribued on he mode choice analysis of ravelers. However, here is no sudy ha incorporaes mode choice modeling in making opimal flee planning decision. 4

28 To mee passengers demand desirably, airlines need o provide an adequae number of service frequency o suppor heir operaing sysem profiably. As such, how airlines deermine a desired service frequency for each operaing roue is imporan as service frequency deerminaion is grealy affeced by ravel demand (Wei and Hansen, 2005; Pifield e al., 2009) and aircraf choice (Zou and Hansen, 2014). Furhermore, he demand flucuaion could affec airline's service frequency o a grea exen. Wen (2013) highlighed ha service frequency deerminaion ha is closely associaed wih aircraf ype is crucial for airlines o assure operaing effeciveness. Wihou his elemen, he resulan flee planning decision may no be appropriae o suppor curren operaing neworks under sochasic demand. However, i is imporan o noe ha he service frequency of airlines is sricly consrained by regulaory limis, especially he arrival/deparure resricion slos a paricular airpors. Therefore, he sraegy of airlines o provide a higher service frequency (o mee increasing demand) may no be workable, unless prior approval (e.g. via slo purchase) is obained. In he case ha an increasing service frequency is no feasible for airlines o mee he demand incremen, airlines may need o selec specific aircraf ype, especially larger aircraf, o accommodae he demand incremen. Ye, he selecion of aircraf by airlines is highly dependen on aircraf specificaion (ype) which is closely associaed wih is corresponding service frequency. As such, service frequency needs o be included in flee planning (Wei and Hansen, 2005; Pifield e al., 2009). Pracically, addiional service frequency could be obained 5

29 by incorporaing slo purchase in making opimal flee planning decision. The lack of desired slos for addiional service frequency may lead o a loss in revenue due o he inabiliy of airlines o mee passenger s demand. While meeing sochasic demand a a profiable level, environmenal issues could no be negleced in view of an increasing concern of green issue nowadays. According o recen saisics, ranspor secor has emerged as one of he major sources of carbon dioxide emission in he world (Janic, 1999; Chapman, 2007; Dekker e al., 2012) which conribues abou 14% of oal global emission recorded (Sern, 2006; European Environmen Agency, 2011). This has caused noorious environmenal problem such as acid rain, global warming and ozone layer depleion (Buon, 1993; European Commission, 1996). Air ranspor secor is claimed o be he mos unsusainable ranspor mode (Chapman, 2007) and here are hree criical environmenal facors, i.e. aircraf emission, noise and fuel efficiency (Janic, 1999; IPCC, 1999; ICAO, 2010; Sgouridis e al., 2011) for airlines. Lee e al. (2009) repored ha he ne effec of nirogen oxides emission from aircraf is esimaed o be 24%. In addiion, he effec of conrails is approximaed o be 21% and he combined effec of waer vapour, sulfur oxides and soo is abou 2.1% of he oal effecs. The carbon dioxide emission is abou 2.5%-3% (Scheelhaase and Grimme, 2007; Anger, 2010). Approximaely, he burning of 1kg of fuel by he aircraf engine would produce abou kg of nirogen oxides, 3.16 kg of carbon dioxide and 1.25 kg of waer vapour (Ralph and Newon, 1996). As such, wih he forecased annual air raffic growh a 5% (Airbus, 2007; Inernaional Air 6

30 Transpor Associaion, 2009; Boeing, 2009), he polluion level will escalae o an alarming level if i is lef unreaed. The aircraf noise is anoher source of aviaion polluion o he environmen and sociey, paricularly o hose who are living in he airpor viciniy. Janic (1999) revealed ha here are wo sources of noise from he aircraf engine, i.e. machinery and primary je noise. Machinery noise is produced by he engine's componens such as fan, compressor and urbine while he generaion of primary je noise is formed when he high-speed gases exhaus from he engine mix wih he surrounding air. Specifically, he main source of noise during ake-off sage is primary je noise while he machinery noise emerges as he major source during landing phase (Ashford and Wrigh, 1979; Horonjeff and McKelvey, 1983). Noise annoyance generaed from aircraf operaions could affec sociology (human) healh from numerous aspecs, including hyperension (Meiser and Donaelle, 2000), high blood pressure (Black e al., 2007) and cardiovascular diseases (Franssen e al., 2004). Besides, aircraf noise especially from nigh flighs has also affeced he qualiy of life of he residens living in he airpor viciniy (Hume e al., 2003; Kroesen e al., 2010). Fuel consumpion is also one of he environmenal issues faced by airlines. I is known ha aircraf emissions are direcly relaed o fuel burn. A more efficien aircraf engine no only save cos, bu also reduce carbon dioxide 7

31 emissions. Each kilogram of fuel saved reduces carbon dioxide emission by 3.16 kg. As such, one of he key areas for airlines o minimize environmenal (green) impac is o operae fuel-efficien aircraf (Inernaional Air Transpor Associaion, 2013). Desirably, airlines could make opimal flee planning decision by acquiring/leasing aircraf ype which could reduce environmenal impacs. In oher words, green aircraf is preferable. Comparaively, a newer aircraf wih advanced echnology is preferred in reducing aircraf emission. For insance, jumbo aircraf A380 is preferred as i is fuel-efficien and emis lesser emission and noise per sea (Airbus, 2013). However, flee planning decision-making does no depend on he environmen issue as he sole facor. Airlines need o consider he operaional issues and more imporanly profi earning. As such, acquiring/leasing new and large aircraf may no always be he preferred choice. In recen years, numerous local governmens and airpor auhoriies, e.g. in Ausralia, Sweden, Swizerland and Germany (Lu, 2009), have implemened sricer environmenal policy and regulaion in order o direc airlines o be greener. Environmenal fines, including emission and noise penaly, are imposed on airlines ha produce excessive polluans. For insance, Briish Airways has paid almos 20,500 per annum as emission charge o Frankfur Airpor (Scheelhaase, 2010) while in he Unied Saes, he penalies 8

32 of aircraf noise violaions a John Wayne Airpor may involve siffer fines as high as $500,000 and he disqualificaion of airline (Girvin, 2009). Undeniably, such policies would affec airline s profi margin. As a resul, i is necessary for airlines o consider he environmenal issue in flee planning. I is foreseen ha, by having green flee in place, a win-win siuaion beween airlines and he environmen could be achieved. In brief, here is a need o develop more effecive flee planning mechanisms and managemen sraegies in order o mee sochasic demand desirably. How o manage flee planning profiably under uncerainy is no a simple ask. There may be more issues and concerns besides hose ha have been highlighed above. Essenially, his research is concenraed on how o opimize flee planning and managemen sraegies of airlines o secure a higher efficiency and profi under various siuaions and pracical consrains as well as subjec o unpredicable uncerainy. Overall, i is anicipaed ha he findings of his research could provide useful guidelines o airlines o operae in a beer and susainable manner which will benefi air ravelers in reurn. 9

33 1.2 Research Objecives This research sudy has five objecives as lised below: 1. To opimize airline s flee planning decision by deermining he opimal quaniy and aircraf ype ha generaes maximum profi (subjec o pracical consrains). 2. To propose a modeling framework for sochasic demand. 3. To model and analyze he impacs of mode choice modeling in flee planning. 4. To compare and assess he impacs of he subjecive judgmen of airline s managemen in making flee planning decision. 5. To promoe green flee planning. 1.3 Research Scope This research comprises four major scopes, namely flee planning decision model under sochasic demand (o capure he occurrence of unexpeced evens), sraegic flee planning modeling framework (o deal wih 10

34 supply-demand ineracion), wo-sage flee planning (o assure adequae service frequency by incorporaing slo purchase) and green flee planning (o incorporae environmenal concerns). Basically, here are wo major elemens, i.e. demand and supply aspecs, ha affec airlines in making opimal flee planning decision o acquire/lease aircraf o mee ravel demand profiably. Specifically, he occurrence of unexpeced evens (e.g. naural disaser, oubreaks of flu disease, fuel price volailiy, ec.) would consiue sochasic demand which behaves uncerain in naure. This would affec he operaions and profi of airlines o a grea exen. As such, how airlines provide a desired service level, wih adequae flee supply o mee sochasic demand is exremely imporan. Mahemaically, an opimal flee planning model (aircraf acquisiion and leasing decision model) is developed wih he aim o find opimal profis while meeing uncerain demand a a desired service level (subjec o various pracical consrains). The decision variables of flee planning decision model are opimal quaniy and aircraf ype ha need o be purchased and/or leased o mee sochasic demand profiably. In order o mee sochasic demand realisically wih sufficien aircraf supply, here are various key aspecs (probable phenomena) ha need o be quanified and incorporaed in opimizing flee planning model. Remarkably, operaional, economy and environmenal aspecs were found o be he hree 11

35 probable phenomena (key aspecs) in making opimal flee planning decision for which he probabiliy of probable phenomena (wih regards o respecive various key aspec) indicaes he likelihood of aircraf possession o mee sochasic demand. In oher words, probable phenomena would assure an adequae flee supply o mee demand flucuaion a a desired level of service. To capure supply-demand ineracion in a beer manner, raveler s response (in erms of mode choice analysis) and he subjecive judgmen of decision makers (airline s managemen) owards numerous decisional crieria of flee planning (including airline s decision policy, exper s consulancy as well as airline s pas performance) are necessarily incorporaed in solving flee planning problem. These elemens are imporan o achieve a argeed level of service profiably from various key aspecs (i.e. operaional, economy and environmenal aspecs). To mee sochasic demand desirably, i is also vial for airlines o assure ha here is a desired service frequency which associaes closely wih he respecive aircraf ype in supporing curren operaing neworks. I is of umos imporance for airlines o assure a higher operaing efficiency and profi margin. However, he service frequency of airlines is sricly conrolled by airpors operaors in compliance o sandard regulaions of airpor in erms of aircraf operaions, especially during peak period or nigh ime. In such a case, how airlines monior and manage heir service frequency o mee demand flucuaion (especially demand incremen) necessiaes a proper flee planning. Specifically, slo purchase plays an imporan role o provide addiional service 12

36 frequency o airlines o mee demand incremen. Wihou his elemen, sochasic demand may no be me desirably and his would affec raveler s expecaion and subsequenly resuls in a loss of airlines no only in erms of operaing revenue/profi bu also he loyaly of ravelers. Environmenal susainabiliy is anoher crucial componen in flee planning. In view of he increasing concerns o preserve he environmen, green performance of airlines (in erms of aircraf emission, noise and fuel efficiency) needs o be moniored closely. Only by having a green flee in place, a win-win siuaion beween he airline and he environmen could be achieved. As such, his requires a well-defined flee planning model o deermine opimal quaniy of respecive aircraf ype in order o yield a greener performance while meeing sochasic demand saisfacorily a a profiable service level. Ideally, he flee supply (for boh aircraf composiion and he corresponding service frequency) of airlines should be in place, righ on ime, o suppor he curren operaing neworks profiably. Besides, he developed environmenal (green) assessmen performance model is able o provide insighful direcion and suggesion o airlines o achieve greener performance, by assessing he effeciveness of respecive miigaion sraegy. The developed approach could capure hree major environmen facors, namely aircraf emission, noise and fuel efficiency. I is also capable o capure he occurrence of unexpeced evens ha could affec airlines' operaions o a grea exen. As such, he developed mehodology is useful no only in flee planning, bu also pracically beneficial for aircraf operaions in real pracice. 13

37 This research disinguishes from he pas sudies as i shows how he environmenal facor (including aircraf emission, noise and fuel efficiency) could be incorporaed ino flee planning model (wih numerous pracical consrain) under uncerainy. In addiion, i shows ha airlines could susain a significan amoun of cos savings if green flee planning is carried ou wih some beneficial improvemen sraegy (o yield a greener performance). For all he above-menioned research scope, he compuaional resuls were verified by making empirical comparisons wih he acual operaing performance of airlines. Overall, he findings of illusraive case sudies show ha his research is pracically viable for which he overall framework o produce opimal flee planning decision-making, as an effecive managemen sraegy for airlines, is displayed in Figure 1.1. Figure 1.1: The Overall Framework of Flee Planning 14

38 1.4 Thesis Overview This research is organized as follows: Chaper 1: Inroducion presens he relevan problem saemen, significance and moivaion of carrying ou his research, ogeher wih he research objecives and scopes. Besides, his hesis overview liss ou sysemaically all he opics ha are included in his research. Chaper 2: Lieraure review discusses pas sudies which are closely relaed o his research. Basically, here are five major discussion areas, namely ravel demand forecasing: deerminisic vs sochasic, airline flee planning approach, sraegic flee planning modeling framework, slo purchase and service frequency in flee planning, as well as green flee planning. A horough and updaed review, including he srenghs and shorcomings of pas sudies, had been addressed accordingly. In Chaper 3: Flee planning decision model under sochasic demand, he firs par of he discussion focuses on a novel modeling framework of sochasic demand in order o deermine he level of sochasic demand realisically under uncerainy. To solve flee planning model, aircraf acquisiion decision model (wihou aircraf leasing) is hen developed and solved opimally wih a realisic case sudy (as linear programming model) 15

39 under sochasic demand. Subsequenly, aircraf acquisiion and leasing decision model is presened o obain opimal flee planning decision hroughou he long-erm planning horizon. An illusraive case sudy in he form of nonlinear programming model is presened o examine he feasibiliy of he developed approach o acquire and/or lease aircraf a opimal profi. Chaper 4: Sraegic flee planning modeling framework mainly covers wo pars, namely mode choice analysis and Analyic Hierarchy Process (AHP) modeling framework. Mode choice analysis focuses on he modeling of raveler s response owards airline s services and marke share under mulimode ransporaion neworks. The analysis of mode choice modeling is hen incorporaed in AHP modeling framework o work ou a sraegic flee planning by assuring an adequae flee supply o mee sochasic demand saisfacorily. To do his, he subjecive judgmen of decision makers (airline's managemen) is incorporaed necessarily. In Chaper 5: Opimal flee planning wih slo purchase, slo purchase decision model is firs discussed (in sage 1), followed by flee planning decision model (in sage 2). In his chaper, influenial impacs of slo purchase in providing addiional service frequency o mee increasing demand are invesigaed explicily so ha airlines could make a proper decision-making (via slo purchase) o obain opimal soluions for flee planning. The relaions of slo purchase, service frequency, flee supply and airline's profi level are 16

40 discussed explicily. Chaper 6: Environmenal performance assessmen for flee planning quanifies he green performance of airlines mahemaically from hree major perspecives, namely Green Emission Index, Green Noise Index and Green Fuel Efficiency Index. The overall green performance of airlines is hen compiled in erms of Green Flee Index (GFI). Besides, some improvemen sraegies (i.e. increasing load facor, operaing new aircraf, reducing service frequency and reducing fuel consumpion) are suggesed o achieve greener performance. Chaper 7: Green flee planning decision model primarily focuses on he problem formulaion and soluion mehods o assis airlines o obain opimal profi while achieving greener performance. Mahemaically, i is formulaed in he form of bi-objecive opimizaion model. By examining a realisic case sudy, effecive improvemen sraegy o yield a greener performance are discussed explicily. Besides, poenial environmenal cos savings by having a green flee is revealed. Chaper 8: Conclusions presens a comprehensive summary of his research. Some possible direcions for fuure research and research accomplishmen are also included. 17

41 CHAPTER 2 LITERATURE REVIEW This chaper discusses pas sudies which are closely relaed o he flee planning problem of airlines. Basically, pas sudies can be caegorized ino five major discussion conexs, namely (i) ravel demand forecasing: deerminisic vs sochasic, (ii) airline flee planning approach, (iii) sraegic flee planning modeling framework, (iv) service frequency and slo purchase in flee planning, and (v) green flee planning. In each conex, a horough and updaed review, which includes he srenghs and shorcomings of pas sudies, has been discussed explicily in order o provide some insighful overviews on he relevan evoluion of flee planning in he airline indusry. 2.1 Travel Demand Forecasing: Deerminisic vs Sochasic In mos of he research sudies peraining o air ranspor, deerminisic demand is forecased and used in he modeling and planning. New (1975) forecased he ravel demand based on he ypes of flighs (shor, medium and long-haul) and number of flighs operaed by airline. Teodorovic and Krcmar- Nozic (1989) esimaed he oal expeced number of passengers based on he marke share of airline which is assumed o follow normal disribuion. Hsu 18

42 and Wen (2003) forecased he demand level of individual operaing roue by adoping grey heory, which is a ime series forecasing approach ha solely requires a small amoun of daa for forecasing. However, is capabiliy is limied o he ime series ha exhibis exponenial growh. Furhermore, i necessiaes regular and new daa o enhance forecasing accuracy. To solve he flee assignmen problem, Barnhar e al. (2002) forecased he level of demand based on he average demand daa and also on he respecive scheduled iinerary as requesed by he passengers. A fundamenal assumpion in deerminisic demand forecasing is ha he demand of passengers is inelasic. Wih he lack of abiliy o handle sochasic feaures, i could no capure demand flucuaions, i.e. he resulan forecasing of deerminisic demand is no responsive o he changes in demand. Thus, deerminisic demand forecasing is no sufficienly robus o reflec he sochasic naure of flee planning problem and hence i may no be a good approximaion for he acual pracice (Lises and Dekker, 2005; Tan e al., 2007; Yang, 2010). This may resul in he loss of opimaliy for a deerminisic modeling in view of he fac ha he impacs of demand variabiliy in acual operaions is negleced (Yan e al., 2008). In comparison, sochasic demand forecasing provides more pracical resuls, wih realiable consisency (Yang, 2010). I is more effecive and useful han deerminisic modeling for which he deailed and realisic daa on demand paerns are no available (Diana e al., 2006; Yan e al., 2008). Noably, Lises and Dekker (2005) highlighed ha airlines would secure a higher profi margin, approximaely o be 11-15% 19

43 more, by capuring sochasic demand. Correspondingly, he load facor would increase abou 2.6% while poenial spill and urned-away passengers would decrease abou 3.3% and 2.3%, respecively. This shows ha sochasic demand modeling is much more beneficial o airlines (compared o deerminisic demand modeling). The ravel demand of air ranspor is sochasic in naure, primarily due o he occurrence of unexpeced even (e.g. economic recession, naural disaser, biological disaser, poliical sabiliy, ec.) which is unpredicable in he real pracice (Malaysia Airlines, 2010a; AirAsia Berhad, 2010a). When hese evens ake place, he level of demand would be affeced o a grea exen and hence resuls in demand flucuaion which behaves in a sae of uncerainy (sochasic). In view of his and recognizing he limiaions of deerminisic demand modeling, researchers had sared o adop sochasic demand in modeling. Lis e al. (2003) used a parial momen measure of risk o inspec he uncerainy of ravel demand. Lises and Dekker (2005) adoped scenario aggregaion-based approach o deermine he bes choice of aircraf by assuming ha ravel demand follows normal disribuion. Yan e al. (2008) capured he demand flucuaions by developing passenger-flow neworks and passenger choice model for which passenger uiliy and marke demand funcions are formed in order o deermine he choice probabiliy funcion of ravelers. Pifield e al. (2009) employed a simulaneous-equaions approach o analyze demand elasiciy and aircraf choice. Hsu e al. (2011a) adoped grey opological models wih Markov-chain o capure demand flucuaions while 20

44 Hsu e al. (2011b) combined grey opological forecass wih Markov-chain model o inspec demand flucuaions and also o deermine he probabiliy of demand. They imposed a penaly cos funcion if he acual demand is more han he forecased demand. In oher areas (no air ransporaion), sochasic demand is assumed o follow cerain disribuion. For example, Berman e al. (1985) and Baa e al. (1989) adoped Poisson disribuion o model sochasic demand for queuing sysems. Du and Hall (1997) proposed a dynamic model o capure he sochasic demand for por operaion. Bojovic (2002) modeled he demand of railroad nework as a Gaussian probabiliy densiy funcion while Tan e al. (2007) assumed ha sochasic demand has a normal disribuion in solving vehicle rouing problem. The proposed mehods used in he pas sudies o capure sochasic demand are remarkable, bu hey have some limiaions. One major shorcoming is ha hey did no quanify he occurrence of unexpeced evens in heir aemps o model sochasic demand. For example, Lis e al. (2003) modeled he demand enirely based on a one-sided risk measure (insead on demand variaion) for which he likelihood of objecive funcion in meeing ravel demand is conrolled no o exceed a hreshold value. Hsu e al. (2011b) adoped Markov-chain model by aking ino accoun only one se of ransiion probabiliy o model ravel demand. Boh sudies ignored he possibiliy of 21

45 evens ha could ake place unexpecedly. Insead of demand flucuaions modeling, he probabiliy of occurrence of unexpeced even should be quanified sysemaically as i could affec sochasic demand o vary differenly. Wihou his elemen, he level of sochasic demand may no be modeled as close o realiy as i is. Moreover, he assumpion of fixed ype of disribuion o quanify demand flucuaion migh be oo resricive. The proposed mehodology migh no be applicable if real demand paern does no follow he ype of disribuion as assumed. Furhermore, demand forecasing mehods as proposed in he exising sudies are only applicable for shor-erm period. For example, Tan e al. (2007) and Yan e al. (2008) modeled he demand flucuaion wihin a day. Lises and Dekker (2005) and Pifield e al. (2009) modeled weekly and monhly demand, respecively. Such shor-erm forecasing mehods are no applicable o model long-erm demand flucuaion which is required in solving flee planning problem. 2.2 Airline Flee Planning Approach To formulae and opimize he flee planning problem, pas sudies had adoped various approaches. Wei and Hansen (2005) buil a nesed logi model o inspec he influence of aircraf size, fligh frequency, sea availabiliy and airfare on airline s demand. They highlighed ha airlines can obain higher reurns from increasing he service frequency han from increasing he aircraf size, i.e. airline s marke share is super-proporional o airline frequency share. 22

46 Therefore, here is a endency for airlines o use smaller aircraf since an increase of frequency can arac more passengers. Despie a closed relaion beween aircraf size and service frequency in making opimal flee planning decision, here is no proper mechanism or clear indicaion on how airlines could acquire and/or lease specific aircraf ype (wih corresponding service frequency) o service he esimaed marke share. Furhermore, he occurrence of unexpeced even is no aken ino consideraion. Wei and Hansen (2007) developed game-heoreic models o invesigae airlines decisions on aircraf size and service frequency under compeiive environmen. They also examined he operaing cos and he demand level of compeing airlines. They revealed ha aircraf size, depending on marke ypes, is a significan facor for flee planning decision-making. The resuls show ha airlines end o use he smalles, ye cos-efficien, aircraf o accommodae differen demand levels, and only increase he service frequency o mee he increasing demand. I was highlighed ha airlines wih more small aircraf can manage flexibly aircraf operaions (including scheduling and roue planning) which are closely relaed wih opimal aircraf acquisiion/leasing decision. However, i is assumed ha compeing airlines know each oher's payoffs, available sraegies and oher releavan informaion in selecing opimal aircraf size and service frequency. As such, he reliabiliy and applicabiliy of heir model migh be quesionable a cerain exen. 23

47 Wei (2006) employed game-heoreical model o invesigae how airpor landing fees affec wo compeing airlines o make decision on aircraf size and service frequency (a opimal profi) in duopoly markes. The resuls show ha a higher landing fee will force airlines o operae larger aircraf and fewer frequency (o reain he same number of passengers). This shows ha airline s opimal aircraf size and service frequency are affeced significanly by landing fees. However, his model assumes ha boh airlines know all he available choices and resulan profi for each oher. This migh no be realisic. Besides, he opimal decision is made solely based on he landing fees. This migh be oo resricive in view of some oher imporan elemens, e.g. demand uncerainy and budge consrain are negleced in making opimal decision. Kozanidis (2009) developed a muli-objecive opimizaion model o maximize aircraf availabiliy. He showed ha fligh and mainenance requiremens are wo imporan facors in flee planning. Besides maximizing he flee availabiliy level, i was found ha i is also vial o minimize is variabiliy in order o assure ha he availabiliy level remains relaively consan over ime. However, his model is limied o miliary operaions insead of commercial fligh applicaion. Givoni and Rieveld (2010) analyzed he environmenal impacs of airlines choice on aircraf size and service frequency. The resuls show ha environmenal impacs could be reduced by operaing a lower service 24

48 frequency (wih larger aircraf). Besides, i was found ha increasing he supply hrough larger aircraf raher han addiional services (wih more frequency) exhibis a beer use of exising capaciy. As such, he resuls highligh ha a large aircraf (wide body) designed for shor-haul fligh would be needed no only o make use he available runway capaciy in a beer manner bu also o reduce he environmenal impac from aircraf operaions. This shows ha he respecive aircraf ype (wih differen size and corresponding service frequency) is an imporan elemen for airlines o make flee planning decision profiably and environmenally. However, hey did no consider possible roue disance and aircraf weigh in heir analysis despie he fac ha aircraf specificaion (including aircraf range and engine weigh) would affec aircraf performance. Hsu e al. (2011a) formulaed sochasic dynamic programming model o opimize airline decisions in purchasing, leasing and disposing of he aircraf. The resuls show ha airlines end o lease raher han purchase aircraf o mee demand flucuaion. Besides, airline ends o form is aircraf composiion by operaing a single ype of aircraf for each operaing period. By considering sraegic alliance beween airlines, Hsu e al. (2011b) developed a dynamic programming model ha deals wih aircraf purchase, dry/we leasing and disposal. The findings reveal ha airline can achieve more cos savings hrough ineracive bargaining (for aircraf acquisiion/leasing) raher han leasing from non-allied airlines. These sudies are ineresing bu posed some limiaions. The mehods proposed by Hsu e al. (2011a, 2011b) are used o 25

49 ackle he flee planning problem wih sochasic demand bu hey did no capure he occurrence of unexpeced evens in modeling sochasic demand. In addiion, heir formulaion migh be oo simplisic by considering demand as he sole consrain. In fac, here are oher crucial consrains, such as budge consrain, lead ime and selling ime consrain, which are imporan in flee planning. 2.3 Sraegic Flee Planning Modeling Framework An efficien flee planning under sochasic demand over a long-erm planning horizon sill remains a major concern for many airlines. This happens mainly due o he supply-demand ineracion ha needs o be handled wih grea care, no only because of sochasic demand ha flucuaes grealy from ime o ime bu also owing o various key aspecs (muli-crieria) of flee planning decision-making ha correlae closely o demand flucuaion. For airlines, he operaions and economy emerge o be he key aspecs when making opimal flee planning decision (AirAsia Berhad, 2010a; Malaysia Airlines, 2010a). Undeniably, hese aspecs are grealy affeced by sochasic demand, i.e. he main facor of airline s services and income. As such, he supply-demand ineracion should be capured explicily in solving he flee planning problem sraegically. This cerainly necessiaes a well-developed and sraegic flee planning model. 26

50 There are some pas sudies ha adoped various approaches o solve he flee planning problems (as discussed in secion 2.2). However, hey did no show how opimal flee planning decision is made wih regard o he influenial key aspecs of flee planning decision-making ha may vary differenly among airlines. As discussed in secion 2.2, he exising sudies primarily focus on he echnical aspec in solving flee planning problem, i.e. hey mainly analyze how airlines make flee planning decision o obain opimal aircraf composiion o mee ravel demand, bu hey did no quanify he key aspecs of flee planning decision-making for which he exen of he respecive key aspec affecing flee planning decision-making is no measurable. Furhermore, he supply-demand ineracion is no sudied explicily by exising sudies. Wihou his elemen, he resulan flee planning decision may no be sraegic o suppor airline s operaing neworks. As such, wo major componens, i.e. demand managemen (in erms of mode choice analysis) and significan key aspecs of flee planning decisionmaking are excepionally crucial for airlines in solving flee planning problems sraegically. In erms of demand managemen, raveler s response is imporan o be undersood and capured explicily by airlines in order o gain a larger marke share (for more profi). Thus, mode choice analysis is required o examine he needs and percepion of ravelers owards airline s services, especially under such a compeiive mulimodal ransporaion sysem (Yan e al., 2008). By doing his, airlines would be able o undersand heir users in a beer manner and hence hey could mee passenger s expecaions desirably 27

51 wih a much beer service qualiy (including an adequae aircraf supply). In order o mee ravel demand a a desired service level, various key aspecs (e.g. operaional and economy) need o be quanified precisely o make opimal flee planning decision-making under uncerainy. To do his, Analyic Hierarchy Process (AHP) which is capable o deal wih uncerainy (Saay and Tran, 2007) plays he role o quaniy he probabiliy of respecive key aspec in solving flee planning problem. I is anicipaed ha by incorporaing mode choice modeling and he AHP in he flee planning model, he supply-demand ineracion could be capured explicily o yield a sraegic flee planning decision-making. The reviews on mode choice analysis and AHP are discussed in he following subsecions Mode Choice Analysis: Air and Ground Transpor The mode choice of ravelers would consiue he marke share of airlines and hence mode choice modeling needs o be analyzed properly. Furhermore, he mode choice of ravelers could be differen nowadays wih he developmen of mulimodal ransporaion neworks (Yan e al., 2008). Thus, mode choice analysis should be done regularly and up o dae in order o undersand he curren ravel rend and raveler s needs in a beer manner so ha an adequae aircraf supply could be provided, righ on ime, o mee raveler s expecaion. For such a compeiive mulimodal ransporaion sysem nowadays, he compeiion is inensifying no only among airlines bu 28

52 also beween airlines and ground ranspor. Globally, he compeiion beween low-cos carriers (LCCs) and full service carriers (FSCs) is escalaing mainly due o he evoluion and subsanial growh of LCCs. Pas sudies, including Mason (2000, 2001), Gillen and Morrison (2003), Barre (2004), Evangelho e al. (2005), O Connell and Williams (2005, 2006), Fageda and Fernandez-Villadangos (2009), Pels e al. (2009) and Abda e al. (2011) repored ha FSCs had los a significan proporion of ravelers o LCCs, and his subsequenly led o subsanial financial losses. Besides, he presence of LCCs had significan impacs in lowering he average fares of airline indusry. Therefore, he compeiion beween LCCs and FSCs has become one of he challenges for airlines in assuring a profiable marke share which is crucial for airlines. In such a case, how o susain and sand ou in such a compeiive airline indusry cerainly requires operaional and managerial efficiency. Recognizing he need o improve he services especially o gain a larger marke share, mode choice decision of ravelers, which is a key policy elemen in demand managemen, should no be negleced. Apar from inensifying compeiion beween he LCCs and FSCs, in fac, here s a direc compeiion beween he air ranspor and ground ranspor. To analyze he demand of ravelers, he compeiion beween highspeed rain (HST) and air ranspor were examined by Gonzalez-Savigna 29

53 (2004) and Roman e al. (2007) for Madrid-Barcelona roue, Givoni (2007) for London-Paris roue, Oruzar and Simonei (2008) for Saniago-Concepcion roue in Chile and Adler e al. (2010) for he European Union nework. Alhough hese sudies examined he compeiion of air ranspor wih ground ranspor a cerain exen, oher ypes of ground ranspor (e.g. bus, car) and specific ype of airlines (e.g. low-cos airlines) were no considered explicily in hese sudies. Furhermore, he sudy area was limied o European counries. Therefore, i could be seen ha exising sudies on he compeiion of air ranspor and ground ranspor are very limied. As repored in People s Daily Online (2011), upon he compleion of Kunming-Singapore High-Speed Railway in 2020, i will ake ravelers abou 10 hours o ravel beween Kunming, China and Singapore (i.e. passing by Bangkok, Thailand and Kuala Lumpur, Malaysia). The compleion of his ranspor nework would hen affec he choice of ravelers in using ground ranspor and air ranspor. The above-menioned insances confirmed he inensifying compeiion beween air ranspor and ground ranspor no only for he presen and bu also for he fuure. As such, i is excepionally essenial for airlines o undersand and o analyze he mode choice of ravelers in order o flourish in a compeiive ransporaion sysem. This aspec is cerainly necessary for airlines in implemening appropriae markeing sraegy o arac more ravelers as well as o increase heir mode share. In addiion, mode choice analysis is significan for airlines in managing heir ravel demand and also in predicing fuure ravel rend. From he social aspec, air ravelers would hen 30

54 benefi by geing a beer service enhanced by he airlines. For airline indusry, he adven of low cos carriers (LCCs) has reshaped he compeiive environmen and has made a significan impac on ravelers mode choice. The LCCs pursue simpliciy, efficiency, produciviy and high uilizaion of asses in order o offer low fares (O Connell and Williams, 2005). Besides, LCCs offer only a single class of service, high densiy seaing, no free food and drinks, no connecing services, and hey commonly use under-uilized secondary airpors (Pels e al., 2009). This reduces he operaional cos overall which could arac ravelers o use LCC s service a a lower price. Such revamp of he airline service has brough vicious compeiion o FSCs. I was found ha LCCs had aken up a large marke share of ravelers who are concerned wih ravel cos. Evangelho e al. (2005) found ha LCCs are preferred by smaller companies wih minimum expendiure policies. Mason (2000, 2001) added ha LCCs are preferred by shor-haul business ravelers, while O Connell and Williams (2005, 2006) found ha i has dominaed leisure rip marke. Apar from ravel cos, he flexibiliy of fligh schedule, convenience in icke booking (hrough inerne), aracive holiday package, and promoional parking a airpors (Barre, 2004; Evanlogelho e al., 2005) are among he facors advocaing he choice of LCC services. The sociodemographic characerisics of ravelers such as ehnics and level of educaion are found o be significan as well (Ong and Tan, 2010). 31

55 The compeiion of LCC is no limied o FSC, as i also has ground ranspor. Numerous sudies have shown ha here is a direc compeiion beween HST and air ranspor. Rus and Inglada (1997) showed ha he inroducion of HST had induced a fall in demand of 20%-50% of he air ranspor while Gonzalez-Savigna (2004) revealed ha over 50% of air ravelers (wih leisure purpose) would diver o HST. The journey ravel ime is one of he significan facors ha affec he mode choice beween HST and LCC. Gonzalez-Savigna (2004) commened ha HST migh be able o compee wih he LCC for he journey which is less han hree hours. Roman e al. (2007) found ha HST is more compeiive for shor journey as he ravelers choose HST wih he aim o reduce delay ime. In addiion, Adler e al. (2010) showed ha HST would arac almos 25% of medium-disance journey (up o 750km) bu only 9% for longer haul markes. Travelers socioeconomic background is found o be one of he significan facors. Oruzar and Simonei (2008) found ha older ravelers prefer o ravel wih HST. In he Malaysian conex, O Connell and Williams (2005) showed ha here is a mode shif from buses and rains when AirAsia was firs launched in They showed ha sudens, who accouned for he second larges non-business marke, have swiched o AirAsia insead of raveling wih buses and rains. Furhermore, a large proporion of AirAsia's ravelers are firs ime flyers and majoriies are youngsers. Neverheless, he sudy was carried ou many years ago and i did no invesigae he conribuing facors ha cause he mode shif. 32

56 To model he mode choice decision of ravelers explicily, saed preference (SP) survey has been used exensively in he pas o invesigae he choice of ravelers. Principally, he SP survey aims o invesigae raveler's response owards hypoheical scenarios in selecing he ravel mode (alernaive) ha is mos beneficial for raveling siuaion and purpose (Train, 2003). To conduc he SP survey, he design of quesionnaire could be oulined wih various raveling aribues (wih differen levels) which could be seleced accordingly based on he findings of he pas lieraures, pilo survey or ranspor operaor s operaional daa and records (Yang, 2005; Hess e al., 2007; Loo, 2008; Wen and Lai, 2010). However, under he circumsances for which he number of raveling aribues and level increase, i is no realisic o presen all possible combinaions of choice o respondens in he real pracice. In such a case, fracional facorial design and confounding facorial design (blocking approach) could be adoped o presen he quesionnaire reasonably o argeed respondens (Train, 2003; Mongomery, 2005). There are several models ha could be esed o model he mode choice decision of ravelers. Some possible models include logi, probi and generalized exreme value (GEV) models (Oruzar and Willumsen, 2001; Train, 2003; Mongomery, 2005). There are some sudies which were conduced and analyzed using SP survey. By underaking SP survey, Hess e al. (2007) modeled airpor and airline choice behavior in he form of mulinomial logi (MNL) srucures while Loo (2008) made use of SP survey o model passengers airpor choice, 33

57 specifically for Hong Kong for which he MNL model was found o be significan o model passengers choice. Besides, Wen and Lai (2010) discovered ha airlines choice of passengers, colleced from SP survey, fied well in he MNL model. To model he inerciy mode choice decision of passengers, Yang (2005) developed several models, including MNL, heerogenous logi kernel (HLK), mixed logi (ML), laen class (LCM), compeing desinaion (CD) and heerogenous compeing desinaions (HCD) models. In comparison o MNL model (as base model), he showed ha HCD model is significan o improve he model s explanaory power by considering muliple-heerogeneiy while he ML model, ha adops he specificaion of coninuous probabilisic disribuion, emerges o be he bes explanaory model in erms of he heerogeneiy of ase variaion. Apparenly, many mode choice models in he exising sudies were oulined based on he MNL model o analyze he behavior of ravelers. This happens mainly due o is inheren propery of Independence from Irrelevan Alernaives (IIA). However, MNL model is governed by Independen and Idenically Disribued (IID) error erm ha assumes homogeneiy in unobservable componens of uiliy. In oher words, he MNL model could no capure heerogeneiy properly (Train, 2003; Yang, 2005). 34

58 2.3.2 Analyic Hierarchy Process: A Tool o Quanify he Key Aspecs of Flee Planning Decision-Making Analyic hierarchy process (AHP) was firs inroduced by Saay (1977) wih he aemp o selec and prioriize a number of acions by evaluaing a group of predeermined crieria in making muli-crieria decision. Concepually, he AHP is originaed from he fuzzy se heory which is developed by Zadeh (1965). As a muliple crieria decision-making approach, he AHP allows he respecive judgmens o vary over he values of a fundamenal scale 1-9. In such a way, he AHP possesses he capabiliy o capure fuzziness (uncerainy) in making muli-crieria decision (Saay and Tran, 2007). As such, flee planning decision-making of airlines which is, in fac, uncerain (primarily due o sochasic demand) and grealy governed by various key aspecs (muli-crieria) could be solved sraegically wih he aid of he AHP. To esimae drivers' preferences owards available ransporaion alernaives, Arslan and Khisy (2006) adoped AHP o explain he roue choice behavior from a behavioral poin of view. Hsu e al. (2009) uilized AHP o examine he preferences of ouriss by idenifying he influenial facors in selecing heir desinaions. Besides, AHP is widely applied in oher secors, including resource managemen, corporae policy and sraegy (Velasquez and Heser, 2013). Apparenly, none of he lieraures apply AHP o solve flee 35

59 planning problem under uncerainy. Specifically, here is no sudy ha applies AHP o quanify he key aspecs of flee planning decision-making. Moreover, i could be seen ha exising sudies did no consider mode choice analysis when making flee planning decision (as discussed in secion 2.3.1). Mos of he sudies in mode choice analysis (e.g. Mason, 2000, 2001) only focused on he raveling aribues of airlines and raveler s preference for which he impacs of raveling aribues in affecing he supply (flee) planning are no inspeced. This (supply-demand) aspec should be considered owing o he fac ha he uilizaion of aircraf and airline's operaions correlae closely wih he rend of ravel marke especially under uncerainy. This necessiaes he incorporaion of mode choice modeling and he AHP in he flee planning model no only o yield a sraegic flee planning decision-making, bu also o capure supply-demand ineracion in a beer manner. 2.4 Service Frequency and Slo Purchase in Flee Planning This secion reviews he relevan papers which discussed and analyzed he service frequency of airlines. Specifically, pas sudies could be grouped ino wo caegories: (i) service frequency deerminaion in flee planning, and (ii) slo purchase. I is anicipaed ha airlines would gain more profi and mee more demand (wih a higher service frequency) by incorporaing slo purchase necessarily in solving flee planning problem. 36

60 2.4.1 Service Frequency Deerminaion in Flee Planning There are some sudies ha inspec airline's service frequency. Pifield e al. (2009) and Mikio (2011) discussed he rade-off beween aircraf size and service frequency. Their resuls showed ha airlines would end o increase fligh frequency (and hence decrease aircraf size) when demand increases (Pifield e al., 2009) and also when runway capaciy expands (Mikio, 2011). However, hey did no show how airlines could provide service frequency adequaely (wih corresponding aircraf ype) o mee increasing demand. Paricularly, from an environmenal aspec, Givoni and Rieveld (2010) discussed airline's choice on aircraf size and service frequency. A lower service frequency (by operaing larger aircraf) was found o produce lesser amoun of emission and noise. This signifies ha aircraf size and service frequency are closely relaed o each oher and his would affec he flee planning decision of airlines. By building a nesed logi model, Wei and Hansen (2005) invesigaed airlines decisions on aircraf size and service frequency. They revealed ha he service frequency and corresponding aircraf size of differen marke ypes could vary differenly according o passengers' choice model esimaion. Similarly, by esimaing passenger s fligh choice ha conribues o varying marke shares (ravel demand), Hsu and Wen (2003) deermined airline's fligh frequency a opimal profi. However, heir soluions of service frequency were 37

61 found o be lower han he acual fligh frequency of airline because he marke shares were underesimaed. This shows ha he changes of demand (demand flucuaion) as well as service frequency could no be capured precisely. Thus, a well-defined model is indeed required o deal wih airline's service frequency in order o mee sochasic demand desirably. Focusing on profi maximizaion of a game-heoreic model (under compeiive environmen), Hansen (1990) deermined he service frequency for he respecive airline, given he fligh frequency of he compeing airlines. However, he assumed he fixed airfare which in fac has an inelasic demand wih respec o price and service. Laer, Wei and Hansen (2007) developed game-heoreic models o analyze airlines decision-making on aircraf size and service frequency. The service frequency of respecive airline is deermined opimally based on maximum profi of airlines in a compeiive marke. Similarly, Wei (2006) employed game-heoreical model o inspec how airpor landing fees could affec he decisions of airlines on aircraf size and fligh frequency in order o produce airline s opimal profi. However, he sole focus on landing fees in affecing airline's decision-making migh be oo resricive. Furhermore, he demand flucuaion ha could affec aircraf selecion and is corresponding service frequency is no ackled. More recenly, airline's decision o deermine weekly fligh frequency (for differen aircraf ype) in response o aircraf emission charges could be 38

62 seen in Wen (2013) for which service frequency on individual roue is deermined by minimizing he operaing cos of a muli-objecive programming model. He showed ha some direc flighs of a paricular aircraf ype were shifed o one-sopover ransi flighs o reduce emission. This reveals ha aircraf ype would affec service frequency deerminaion. Wih he aim o maximize airline's profi, Lises and Dekker (2005) adoped scenario aggregaion-based approach o deermine he flee composiion (aircraf choice) o mee shor-erm sochasic demand. Apparenly, long-erm flee planning model may no be solved sraegically in view of he fac ha he developed model only capure he shor-erm demand. Despie selecing a paricular aircraf ype o mee ravel demand, service frequency of respecive operaing roue which deals closely wih aircraf choice is no deermined opimally o mee demand flucuaion. Besides, he accuracy of heir model o deermine he opimal profi may no be accurae in view of he specific airfare of passenger's class (business and economy) is negleced. From he airline's business principles, he airfare of differen passenger's class, as a major income for airlines, is a crucial elemen and hence his componen needs o be ackled appropriaely in flee planning. Hsu e al. (2011a, 2011b) formulaed sochasic dynamic programming model o solve flee planning problem by minimizing airline s cos. In spie of he respecive service frequency of exising operaing neworks, how airlines 39

63 make opimal decision o operae addiional service frequency o mee increasing demand is no explored explicily. Besides, heir formulaions migh be oo simplisic by considering ravel demand as sole consrain in flee planning. This may affec airline's services in providing a desired service frequency o mee demand flucuaion. Furhermore, he airfare of specific passenger's class is ignored and hence he accuracy of airline's profi and revenue are quesionable a some exen. From he aforemenioned sudies, i could be seen ha he service frequency of airlines is closely relaed no only o aircraf size and ype bu also o he demand flucuaion. In oher words, here is a srong ineracion beween supply (service frequency and flee composiion) and demand. This poins ou ha he service frequency of airlines which associaes closely wih flee combinaion (aircraf size/ype) needs o be moniored wisely o mee he demand flucuaion a a desired service level. This necessiaes he inclusion of service frequency in he flee planning model. Neverheless, unil oday, none of he exising sudies capure he service frequency explicily in solving he flee planning problem which deals closely wih aircraf composiion. Moreover, sochasic demand which has a grea impac on aircraf ype and service frequency is no aken ino consideraion by mos of he exising sudies. 40

64 2.4.2 Slo Purchase In view of he consan growh of air raffic (ravel demand), i.e. approximaely 5% annually (Inernaional Air Transpor Associaion, 2009), and he consrain of service frequency a paricular airpors, here will sill be a challenging issue o airlines o mee demand flucuaion profiably in such a compeiive airline indusry. In order o capure he service frequency explicily and o suppor he curren operaing neworks, slo purchase offers a greaer opporuniy o increase airline's service frequency in order o mee he ravel demand a a desired service level (Fukui, 2010; Babic and Kalic, 2011, 2012). For insance, US Airways had increased a oal of 142 flighs via slo purchase decision a LaGuardia Airpor, New York from year 1992 o 2000 (Fukui, 2010). Basically, when he airpor capaciy is no able o accommodae he requess of all airlines, he number of aircraf movemens is regulaed by airpor slos on a specific dae (Jones e al., 1993; Pellegrini e al., 2012). Slo conrols, he mos effecive demand-managemen ool, have been widely used a some major airpors especially in Europe and he Unied Saes (Mehndiraa e al., 2003). Based on he IATA sysem, slo allocaion is made wice a year, a he IATA Scheduling Conference (Babic and Kalic, 2012). Subsequenly, secondary slo rading was inroduced based on airline's willingness o pay for slos (Mo, 2006) for which slo price is commonly airpor-specific and grealy influenced by ime of he day, airline regulaion, ravel demand, ec. (Gillen, 2006). Due o he fac ha he service frequency of airlines is limied o a fixed number for a paricular ime period, airlines mus possess a slo for he 41

65 provided ime period for arrival/deparure (Mehndiraa e al., 2003). For airlines, slo purchase is exremely useful and vial o increase fligh frequency and operaions efficiency (Brueckner, 2009), o reduce delay (Mehndiraa e al., 2003; Gao e al., 2011), o mee flucuaing demand (Fukui, 2010) as well as generae more profi (Babic and Kalic, 2011, 2012). For air ravelers, Swaroop e al. (2012) highlighed ha slo purchase of airlines could improve ravelers' welfare, by providing a beer conneciviy beween flighs (wih lesser delay ime and more service frequency from airlines). There are some sudies ha discuss slo purchase, especially on he underlying benefis of slo purchase decision-making. Focusing on new fligh scheduling in order o expand airline roue nework, Babic and Kalic (2011, 2012) opimized he slo purchase decision-making (wih maximum revenue). They found ha slo purchase could increase airline's profi and service qualiy by adding new desinaion, increasing fligh frequency and improving schedule conneciviy. By using welfare based approach, Swaroop e al. (2012) analyzed he welfare effecs of slo conrols including he benefis from queuing delay reducion and coss. They showed ha slo conrol is effecive and i would improve ravelers' welfare by reducing sysem delays (wih addiional service frequency). Mehndiraa e al. (2003) esimaed he impac of slo conrols by adoping a marke-based allocaion mechanism. They showed ha slo conrol is consrucive for demand managemen as well as o alleviae unnecessary delay. By adoping price and quaniy-based approaches, Brueckner (2009) discussed he benefis of slo purchase o manage airpor congesion. Slo 42

66 purchase was found o be beneficial o airlines in providing efficien operaions as long as slo purchase decision is opimally made. Similarly, by comparing congesion pricing and slo rading, Basso and Zhang (2010) revealed ha he oal air raffic is higher under slo aucions and his in fac signifies ha slo purchase is able o mee more ravel demand. Focusing on compeiive markes, Fukui (2010) used regression analysis o examine wheher if sloholding airlines have resriced service expansion and marke enry by oher airlines. I was found ha alhough slo markes migh possess he poenial o enhance compeiion, here are sill pleny of improvemen areas in he slo markes. Thus, he resuls highlighed ha i is necessary o design addiional enhancemen mechanisms for slo rading sysem o yield more benefis o airlines. As revealed by he afore-menioned pas sudies, i could be seen ha slo purchase is cerainly beneficial o airlines in assuring higher profi, via a higher service frequency in meeing more demand. However, here is no exac approach or proper model in he exising sudies ha could assis airlines o make use of slo purchase wisely in providing appropriae fligh frequency o mee sochasic demand profiably. As such, a suiable and well-defined model is required o deermine opimal slo purchase as well as flee planning decision so ha airlines could mee sochasic demand desirably (wih opimal service frequency and flee composiion). 43

67 In overall, i could be inferred ha he service frequency of airlines (o mee corresponding demand level) is associaed closely wih he aircraf ype and size, hus i can' be denied ha slo purchase, a vial elemen o provide more services (wih addiional service frequency), would also influence he flee planning decision of airlines o a grea exen (in erms of opimal quaniy of respecive aircraf ype). In view of he implicaion of slo purchase in providing more services, i.e. a higher service frequency o mee increasing demand, which necessiaes he inclusion of slo purchase (wih associaed service frequency) in solving he flee planning problem. By having slo purchase, airlines would no only be able o improve heir services qualiy by providing a desired service frequency, via an adequae flee supply, o mee demand incremen (Brueckner, 2009; Fukui, 2010), bu would also be able o achieve passenger's saisfacion desirably. More imporanly, he incorporaion of slo purchase in flee planning would increase airline's revenue and profi (Babic and Kalic, 2011, 2012). This is definiely crucial for airlines o susain is profiabiliy in such a challenging airline indusry. In view of he fac ha flee composiion and airline's profi may vary o a grea exen by incorporaing slo purchase and service frequency of each operaing roue, here is a need o improve he exising approaches in solving he flee planning problem. I could be seen from he exising sudies, he availabiliy of slo purchase was negleced. This may resul in he unfeasibiliy of airline's flee supply o mee sochasic demand saisfacorily. Besides, he specific airfare of each passenger class, which is negleced by many pas 44

68 sudies, should be incorporaed necessarily o solve he flee planning problem in order o capure airline's business principles in a beer manner. 2.5 Green Flee Planning Three major environmenal issues peraining o air ranspor sysem are aircraf emission, noise, and fuel consumpion (Janic, 1999; IPCC, 1999; ICAO, 2010; Sgouridis e al., 2011). In he following subsecions, he conribuing facors ha cause environmenal issue are discussed accordingly. Subsequenly, several miigaion sraegies ha correspond o he respecive environmenal issues are discussed. Besides, some relevan sudies ha examine environmenal impacs are reviewed Environmenal Issues of Air Transpor Sysem In view of he increasing concern on green issues, i is crucial for airlines o idenify he conribuing facors ha could affec heir environmenal (green) performance. This is necessary no only o quanify he overall green impacs accuraely, bu also o carry ou improvemen sraegies effecively (for greener performance). I was found ha aircraf cruising aliude (Williams e al., 2002), load facor, aircraf age, cabin densiy configuraion (Miyoshi and 45

69 Mason, 2009), aircraf size and service frequency (Givoni and Rieveld, 2010) are he facors ha could affec aircraf emission level. The aircraf emission level varies significanly on cruising aliudes and fligh pahs (Williams e al., 2002). Generally, a lower aliude and longer cruising sage will end o generae more polluans. However, a high aliude of fligh may resul o he formaion of conrails ha causes negaive impac o he environmen. The aircraf load facor, i.e. a measure of uilizaion amoun of oal available capaciy of aircraf, has been recognized as one of he significan conribuing facors o aircraf emission. For a higher load facor, he fuel consumpion of aircraf is lower (in erms of uni load facor) and hence he corresponding emission level ends o be lesser. Therefore, an increasing load facor was found o be more environmenal beneficial, paricularly due o a lower amoun of polluans per uni load facor (Miyoshi and Mason, 2009; Givoni and Rieveld, 2010). Anoher conribuing facor worh menioning is he cabin densiy configuraion, i.e. he srucure of seas supplied for which he aircraf wih a higher seaing densiy would increase aircraf weigh and hence more emission would be produced (Miyoshi and Mason, 2009). Besides, aircraf age have an influenial impac in aircraf emission. Miyoshi and Mason (2009) menioned ha aircraf echnology could be a deermining facor in his aspec. Usually, newer aircraf wih advanced echnologies (by incorporaing a beer fuel efficiency sysem) would emi lesser emission compared o aging aircraf (Janic, 1999). In addiion, aircraf size also influences he emission level. Usually, a smaller aircraf (single-aisle) 46

70 which is operaed for shor-haul neworks emi lesser polluans compared o large aircraf (win-aisle) for long-haul neworks (Givoni and Rieveld, 2010). Larger aircraf produces more emission, mainly due o a large proporion of carbon emission, especially from cruising sage of long-haul flighs (Morrell, 2009; Miyoshi and Mason, 2009). As such, i could be deduced ha aircraf size would produce differen emission level. Furhermore, service frequency is also one of he conribuing facors. By having a higher service frequency, airlines would consume more fuel o suppor heir operaing neworks and hence he level of aircraf emission would increase proporionally. In oher words, a higher service frequency (more flighs) would consequenly emi more aircraf emission (due o more fuel burning). The level of aircraf noise emied from aircraf operaions depends on several facors, such as aircraf ype (Janic, 1999) and aircraf rajecories (Clarke, 2003; Visser, 2005; Pras e al., 2010, 2011). Heavier aircraf usually generaes louder noise due o more powerful engine seing (ICAO, 2011). I was found ha he engine pars such as fan, compressor and urbine are he main sources of aircraf noise. Aircraf rajecories conribue o noise during he ake-off and landing sage by having differen fligh speed, hrus seing as well as flap and sla configuraion (Pras e al., 2011). Besides, he aircraf noise level produced by a paricular fligh rajecory (during ake-off and landing sages) is also relaively influenced by he navigaion sysem and erminal airspace. 47

71 In erms of fuel consumpion, Janic (1999) and Morrell (2009) highlighed ha echnological innovaion is one of he significan facors affecing aircraf fuel consumpion level. They revealed ha improved echnology on engine propulsive and hermal efficiency could resul in more fuel savings. Aircraf engine wih a higher bypass raio would also have lower fuel consumpion (Janic, 1999). Besides, advances in srucures or maerials o develop a new generaion of aircraf would be able o reduce aircraf weigh and fuel consumpion. Airbus (2013) claimed ha he fuel consumpion of A380 is abou 17% lesser (per passenger) han is compeior. This is achieved by having a highly aerodynamic and efficien fuselage design and also he usage of innovaive composie maerials o reduce weigh. For aircraf operaions, i is imporan o noe ha fuel consumpion direcly conribues o aircraf emission level. In erms of aircraf size, Morrell (2009) showed ha fuel efficiency appears o be higher for smaller aircraf (especially for shor/medium-haul) comparing o a larger size of aircraf (for long-haul). Smaller aircraf was found o be more fuel-efficien mainly due o is sea densiy and load facor which is usually higher han larger aircraf. This shows ha he aircraf ype in erms of aircraf size wih varying sea densiy and load facor would affec he fuel efficiency of airlines. Specifically, Tsai e al. (2014) showed ha a lower fuel consumpion (and hence emission level) could be achieved by reducing he weigh of seas in passenger cabins. Abdelghany e al. (2005) showed ha fuel managemen sraegy in response o aircraf's operaional condiions would 48

72 affec he fuel efficiency of airlines. In general, excessive fuel loading (paricularly o serve subsequen fligh) would add on o he aircraf weigh and heavier aircraf would consume more fuel. In addiion, Nikoleris e al. (2011) showed ha idling and axiing saes a consan speed or braking emerged o be wo larges sources of fuel burn during landing and ake-off (LTO) cycle, which accouns abou 18% of fuel consumpion. The fuel efficiency is relaively sensiive o hrus level assumpions (i.e. 5% and 7% respecively for axiing and urning saes) and depends very much on he number of sops during axi, duraion of each sop, number of urns on axiway as well as acceleraing ime Miigaion Sraegies Various miigaion sraegies are proposed and in-place o alleviae deerioraing environmenal problems resuling from aircraf aciviies. These sraegies could be caegorized ino hree caegories, namely echnological innovaion, operaional and flee, policy and rules & regulaions. In erms of echnological innovaion, an improvemen in engine and aerodynamics design as well as using lighweigh maerial o reduce aircraf weigh are found o be beneficial o he environmen (Hellsrom, 2007; Sgouridis e al., 2011). Miyoshi and Mason (2009) showed ha a newly 49

73 developed aircraf hrough he incorporaion of echnological advances ino he flee (such as B and B777) produced lower emission han older generaion aircraf. Besides, Morrell (2009) repored ha mos of he efficiency gains of B come from new echnology. Janic (1999) highlighed ha he improvemen in he engine s propulsion and hermal efficiency has increased he engine pressure raio and urbine emperaure for which he engine wih a higher bypass raio (e.g. B777) has lower fuel consumpion. Generally, fuel consumpion decreases by abou 4% for each incremen in he engine's bypass raio. Besides, he inroducion of he high bypass echnology o he aircraf engine has reduced he engine noise significanly (Air Transpor Acion Group, 1996). A he same ime, he engine has become bigger and sronger o propel bigger and faser aircraf. The larger and faser (more producive) aircraf which is powered by sronger urbofan and high bypass engines have generaed a lower level of noise. Furhermore, urbofans wih ulra-high-bypass raio and open roor prop-fans are idenified as possible soluions o reduce aircraf noise (Smih, 1992). Besides, Airbus (2013) repored ha he larges aircraf A380 is producing ulra low noise wih a significan reduced aircraf weigh hrough he use of lighweigh maerials. In erms of operaional efficiency, he improvemen effors include fligh opimizaion (e.g. by generaing opimal aircraf rajecories) and ground operaion opimizaion such as aircraf axiing operaion (Sgouridis e al., 2011). Recen advances in navigaion echnology have guided he cockpi crew o operae effecively and safely under Insrumen Fligh Rules (IFR) 50

74 environmen. For insance, he Area Navigaion (RNAV) sysem allows he pilo o creae aircraf rajecory based on a series of arbirary reference poins while he Global Posiioning Sysem (GPS) generaes precise esimaes for a paricular posiion a any locaion around he world (Logsdon, 1992). By combining RNAV and GPS, he sysem enables he approach and deparure rajecories o be adjused for noise reducion. Similarly, Visser (2005) inroduced a noise opimizaion ool ha could generae aircraf rajecories or fligh pahs for boh arrivals and deparures which reduce aircraf noise impac under operaional and safey consrains. Furhermore, Clarke (2003) developed a simulaion sysem ha aims o assis air raffic conrollers in deermining appropriae sequencing and spacing for opimal (maximum) akeoff and landing raes in heavy raffic condiion. Besides, by increasing he densiy cabin configuraions and load facor, he emission rae per passenger could be reduced (Miyoshi and Mason, 2009). In addiion, he lower service frequency (by operaing larger aircraf) could produce lesser emission and hus a larger aircraf for shor-haul operaion is encouraged in order o reain similar capaciy in meeing ravel demand (Givoni and Rieveld, 2010). In erms of policies and rules & regulaions, noise charge is imposed on airlines ha have generaed noise level over allowable limi. Generally, noise charges are imposed based on individual aircraf or cumulaive noise recorded. Individual aircraf noise charge is compued based on aircraf s maximum akeoff weigh (MTOW) in accordance wih he Federal Aviaion Adminisraion (FAA) noise cerificaion sandards. Heavier aircraf 51

75 cusomarily incur higher landing and noise charges and busier airpors charge higher noise fee per landing. Noise charges, however, vary across airpors, i.e. depending grealy on operaion ime of deparure and/or arrival (Girvin, 2009). On he oher hand, cumulaive noise limi (in he form of noise quoas/limis) refers o noise exposure over a specific period. I is imposed by airpors o conrol he oal noise generaed by airlines. Based on yearly maximum oal noise level for each planning year a airpors, he airpor auhoriy would curail he operaions of airlines if heir operaions exceed he regulaed limi of noise volume. In he Unied Saes, he airpor auhoriy may increase or reduce he fligh slos available o airlines based on heir cumulaive noise exposure from previous year (Girvin, 2009). Nighime curfew is anoher mos common noiseabaemen measure for which airpors ban aircraf operaions over a predeermined nigh ime period and enforce penalies on airlines ha violae he curfew. Operaions during curfew hours are limed o a maximum number and curfew regulaions are highly airpor-specific. Some of he examples of airpors ha regulae noise limis for dayime and nigh ime operaions are he UK s Leeds, Czech Republic s Prague and Ausria s Salzburg. Emission charge is inroduced o reduce aircraf emission level. The charge is generally compued based on nirogen oxide and hydrocarbon emission level a airpors. This concep was inroduced in Swizerland and Sweden in 1997 and 1998, respecively. There were five classes of emission charges in Swizerland and seven classes in Sweden ha are ranked according o a specific emission level of urbofan engines (Scheelhaase, 2010). In 2003, 52

76 he European Civil Aviaion Conference (ECAC) creaed ERLIG formula ha provides a sandard approach o compue emission level from aircraf engines (ECAC, 2003). This was hen adoped by he European airpors, including London Heahrow airpor in 2004 and Munich airpor in However, emission charges vary across airpors. For insance, 3 per emission value uni (in on) is charged in Germany, 5.5 in Sweden, 1-3 in Swizerland and 1.60 in he UK. Generally, emission charge differs depending on he ype and number of engines of aircraf. In Germany, emission charges of B (wih JT9D-7FW engines) a are much higher han he emission charges of A (wih CFM56-5C4 engines) a Generally, low emission charges are levied for small urboprops and high emission charges are se on bigger and heavier aircraf, mainly due o more powerful engines han smaller aircraf (Scheelhaase, 2010). Alhough aircraf engine was shown o be a major facor on emission charge, Scheelhaase (2010) highlighed ha he decision on he engine ype used on aircraf depends on a bundle of managerial facor, no jus on he implemenaion of emission charges. More recenly, he European Union (EU) implemened Emission Trading Scheme (ETS) o alleviae global warming by reducing carbon dioxide emission. Saring wih inra-european flighs in 2011, i's required for he airlines o hold allowances for heir carbon dioxide emission and non-european airlines was included from 2012 for heir aircraf operaions ha operae in and ou of he EU (Albers e al., 2009). Under his scheme, airlines will obain an iniial se of free-of-charge allowances, i.e. 85% based on he

77 average emission while he remainder (15%) being aucioned (Wen, 2013). However, airlines have o acquire addiional allowances if hey require more. The allowance (emission) price of ETS ranges beween 10 and 30 for each on of carbon dioxide emission (European Commission, 2005; Erns and Young, 2007; Scheelhaase and Grimme, 2007). Approximaely, his would resul in an addiional cos of 20/on of carbon dioxide per passenger. However, he implemenaion of ETS sysem in aviaion secor received responsive debaes since is proposal Environmenal Assessmen Approaches Till o dae, here are very limied sudies ha quanify he environmenal impacs of ransporaion secor. To assess he environmenal impacs for a highway roue and paving projec, Boclin and Mello (2006) presened a decision suppor mehod by using fuzzy logic approach. They showed ha park-highway is he mos promising alernaive in giving he bes ecological, economic and social performance. Rossi e al. (2012a, 2012b) examined hree-dimensional concep of susainabiliy o idenify he preferences of decision makers and also o obain he mos imporan characerisics of alernaive ransporaion policies. The limiaion of hese sudies is ha hey primarily focused on ransporaion alernaives analysis and forecasing for which here is no exac quanificaion approach ha could be used o evaluae he green performance of ranspor operaors. In oher fields 54

78 (no ransporaion sysem), Silver (2000) evaluaed he impacs of finfish mariculure on coasal zone waer qualiy by adoping he fuzzy logic approach. Four fuzzy ses (nil, moderae, severe and exreme impacs) were defined and he corresponding parial memberships have been combined o yield a single comprehensive score as an overall measure of environmenal qualiy. Valene e al. (2011) caegorized old mining sies and described heir environmenal impac as low, medium and high. They showed ha he use of fuzzy logic o obain he environmenal impac index allowed he inegraion of quaniaive and qualiaive componens. Some oher relevan sudies o evaluae he environmenal impacs by employing fuzzy logic could be seen in Andrianiasaholiniaina e al. (2004), Shepard (2005) and Peche and Rodriguez (2009). However, hese sudies grealy depend on fuzzy membership of he concerned variables which, in fac, does no possess a clear and specific mechanism for exac formulaion or combinaion. The way o inegrae he memberships of variables depends on real applicaion and his would be geing difficul for complex siuaions (Silver, 2000). As poined ou by Valene e al. (2011), he membership funcions are generally formed wih he aid of some probabiliy disribuion. In oher words, he resuls are disribuion-oriened. Ye, i is imporan o noe ha in real pracice, some concerned variables may no possess specific disribuion. Even if hey do, he way o idenify he bes disribuion may no be easy and sraighforward. Besides, Singh e al. (2012) presened an overview of susainabiliy assessmen mehodologies, including environmenal susainabiliy index (ESI), 55

79 environmen qualiy index (EQI) and environmenal performance index (EPI). However, mos of he assessmen approaches generae composie index merely based on aggregae value or he weighed sum value of relevan indicaors. These approaches are relaively simplisic o a cerain exen for which here is no clear indicaion on he applicaion for more complicaed problem. Furhermore, hese indices did no capure he occurrence of unexpeced even. The occurrence of unexpeced even should be incorporaed as i would affec he operaions of ranspor operaors and hence he environmenal performance will vary differenly. None of he indices address his aspec in quanifying he environmenal performance. Apparenly, many research sudies had shown ha he choice of aircraf ype, size, age, and aircraf echnology are among he key facors in addressing environmenal issue. As such, he firs and he bes sep o deal wih aviaionrelaed environmenal problem is o consider having a green flee ha produces he leas polluion impac o he environmen. By having green flee, airlines could hen furher opimize heir operaions o minimize he environmenal impac. In he ligh of his, here is a need o consider he environmenal facor during flee planning (Rosskopf e al., 2014). Pas sudies such as Lises and Dekker (2005), Wei (2006), Wei and Hansen (2007), Pifield e al. (2009), Hsu e al. (2011a, 2011b) and Wen (2013) have a limiaion in addressing he environmenal issue in flee planning. Mos of hem considered revenue and profi only as he main objecive when making opimal decision in flee planning. Recenly, Rosskopf e al. (2014) formulaed flee opimizaion model 56

80 as a muli-objecive, mixed-ineger programming model wih he aim o balance he economic and environmenal goals in flee planning. By maximizing airline's asse value and minimizing oal nirogen oxide emissions from fligh operaions, hey showed ha airline would have o deviae abou 3% from is economic opimum o improve a 6% of he environmenal goal. However, he occurrence of unexpeced even ha would affec aircraf operaions is no ackled. Furhermore, only nirogen oxide is considered o reduce environmenal impacs. In fac, aircraf noise and fuel efficiency are also crucial environmenal facors ha ough o be aken ino consideraion o improve he green performance of airline. Some crucial pracical consrains for flee planning, e.g. aircraf range consrain, lead ime consrain and selling ime consrain, are also lef ou. In addiion, here is no clear indicaion on how o quanify he weighs for he economic and environmenal goals in solving he flee planning problem. As such, here is a need for furher research effor peraining o his issue. 2.6 Summary In overall, he limiaions of he exising sudies and he needs for improvemen could be summarized below: For ravel demand forecasing: Mos of he sudies forecased deerminisic demand (inelasic) which could no capure demand flucuaions. Hence, he resulan forecas is no robus 57

81 and his may resul in he loss of opimaliy. Pas sudies ha capured sochasic demand did no consider he occurrence of unexpeced evens in modeling sochasic demand. Some sudies assumed he fixed ype of disribuion o quanify demand flucuaion. This migh no be realisic. Some sudies focused on shor-erm demand forecasing. This may no be applicable o solve long-erm flee planning problem. In view of he fac ha he ravel demand of airlines behaves in a sae of uncerainy (sochasic) primarily due o he occurrence of unexpeced evens which is unpredicable in he real pracice, airlines would require a welldefined modeling framework o model sochasic demand. The developed modeling framework of sochasic demand (as described in Chaper 3) is no limied o any saisical disribuion in solving he long-erm flee planning problem under uncerainy. For airline flee planning approach: Alhough many sudies focused on he analysis beween aircraf size and service frequency, here is no proper mechanism on how airlines could acquire/lease specific aircraf ype (wih corresponding service frequency). The occurrence of unexpeced even is negleced in solving flee planning problem. Some sudies assumed ha compeing airlines know all he available informaion in selecing he opimal aircraf size and service frequency. This migh no be sensible. 58

82 I migh be oo resricive wih sole dependence on paricular consrain (e.g. landing fees, demand consrain) o obain opimal flee planning decision. Pas sudy did no consider possible roue disance and aircraf weigh despie he fac ha aircraf specificaion (including aircraf range and engine weigh) would affec he aircraf composiion and performance. Apparenly, here is no sudy ha formulaes a proper flee planning model o opimize aircraf acquisiion/leasing decision o mee sochasic demand under uncerainy. To obain opimal flee planning decision for each operaing period hroughou he planning horizon, numerous pracical consrains ha realisically capure various echnical and operaional consideraions of airlines (including aircraf performance) have o be included necessarily. This is vial o assure ha he aircraf operaions of airlines are pracically viable o suppor he curren operaing neworks a a desired and profiable service level. For sraegic flee planning modeling framework: Pas sudies primarily focus on he echnical aspec in solving flee planning problem bu did no quanify he key aspecs (probable phenomena) of flee planning decision-making. Exising sudies did no consider mode choice analysis in making flee planning decision as hey only focused on he raveling aribues of airlines and raveler s preference. For mode choice analysis, he sudy area of mos of he pas sudies was limied o European counries. 59

83 Oher ypes of ground ranspor (e.g. bus, car) and specific ype of airlines (e.g. low-cos airlines) were no considered explicily by pas sudies. Some sudies were carried ou many years ago and he findings may no reflec he curren demand rend and raveler's response. From he aforemenioned limiaions, i could be seen ha supplydemand ineracion is no sudied explicily by he exising sudies. For airlines, he supply-demand ineracion is crucial in view of he needs and expecaions of ravelers (demand) which would affec airline s service (supply) o a grea exen. Therefore, a sraegic flee planning modeling framework is developed o opimize he flee planning decision of airlines by incorporaing mode choice analysis and subjecive percepions of airline s managemen (decision makers) for which he key aspecs (probable phenomena) of flee planning decisionmaking plays he role o assure he feasibiliy of aircraf operaions in supporing he operaing neworks. By providing a desired service level (i.e. opimal supply via flee planning decision), airlines could reain no only a higher profi level bu also he ineres or loyaly of heir passengers. For service frequency deerminaion in flee planning: Exising sudies did no show how airlines could provide service frequency adequaely (wih corresponding aircraf ype) o mee increasing demand. Some sudies highly depended on paricular consrain (e.g. landing fees, demand consrain) o deermine service frequency. This migh be oo resricive. 60

84 Demand flucuaion ha could affec aircraf selecion and is corresponding service frequency is no ackled. Pas sudies only assigned a paricular aircraf ype o mee ravel demand for which he service frequency of respecive operaing roue which deals closely wih aircraf choice is no deermined opimally o mee demand. Some sudies assumed he fixed airfares which imply an inelasic demand wih respec o price and service. The specific airfare of passenger's class (business and economy) is negleced and hence he accuracy of airline's profi and revenue are quesionable. There is no proper flee planning model in he exising sudies ha could assis airlines o make use of slo purchase wisely in providing appropriae fligh frequency o mee sochasic demand profiably. For flee planning sudies, exising sudies ignored he availabiliy of slo purchase. This may resul in he unfeasibiliy of airline's flee supply. To mee sochasic demand a a desired service level, airlines would need o possess a proper aircraf composiion ha could provide an appropriae service frequency, righ on ime, o suppor he operaing neworks under uncerainy. This is paricularly imporan for airlines o mee increasing demand no only o assure a higher profi level bu also o susain compeiively. However, i could be seen ha here is no exising sudy ha could provide proper mechanism o assis airlines o provide addiional service frequency under numerous pracical consrains (including he regulaed limis of aircraf operaions a paricular airpors). As such, slo purchase plays a vial role o provide addiional service frequency (paricularly o mee increasing demand) and a well-defined flee planning model is indeed required o 61

85 deermine opimal aircraf composiion and corresponding service frequency. Besides, he developed flee planning model should include he specific airfare of passenger's class (business and economy). From he airline's business principles, he airfare of differen passenger's class, as a main income for airlines, is an essenial elemen and hence his componen needs o be ackled necessarily in flee planning. For green flee planning: Many sudies considered he single environmenal facor (aircraf emission, noise and fuel efficiency) or miigaion sraegy a one ime. Hence, he impac of he respecive sraegy on oher green issues could no be capured. Pas sudies focused on ransporaion alernaives analysis and forecasing, i.e. here is no exac quanificaion approach ha could be used o evaluae he green performance. Exising sudies are grealy dependen on fuzzy membership which is generally formed by some probabiliy disribuion, i.e. he resuls are disribuion-oriened. Mos of he exising assessmen approaches generae a composie index based on aggregae value or weighed sum value of relevan indicaors. These approaches are relaively simplisic o a cerain exen. Mos of he pas sudies of flee planning considered revenue and profi only as he main objecive when making opimal decision in flee planning. 62

86 The occurrence of unexpeced even ha would affec aircraf operaions is no ackled. Only a paricular polluan (e.g. nirogen oxide) is considered o reduce environmenal impacs. Some oher environmenal facors, e.g. aircraf noise and fuel efficiency are negleced. Some crucial pracical consrains for flee planning, e.g. aircraf range consrain, lead ime consrain and selling ime consrain, are lef ou. There is no clear indicaion on how o quanify he weighs for he economic and environmenal goals in solving flee planning problem. While meeing sochasic demand a a profiable level, environmenal issues could no be negleced in view of he increasing concerns of green issue nowadays. However, here is no sudy ha incorporaes green concern in solving he flee planning problem. By formulaing green flee planning, airlines would deermine he opimal aircraf ype and quaniy ha could minimize environmenal impacs while aaining maximal profi. Insead of a single environmenal facor, he developed model could quanify he overall green performance of airlines. I would also evaluae he effeciveness on respecive environmenal facor (aircraf emission, noise and fuel efficiency). Furhermore, airlines could susain a significan amoun of cos savings if green flee planning is carried ou wih some beneficial improvemen sraegy (o yield a greener performance). 63

87 CHAPTER 3 FLEET PLANNING DECISION MODEL UNDER STOCHASTIC DEMAND 3.1 Making Opimal Aircraf Acquisiion and Leasing Decision under Sochasic Demand This chaper (wih hree major secions) oulines he developmen of long-erm flee planning decision model o mee demand flucuaion which is sochasic in naure. To capure he demand flucuaion (for firs secion), a novel modeling framework of sochasic demand is necessarily developed o deermine he level of ravel demand under uncerainy. In he framework, he probabiliy of possible occurrence of demand uncerainy is quanified in erms of Sochasic Demand Index (SDI). To solve he flee planning problem of airlines, an aircraf acquisiion decision model (in second secion) is formulaed o deermine he opimal quaniy and aircraf ype ha is o be acquired (wihou aircraf leasing) o mee sochasic demand under numerous pracical consrains. This model is able o assure ha sochasic demand (represened by a paricular probabiliy disribuion) is me profiably a a desired service level. In view of he fac ha aircraf leasing also plays a vial role in providing an adequae flee supply o airlines in supporing heir operaing neworks, an opimal aircraf acquisiion and leasing decision (in hird secion) is hen developed accordingly wih he aim o deermine he quaniy and aircraf ype 64

88 ha is required by airlines (via acquisiion and/or leasing) o mee he sochasic demand desirably. The level of sochasic demand is quanified by using he SDI (based on he modeling framework of sochasic demand). The developed models are able o make opimal flee planning decision for each operaing period hroughou he long-erm planning horizon. In order o examine he feasibiliy of he developed mehodologies, illusraive case sudies were presened and solved in he form of linear and nonlinear programming models (depending on airlines' operaional daa). Concisely, he findings empirically deduced ha he developed mehodologies are effecive and beneficial for airlines o mee sochasic demand profiably under uncerainy. 3.2 Modeling Sochasic Demand under Uncerainy Globally, airlines forecas he fuure growh of ravelers annually in order o obain he laes rend of ravel demand. Typically, forecasing (or predicion) of demand growh is found o be posiive (i.e. implying posiive growh) in accordance wih he increase in populaion size and income level (Malaysia Airlines, 2010a; Inernaional Air Transpor Associaion, 2010). However, when here is an occurrence of an unprediced even which could affec raveler s decision, here would be a reducion in demand during a cerain period of ime. This is referred o as a negaive effec. A 5-sep modeling framework (as displayed in Figure 3.1) is developed o deermine he 65

89 level of demand flucuaion. In he framework, a Sochasic Demand Index (SDI) is defined o quanify he probabiliy of possible occurrence of demand uncerainy. I is assumed ha he value of SDI for he base year (year 0) is 1. The Mone Carlo simulaion (Taha, 2003; Winson, 2004) is used o deermine he occurrence probabiliy of posiive and negaive effecs wih no prior assumpion of a fixed disribuion. The general procedure of he developed framework is elaboraed as follows: Sep 1: Deermine possible occurrence of unexpeced even Consider a se of uncerain evens ha could affec ravel demand. For example, he occurrence of biological disease, economic downurn and naural disaser which could ake place unexpecedly in real life. The probabiliy disribuion of hese evens is deermined. To form he respecive probabiliy disribuion, airlines could obain and analyze he hisorical daa of unexpeced evens over a period of ime. Sep 2: Deermine he probabiliy of unexpeced even (negaive effec) Based on he pre-deermined probabiliy disribuion in Sep 1, he probabiliy of unexpeced evens is simulaed by using he Mone Carlo simulaion. The probabiliy of occurrence can be expressed as follows: C H c1 h1 1,if h happens 0,if h does no happen for 1,...,, (3.1) PP P hc T for which P hc is he probabiliy ha he unexpeced even, c happens wih a possible occurrence of h. 66

90 Figure 3.1: Modeling Framework of Sochasic Demand Sep 3: Deermine he possible incremen of forecased demand The possible incremen of forecased demand (posiive effec) needs o be esimaed. Demand growh is esimaed based on pas ravel rend (from he hisorical daa published by airlines or Non-Governmen Organizaion) and fuure ravel rend forecasing. The probabiliy disribuion ha describes he projeced growh of demand needs o be modeled as well. 67

91 Sep 4: Deermine he probabiliy of he possible incremen of forecased demand Based on he demand growh as projeced in Sep 3, he possible incremen of forecased demand for each operaing period as well as is probabiliy is deermined accordingly wih he aid of Mone Carlo simulaion. Sep 5: Deermine he value of SDI for each operaing period For each operaing period, he SDI, Index, is deermined subjec o boh posiive and negaive effecs. The probabiliy of boh effecs are compiled ogeher o work ou he SDI owing o he fac ha he level of sochasic demand is affeced no only by he occurrence of unexpeced even (negaive effec) bu also influenced posiively by demand growh (posiive effec). By considering boh effecs (i.e. o sum up boh effecs), he SDI could be expressed as follows: Index PP D 1 for 1,..., T (3.2) f ( inc) for which he consan of 1 is he index value for he base period (year 0). Specifically, Index 1 means ha in overall (due o boh posiive and negaive effecs), he level of sochasic demand of year is higher han he level of demand in previous year (i.e. year 1). Similarly, Index 1 indicaes ha he level of sochasic demand of year is lower han he level of demand of previous year, 1. Index 1 implies ha he demand of year and is previous year (i.e. year 1) is he same. This is possible due o he naure of uncerainy and he growh of demand which is sochasic. 68

92 By using he SDI, he demand level of each operaing year, D, is deermined based on he following equaion: D Index D0, for 1 Index D 1, for 2,..., T (3.3) For operaing year 1, he demand level is deermined by using he convoluion algorihm as described by Winson (2004) (as oulined in Appendix A). According o Winson (2004), convoluion algorihm can be adoped o generae normal random variaes. Besides, his algorihm incorporaes random number, which is a significan componen for simulaion o capure he vagueness and randomness. The demand level of airline's projeced demand could be defined as follows: D for which he forecased demand, 0 12 f f Rr 6 (3.4) r1 D has mean f f and sandard deviaion f. For Equaion (3.4), he componen of Rr signifies he respecive random number ha is needed o work ou he modeling of sochasic demand. The componen of R, as pars of he Convoluion Algorihm (Winson, 2004), is r needed mainly o capure demand flucuaion which is sochasic in he naure. Noe ha for subsequen operaing period, he level of sochasic demand is deermined by considering he curren SDI and he level of sochasic demand of previous operaing period. 69

93 3.2.1 An Illusraive Example (To Deermine he Probabiliy of he Occurrence of Unexpeced Evens) In order o deermine he probabiliy of he occurrence of unexpeced evens by using Equaion (3.1), consider wo ypes of unexpeced evens, i.e. c 1, 2 which respecively represens biological disaser (flu) and economic recession for he operaing period 5 (i.e. 5 ). The relevan informaion for each unexpeced even is summarized in Table 3.1. For he case ha biological disaser (flu) and economic recession o happen simulaneously, he probabiliy of occurrence of hese unexpeced evens can be compiled accordingly (based on Equaion (3.1)) as below for he operaing period 5: PP c1 h x P h2 5 5 h1 h1 P P P P (1) 21 (0) x 12 (0) 22 (1) 0.60 (1) 0.40 (0) x 0.89 (0) 0.11 (1) % P hc P h The componen of 7% indicaes he probabiliy for which he biological disaser and economic recession o happen simulaneously in he operaing period 5. 70

94 Table 3.1: The Informaion of he Respecive Unexpeced Even Unexpeced even, c Saisical disribuion (Poisson disribuion wih mean ) Possible occurrence, h and is probabiliy (where X is he occurrence quaniy of unexpeced even) Acual occurrence of h in he operaing period 5 (can be deermined wih he aid of simulaion) Remarks: Biological disaser Economic recession, (flu), c = 1 c = 2 Pois 7 1 Pois 9 P X h P X P X h P X For h 1, For h 2, For 1, For 2, h 1 h 2 1. For biological disaser (flu), he acual occurrence of h = 1 indicaes ha he probabiliy for which he biological disaser (flu) o happen a mos 7 imes (in he operaing period 5) is For economic recession, he acual occurrence of h = 2 indicaes ha he probabiliy for which he economic recession o happen in he operaing period 5 is Aircraf Acquisiion Decision Model Assume ha here is a choice of n ypes of aircraf ha could be purchased and operaed for a se of origin-desinaion (OD) pairs. The objecive of aircraf acquisiion decision model is o find opimal quaniy and ype of aircraf ha should be purchased in order o maximize he operaional profi of airlines. The level of demand is sochasic and i could be expressed by some random disribuions. To deal wih his sochasic elemen, aircraf acquisiion problem is formulaed as a probabilisic dynamic programming problem. This approach is adoped primarily due o is abiliy o decompose he long-erm flee planning problem ino a chain of simpler sub-problems for more racable opimal soluions. The objecive funcion is o maximize he expeced profi of airlines by considering various pracical consrains in flee planning. For he developed model, he operaing period,, in erms of years is he sage variable 71

95 of he model while he sae variable a each sage consised of various inercorrelaed variables, namely he quaniy of aircraf o be purchased (i.e. main decision variable), iniial quaniy of aircraf owned, quaniy of aircraf o be sold, quaniy of aircraf o be ordered and quaniy of aircraf o be released for sales. Wih he aim o maximize airline's expeced profi, he opimal decision (i.e. alernaives a each sage) is he acquisiion decision of new aircraf o mee sochasic demand while making decision o sell aging aircraf. The suiabiliy of he developed long-erm flee planning model could be explained by he lead ime and order placing ime (in advance) of aircraf acquisiion. According o some airlines (Malaysia Airlines, 2010a; AirAsia Berhad, 2010a), he acquisiion of new aircraf requires a period of five years (in average) o be compleely delivered by he aircraf manufacurer. Under cerain circumsances (e.g. manufacuring issues), he acual lead ime migh be longer han he agreeable lead ime (beween he airline and aircraf manufacurer). This will resul in he lae or delay of aircraf delivery and hence he airline would receive he new aircraf much more laer. Besides, he airlines also have o place heir acquisiion/leasing order in advance (earlier) in order o receive he respecive aircraf on ime for operaions. As such, i could be deduced ha he developed long-erm flee planning model (wih he corresponding demand forecasing) is reasonably and pracically needed o opimize aircraf acquisiion/leasing decision. 72

96 3.3.1 Consrains There are some pracical consrains ha need o be considered in opimizing aircraf acquisiion decision model. They are explained as follows: Budge consrain This is he mos pracical consrain in order o ascerain ha he soluion obained is financially feasible for airlines. Accordingly, oal purchase cos of aircraf should be less han or equal o airline s allocaed budge for aircraf acquisiion. This consrain could be expressed as follows: n P purc x MAX i i budge () for 1,..., T (3.5) i1 Demand consrain Le indicaes he significance level o mee sochasic demand, he following expression can be formulaed o achieve airline's argeed level of service. where 1 n S i S P SEATi f D, A D 1 for 1,..., T; S s,..., sk (3.6) i1 1 is he confidence level (i.e. argeed service level) of airlines o mee sochasic demand while P is he probabiliy of occurrence of a desired service level. Noe ha sochasic demand can be represened by some probabiliy disribuions. If ravel demand is assumed o follow normal disribuion wih mean and sandard deviaion, demand consrain could 73

97 be expressed as follows: 1 S i n SEATi f D, A F (1 ) for 1,..., T; S s1,..., sk (3.7) i1 where F 1 (1 ) is he inverse cumulaive probabiliy of 1-. Parking consrain When aircraf is off-duy, i has o be parked a he hangar or apron of airpor. In such a case, aircraf selecion would someimes be consrained by he geomery layou of airpors. As such, parking consrain is ough o be considered feasibly. This consrain could be oulined as follows: n m i1 y0 P P I x size PARK for 1,..., T (3.8) iy i i Sales of aircraf consrain For some airlines, aging aircraf which is less cos-effecive migh be sold a he beginning of a cerain operaing period when airlines make decision o purchase new aircraf. However, o mainain a cerain level of operaional efficiency, he quaniy of aircraf sold should no be more han wha was possessed by airlines. This consrain can be expressed as follows: sold I for 1,..., T; i 1,..., n; y 1,..., m (3.9) P iy ( 1) i( y1) Order delivery consrain The delivery of new aircraf is grealy dependen on he efficiency of aircraf manufacurer. Someimes, here migh 74

98 be a delay in delivering new aircraf. As such, he aircraf ha one could purchase should no be more han he number of aircraf available in he marke. This consrain can be expressed as follows: P x ORDER for 1,..., T; i 1,..., n (3.10) i Lead ime consrain I is imporan o noe ha in real pracice, airlines would ge an agreeable lead ime (he period beween placing and receiving an order) from aircraf manufacurer when hey order new aircraf (o be purchased). However, he real lead ime migh be longer han he agreeable lead ime and his will resul in he delay of aircraf delivery. This signifies ha lead ime consrain is necessary as i is able o indicae when airlines are supposed o place an order for new aircraf. This consrain can be expressed as follows: for 1,..., ; 1,..., P RLT DLT T i n (3.11) i i By assuming ha lead ime is normally disribued wih mean LT and sandard deviaion where 1 1 LT, his consrain could hen be saed as follows: DLT F 1 (1 ) for 1,..., T; i 1,..., n (3.12) i LT LT F is he inverse cumulaive probabiliy of 1-. Selling ime consrain Airline's aging aircraf which is less effecive migh be sold during a paricular operaing period. In such a case, airlines need o know he mos suiable ime o release heir aging aircraf for 75

99 sales paricularly o look for prospec buyers in advance. In real pracice, he real selling ime migh be longer han he desired selling ime. Therefore, his consrain is formed wih he aim o reduce he possibiliy of his inciden as leas as possible. This consrain could be defined as follows: for 1,..., ; 1,..., P RST DST T i n (3.13) i i Subsequenly, his consrain could be saed as follows by assuming selling ime is normally disribued wih mean ST and sandard deviaion ST : where 1 1 DST F 1 (1 ) for 1,..., T; i 1,..., n (3.14) i ST ST F is he inverse cumulaive probabiliy of Objecive Funcion The objecive of he aircraf acquisiion decision model is o maximize he expeced operaional profi of airlines for which he profi could be derived by geing he subracion of he oal operaing cos from he oal revenue obained by airlines. For airlines, he oal revenue comes from he operaional income (i.e. sales of air ickes) and he sales of aging aircraf. Conversely, he oal operaing cos comprises of operaional cos, aircraf purchase (acquisiion) cos, mainenance cos, depreciaion expenses and payable deposi for new aircraf o be purchased. In general, he oal revenue of operaing P period, TR( I ), can be expressed as follows: n m P S S 1 TR I E fare D sold resale for 1,..., T; S s,..., s (3.15) iy iy k i1 y1 76

100 The firs erm of he righ hand side of Equaion (3.15) indicaes he expeced income obained from he sales of fligh ickes by considering he level of sochasic demand D for which F 1 S D S 1. The second erm indicaes he revenue obained by selling aging aircraf. On he oher hand, he P oal operaing cos of operaing period, TC( I ) can be expressed as follows: n n n m P S S P S i P P cos, i i i iy iy TC I E D u purc x hgf D A I dep n i1 i1 i1 i1 y1 dp x for 1,..., T; S s,..., s P i i 1 k (3.16) The erms of he righ hand side of Equaion (3.16) respecively indicae he expeced operaing cos, seup cos of aircraf acquisiion, aircraf purchase cos, mainenance cos, oal depreciaion expenses, and oal payable deposi for n ypes of aircraf Probable Phenomena in Flee Planning Airlines encouner many challenging unexpeced evens, for insance he occurrence of naural disaser, economic downurn and oubreak of diseases which are unpredicable in naure. In accordance o he occurrence of unexpeced evens (risks), an efficien risk managemen is necessary. According o Malaysia Airlines (Malaysia Airlines, 2010a), risk managemen process produces a risk map and likelihood scale for airline's managemen o prioriize he acion plans in miigaing possible risks. This highlighs ha 77

101 differen acion may be required o solve differen issues and a paricular issue may be handled differenly a differen imes. This signifies ha he flee supply o mee sochasic demand which is relaively influenced by he risks (unexpeced evens) could be oulined similarly, i.e. in erms of he likelihood scale. As such, he probable phenomena, s,..., s for a oal of k phenomena, 1 k are defined o describe he possible scenario of aircraf possession in meeing sochasic demand under uncerainy. The probabiliy of probable phenomena, p,..., p s1 sk quanifies he likelihood (probabiliy) of aircraf possession o mee sochasic demand. In oher words, hey define how well he respecive aspec of flee supply of airline in meeing demand. Preferably, he quaniy of operaing aircraf should be available adequaely o mee a desired level of service. If probable phenomena and is probabiliy are no defined, i means ha airlines only deals wih one possible scenario o mee sochasic demand, i.e. hey have perfec confidence ha a cerain level of sochasic demand will be me perfecly for a paricular operaing period during he planning horizon. However, his should no be he case as here is no perfec assurance of he fuure. As such, his indicaor is necessary o ake ino consideraion he respecive key aspec in making flee planning decision under uncerainy. The number of probable phenomenon varies depending on he percepion and consideraion of airlines in decision-making. Generally, wo probable phenomena (key aspecs) are considered for wo major aspecs, i.e. operaional and economic aspecs. The operaional aspec refers o he relevan 78

102 perspecives such as operaing roues ha could be flown wih a paricular aircraf and raffic righs. The economic aspec may cover he cash balance and deb/lease financing of airlines. These are he key consideraions of airlines in flee planning (AirAsia Berhad, 2004; Malaysia Airlines, 2010a). In fac, flee planning model is a muli-crieria decision-making problem in which several key aspecs (such as operaional and economic aspecs) have o be considered. The erm phenomenon is used o represen various siuaions ha occur owing o he impacs of hese differen aspecs on he flee planning model. Differen risk consideraion of he managemen would lead o differen possible scenarios of aircraf possession because he associaed flee planning oucome would be differen. For example, if he airline perceives ha local fligh is less risky (as i is operaed a home counry), his would cause he airline o purchase more aircraf wih smaller capaciy and offer higher service frequency. However, if i is an esablished airline wih good record of long-haul fligh, i would consider buying larger aircraf wih higher capaciy and adjus is frequency. As such, differen risk consideraion would cerainly conribue o differen scenarios. In he firs case (local fligh), he airline will own more aircraf wih smaller capaciy while he second case (long-haul fligh), he airline will have bigger size aircraf. Probabilisic approach can be adoped o quanify he risk as he oucome is no deerminisic a he poin of planning. The airline would amend 79

103 heir sraegy by considering various risk aspecs. As such, i is reasonably for he airline managemen o possess varying possible scenarios of aircraf possession (wih differen flee composiion) by considering all key aspecs (phenomena) in place. Noably, he consideraion of all relevan phenomenon (key aspec) in flee planning consiues he formaion of 'probable phenomena', i.e. 'probable phenomena' is ermed o reflec all key aspecs (including operaional and economic aspecs) which could impac opimal flee planning decision (noe: he erm of 'probable phenomena' is no used o imply he risk consideraion of he airline managemen ha lead o he possible scenarios of aircraf possession). In view of he probable phenomena is grealy driven by he risk consideraions which associae closely wih he operaing aircraf of he airline (as explained in he abovemenioned example), he resulan probable phenomena may vary (wih differen impacs) across a variey of aircraf ype. In oher words, i is likely for he probable phenomena o vary in accordance o he exising aircraf composiion of he airline, i.e. he exising flee supply (aircraf composiion) of he airline is an influenial inpu which consiues he formaion of he probable phenomena. Airline would consider exising flee composiion during he flee planning. This is o mainain he flee homogeneiy o ease he raining of pilo and fligh aendan, as well as he raining of aircraf mainenance engineer and echnician. In addiion, he faciliy (such as mock aircraf and mainenance garage) has o be mainained as well. As such, an airline which is owning a flee of Airbus is less likely o 80

104 purchase or lease Boeing s aircraf, unless he airline has he plan o expand is personnel (pilo and engineer eam) and infrasrucure. This is subjeced o he airline sraegy upon risk consideraion. Accordingly, he probable phenomenon is defined o capure he occurrence of various possible scenarios under he risk consideraion. Pracically, he probable phenomena and is corresponding probabiliy could be esimaed based on he decision policy of airlines, qualiaive judgmen of expers or consulans, pas performance of airlines and ravelers' response. Generally, he decision policy of airlines refers o he complian business sraegies and corporae framework which have been praciced closely by he decision makers (i.e. airline's managemen). Decision policy is playing he role o assure ha he managerial and operaional decision-making is praciced under he documened rules, in line wih he mission and vision of airlines. As a rule, airlines have o obain governmenal approvals o operae heir business neworks. As such, hey have o ensure ha heir flee operaions (which are driven by he business srucure (framework) and sraegies) are always in compliance wih he obained approvals while accomplishing business goals (mission and vision) in achieving a desired service level. On he oher hand, he consulancy of he expers refers o he advisory of a group of expers/panels owards he performance of airlines which may range from he financial managemen o he operaional pracices as he key 81

105 consideraions for decision-making. The consulancy of he expers could be obained via conrac-basis or permanen employmen. Besides, a quesionnaire survey sudy could be underaken appropriaely o obain he professional opinions/judgmens of he expers in he relevan field. For insance, when airlines plan o expand heir operaing neworks, a comprehensive analysis is cerainly needed paricularly o analyze he poenial of expansion as well as he possible difficulies or risks for expanding. Thus, he employmen of a group of expers or consulans is necessary for a professional, complee and deailed analysis on he business planning o achieve he level of service saisfacorily. The pas performance of airlines includes boh he demand and supply aspecs of airlines for which he aspec of demand primarily focuses on he saisical daa or operaing records of he number of ravelers as well as he ravel rend which is associaed closely o he flee planning of airlines. From he perspecive of supply, he performance and capabiliy of he flee in servicing he operaing neworks has o be aken ino consideraion in analyzing he pas achievemen of airlines. The adequacy of flee in meeing he ravel demand of airlines is paricularly imporan o achieve a desired service level. Boh of hese aspecs (demand and supply) have o be considered due o heir grea effec on each oher. 82

106 Travelers' response refers o he reacion of he ravelers which may change from ime o ime owards he services of airlines. This componen has o be considered for he reason ha i reveals he behaviors and expecaions of he ravelers owards he provided services. By having his componen in place, airlines can cerainly capure he needs of heir ravelers in a beer manner for he enhancemen of heir services. Traveler's response can be obained by conducing mode choice modeling analysis via ravel survey. For insance, some airlines had underaken regular ravel surveys in order o monior heir services as well as o idenify he area of improvemen (AirAsia Berhad, 2010a) Problem Formulaion The exisence of indeerminacy (sochasiciy) and he variabiliy of unexpeced evens o occur inroduce a probabilisic elemen. The absence of deerminism implies ha fuure evens are unpredicable. Aircraf acquisiion decision model is probabilisic as ravel demand is sochasic (no deerminisic) due o he occurrence of unexpeced even which is unpredicable in he real pracice. The characerisic of flee planning problem is ha some elemens are random, including he level of ravel demand which is uncerain ha giving rise o he elemen of sochasic demand and his resuls in a probabilisic issue. As such, a probabilisic dynamic programming model is adoped o solve he flee planning problem. Wih he aim o maximize airline's 83

107 expeced profi by acquiring new aircraf o mee sochasic demand under uncerainy, he formulaion of aircraf acquisiion decision model can be phrased as follows: For 1,2,, T P I P 1 n m n s1 s1 s1 s1 P E( fare ) D soldiyresaleiy E cos D ui purci ( xi ) i1 y1 i1 p s... 1 n n m n S i P P P hgf ( D, A ) Iiy depiy dpi ( xi ) i1 i1 y1 i1 n m n sk sk sk P E fare D soldiyresaleiy E cos D ui purci ( xi ) i1 y1 i1 P p s P k n n m n 1 I S i P P P hgf ( D, A ) Iiy depiy dpi ( xi ) i1 i1 y1 i1 (3.17) max X 1 r sk ( ) subjec o consrains (3.5)-(3.10), (3.12) and (3.14) for which S D, X, I, SOLD, O, R 0 Z. The erm, 1 1 r is needed o obain discouned value across he period of ime while k indicaes he k-h probable phenomenon for owning P I as iniial flee size. I is imporan o noe ha flee planning model is solved by assuming ha he developed model would subsequenly lead o sraegic operaional decision of airlines (e.g. fligh rouing and scheduling). 84

108 3.3.5 Soluion Mehod The developed aircraf acquisiion decision model, in he form of probabilisic dynamic programming model, can be solved by decomposing i ino a chain of simpler sub-problems. By using working backward mechanism, he soluion mehod commences by solving he sub-problem a he las period of planning horizon, T. The opimal soluions found for he saes a curren sage leads o he problem solving a he period of T 1, T 2,..., 1. This procedure coninues unil all sub-problems have been solved opimally so ha he decision policy o acquire new aircraf can be deermined profiably. For he developed opimizaion model (3.17), he ype of soluion mehod (i.e. linear programming problem or non-linear programming problem) can be idenified clearly wih a careful inspecion paricularly from he key componens as follows: funcion of number of flighs, S n f D, A funcion of mainenance cos, S n hgf D, A pracical consrains (3.5)-(3.10), (3.12) and (3.14) In general, he developed model could be convered equivalenly eiher o he linear programming model or nonlinear programming model based on he naure of lineariy. For model (3.17), modeling parameers would appear o be discree or coninuous variables while he objecive funcion and pracical consrains could be a linear or nonlinear funcion. If hey are in he form of 85

109 linear funcion in erms of decision variables, hen model (3.17) can be solved as a linear programming model, or else i is solved as a nonlinear programming model. In realiy, he lineariy of hese componens could be confirmed based on he operaional daa of airlines. I shall hen be validaed by using a regression es wih he aid of some mahemaical sofware. For he illusraive case sudy as shown in he following secion, linear relaionship was adoped for he above menioned componens and hence i is solved as a linear programming model An Illusraive Case Sudy: Linear Programming Model An illusraive case sudy is shown o examine he developed model. To make an opimal aircraf acquisiion decision, assume ha here are wo ypes of aircraf where n 1 for A and n 2 for A Airlines need o decide when and which ype of aircraf should be purchased over he planning horizon, i.e. eigh years. To avoid choosing some unrealisic value for modeling parameers and funcions, realisic daa and relevan informaion are compiled accordingly from published repors and accessible websies of airlines. Tables 3.2 and 3.3 show he daa inpu of he model. From Airbus published saemen (Airbus, 2010a, 2010b), i is obained ha he capaciy of aircraf A and A is 180 (wih a oal size of 1300 m 2 ) and 295 (wih a oal size of 3900 m 2 ) respecively. The expeced fligh fare and cos as shown in Table 3.2 is generaed based on he available financial repors of Malaysia Airlines 86

110 (MAS) (Malaysia Airlines, 2010a). In addiion, he purchase prices of aircraf as shown in Table 3.3 were obained from he published daa of Airbus (Airbus, 2010c). Wih he purchase price of aircraf and esimaed useful life of aircraf, i.e. five years, he depreciaion value of aircraf are calculaed accordingly by using he sum of he years digis approach. The resale prices and depreciaion values of aircraf as shown in Table 3.3 are obained based on assumed residual value, i.e. salvage cos of aircraf, which is 10% of aircraf purchase cos. Table 3.2: The Expeced Value of Fligh Fare and Cos per Passenger Operaing period, E fare, $ s 1 s 2 E fare, $ cos s 1 E, $ cos s 2 E, $ Table 3.3: Aircraf Resale Price, Depreciaion Value and Purchase Price ($ millions) y resale 51y resale 52 y dep 51y dep 52 y purc 51 purc Average There are many variables and parameers in aircraf acquisiion decision model. Since no all real daa can be obained, i is ineresing o invesigae how he resuls vary if he values of variables and parameers are changed. The daa inpu for benchmark scenario is lised as follows: Two probable phenomenon are considered, where k 2 87

111 P P A = 1, iniial flee size is I11 50 and I12 50 P P A =1, iniial flee size o be wo years old is I112 I122 2 The probabiliy of aircraf possession is p 0.5 and s1 p 0.5 s2 The budge, MAXbudge () $6,500,000,000 Area of parking space, PARK 500, 000m 2 Order delivery, ORDER 25 Discoun rae, r 5% per annum Confidence level of demand consrain, 1 95% Significance level of lead ime consrain, 5% Significance level of selling ime consrain, 5% Salvage cos of aircraf = 10% of aircraf purchase cos s2 s1 D 0.95D (3.18) The funcion of number of flighs is n n 2 f A A [R 0.81%] (3.19) The funcion of mainenance cos is n n 2 hgf A A [R 0.85%] (3.20) The funcion of number of aircraf is NA x10 NP [R 0.82%] (3.21) where NP is he number of passengers. Equaion (3.19) indicaes ha 483 flighs are operaed pracically for each addiional aircraf. The consan in his equaion has no pracical 88

112 inerpreaion. Equaion (3.20) denoes ha $705 is he esimaed increase of mainenance cos for each addiional aircraf and $81031 is he overall mainenance cos wihou considering addiional aircraf. These funcions signify ha respecive funcion is srongly affeced by he quaniy of aircraf owned, n A. Equaion (3.21) implies ha each addiional 500,000 passengers require one addiional aircraf (or one passenger requires aircraf). By using backward working mechanism, model (3.17) is simplified o model (3.22)-(3.30) when T 8. PI ( ) P 8 1 s P 8 P D8 sold815 sold825 x81 x82 7 P 7 P 6 P 7 P A8 1.44x10 I81 4.1x10 I82 8x10 x x10 x P 8 P D8 sold815 sold825 x x10 x82 7 P 7 P 6 P 7 P A8 1.44x10 I81 4.1x10 I82 8x10 x x10 x x x10 8x x10 p s1 max X s x x10 8x10 subjec o p s2 (3.22) 80x 228x 6500 (3.23) P P I I x x (3.24) P P P P s2 D, D (3.25) s I 13x 39I 39x 5000 (3.26) P P P P sold P I, sold825 I82 (3.27) P x P P 81 x82 25 (3.28) DLT81 30, DLT82 30 (3.29) DST81 30, DST82 30 (3.30) where S D, X, I, SOLD, O, R Z 0. Equaion (3.23) akes he budge 89

113 consrain of $6500 million. The oal demand simulaed for 8 follows normal disribuion, i.e ~ 9x10,1x10 D N. Wih a 95% confidence level, i is found ha oal aircraf owned a his period mus be greaer han 93, i.e. A8 93, which is indicaed in Equaion (3.24). Equaion (3.25) indicaes ha wih verified normal disribuion, acual level of demand for = 8 is prediced o be a leas a a confidence level of 95%, which is derived by Equaions (3.6)-(3.7). Equaion (3.26) is parking consrain; Equaion (3.27) is he sales of aircraf consrain, which is derived wih he assumpion ha aircraf a he age which is equal o or greaer han five years old are considered P P o be sold, hus: sold815 I, and 714 sold825 I. Equaion (3.28) indicaes order 724 delivery consrain. Wih assumed normal disribuion of RLT N and 8 n ~ 1.918, RST8 n ~ N 1.918,0.3613, Equaions (3.29) and (3.30) represen lead ime and selling ime consrains respecively for which he desired period o order new aircraf as well as he period o release aging aircraf for sales is a leas 30 monhs (i.e. 2.5 years 3 years) in advanced. The objecive funcion and pracical consrains are boh linear funcions in erms of decision variables and hence he developed model (3.22) is solved as a linear programming model. Ieraively, he procedure is repeaed o formulae he opimizaion model for operaing period, 7,6,5,4,3,2,1. Anoher six scenarios (wih variaions o some of he modeling parameers used in he benchmark scenario) are developed o invesigae he impac of he changes on he resuls. The following liss he developed 90

114 scenarios and he values of parameers used for sensiiviy analysis. Scenario A and B has confidence level of 90% and 99% respecively Scenario C and D has he probabiliy of aircraf possession a 0.6:0.4 and 0.4:0.6 respecively Scenario E and F has order delivery consrain, ORDER 20 and ORDER 30 respecively Resuls and Discussions The resuls of he benchmark scenario are shown in Table 3.4. Table 3.4 shows a consisen increasing rend of discouned annual profi excep he period for which here s a decrease in sochasic demand or when a paymen is charged for he acquisiion deposi and purchase cos of new aircraf. This shows ha he developed model is able o capure he demand uncerainy in a fairly beer manner. In addiion, he findings provide an insighful view for airlines in making an opimal aircraf acquisiion decision o accoun for he demand flucuaion. For Scenario A and B, he resuls show ha he confidence level has an impac on airline s oal demand and profi level. The confidence level indicaes he level of service argeed by an airline and hence airline s profi is affeced if he argeed level of service changes. The resuls of Scenario A and 91

115 B esablished he fac ha a higher profi is gained when he value of confidence level is on he rise. Apar from his, he resuls show ha here is a endency for airline o acquire more aircraf o mee he increase in demand, ye subjec o pracical consrains as elaboraed earlier. In overall, he resuls show ha airlines have o se heir arge properly in order o maximize heir operaional profi. Table 3.4: The Resuls of Benchmark Scenario (Aircraf Acquisiion Decision Model) Operaing period, Annual profi (millions) $1,752 $1,317 $1,433 $1,068 $264 $1,773 $659 $2,861 Quaniy of aircraf A o be ordered A Quaniy of aircraf A o be received A Iniial quaniy A of aircraf A Quaniy of aircraf o A be released for sales A Quaniy of aircraf A o be sold A Toal demand (million) From he resuls of Scenario C and D, i could be observed ha he profi level of airline has a smaller effec when he probabiliy of aircraf possession changes. Conrary o Scenario D, he expeced profi generaed by Scenario C is higher as i is oulined a a higher probabiliy of s 1, i.e. ps which is 20% higher han p s 1 of Scenario D. Similarly, he profi gained by Scenario C is higher han benchmark scenario hroughou he planning horizon. This shows ha a higher value of p s 1 which corresponds o a higher level of demand subsequenly resuls in a higher reurn. Therefore, he developed model is sensiive o he seing of aircraf possession of airline (probable phenomena) o mee sochasic demand. 92

116 The resuls of Scenario E and F show ha he order delivery consrain could affec he opimal decision of aircraf acquisiion. This happens mainly due o he consideraion (or decision) of airlines in purchasing he leas quaniy of aircraf as long as he oal quaniy of aircraf owned is sufficien o provide he argeed service level. Hence, i is imporan o noe ha i s no cerainly profiable o acquire more aircraf as a higher aircraf purchase cos and mainenance cos will occur. In oher words, o purchase less aircraf probably conribues o higher expeced profi (due o he less charged coss). In a nushell, i could be seen ha he seing up of modeling parameers in he developed model could affec opimal resuls, o some exen. Comparaively, he resuls are more sensiive o he confidence level compared o oher parameers. Besides, he findings revealed ha here is no ideal means o obain a supreme profi as opimal acquisiion decision is decidedly dependen on several facors, i.e. managemen policy of airlines, he desired scenarios o be opimized and also he occurrence of unpredicable unexpeced evens. Therefore, in order o improve he decision making in flee planning, hose aspecs as menioned and illusraed earlier should be aken ino consideraion favourably. 93

117 3.3.8 Summary An opimal aircraf acquisiion decision model is formulaed wih he aim o maximize airline's profi. To do his, a mahemaical opimizaion model is developed by using probabilisic dynamic programming approach in order o capure sochasic demand which is assumed o be normally disribued. The proposed model and soluion mehod is esed wih an illusraive case sudy, in which mos of he inpu daa and funcions are eiher obained or simulaed by using airline's real daa. The developed model is solved opimally o deermine airline s decision for he quaniy and ype of new aircraf ha should be purchased during he planning horizon. I is observed ha he compuaional oupus are sensiive o he values of he modeling parameers, a varying degrees, and he resuls indicaed ha he proposed mehodology is viable. Wih reasonable assumpions ha perain closely o realisic pracice, he resuls revealed ha aircraf acquisiion decision is srongly influenced by sochasic demand as well as he policy of airlines (for insance, he predeermined age of aircraf o be sold). Generally, he profi of airlines is increasing when he level of demand is on he rise excep for an unexpeced drop in demand, which could ake place unpredicably in real pracice or when acquisiion deposi and purchase cos are charged for new aircraf. In addiion, six scenarios are creaed o es he sensiiviy of he parameers seing o he oucome. Remarkably, he order delivery consrain has a lile impac for he 94

118 aircraf acquisiion decision. Noneheless, he aircraf acquisiion decision is comparaively influenced by he confidence level and he probabiliy of aircraf possession. I is shown ha he resulan findings are able o seer he relevan auhoriies a managemen level as well as he decision makers in making a wise and profiable flee planning decision. 3.4 Aircraf Acquisiion and Leasing Decision Model To opimize he flee planning decision of airlines (via aircraf acquisiion and leasing), he aircraf acquisiion and leasing decision model is formulaed as a probabilisic dynamic programming model. Specifically for a se of origin-desinaion (OD) pairs, assume ha here is a selecion of n ypes of aircraf ha could be purchased or leased. The decision variables of he model are he quaniy and ype of aircraf o be purchased or leased in order o maximize he operaional profi of airlines. The sochasic demand modeled from secion 3.2 is used as one of he inpus o he developed model. The opimal decision, i.e. he alernaive a each sage is aircraf acquisiion and leasing decision o mee sochasic demand while making decision o sell he aging aircraf. For a paricular operaing period, alhough he sae variables and he corresponding opimal soluions could be obained, he opimal decision for he nex operaing period is unknown due o uncerainy. In fac, he saes of he nex operaing period are uncerain given he curren decision because many facors may no be known wih cerainy in pracice (Taha, 2003; 95

119 Winson, 2004). In oher words, he developed model is probabilisic as is inpus (sochasic demand) and oupus (opimal soluions) are subjec o possible occurrence of unexpeced evens, which is sochasic (probabilisic) in naure Consrains The pracical consrains considered for aircraf acquisiion and leasing decision model are oulined as follows: Budge consrain Budge consrain ascerains wheher or no he soluion is financially feasible for airlines. For his consrain, he sum of aircraf purchase and lease cos should no be more han airline's allocaed budge. This consrain could be expressed as follows: n n P L i i i i budge () i1 i1 purc x lease x MAX for 1,..., T (3.31) Demand consrain To ensure ha sochasic demand could be me desirably a a argeed level of service, he demand consrain could be expressed as follows: n i1 1 S i S SEAT f D, A 1 D for 1,..., T; S s,..., s (3.32) i k for which he level of demand could be derived by using he 5-sep modeling 96

120 framework while 1 is he confidence level (argeed service level) o mee sochasic demand. Equaion (3.32) assures ha he service frequency provided by airlines (wih available number of aircraf seas) would be sufficien adequaely o mee ravel demand saisfacorily. Parking consrain When an aircraf is no in operaion, i has o be parked a he hangar or a he apron of he airpor. In such a case, he choice of aircraf would someimes be consrained by he geomery layou of he hangar or he apron of he airpor. As such, parking consrain is ough o be considered feasibly. This consrain could be oulined as follows: n m i1 y0 P L P L Iiy Iiy xi xi sizei PARK for 1,..., T (3.33) Sales of aircraf consrain For some airlines, aging aircraf which is less cos-effecive migh be sold a he beginning of a cerain operaing period when airlines make he decision o purchase new aircraf. However, he quaniy of aircraf sold should no be more han he aircraf owned by airlines. This consrain can be expressed as follows: sold I for 1,..., T; i 1,..., n; y 1,..., m (3.34) P iy ( 1) i( y1) Order delivery consrain The delivery of new aircraf depends on he producion and he supply of aircraf manufacurers. Someimes, here migh be an availabiliy issue in delivering new aircraf. As such, aircraf o be 97

121 purchased should no be more han he number of aircraf available in he marke. This consrain could be formed as follows: P x ORDER for 1,..., T; i 1,..., n (3.35) i For aircraf leasing, i is assumed ha order delivery consrain is no relevan due o is possible availabiliy wihin one year (shor-erm duraion) for some circumsances. In addiion, he quaniy of leased aircraf is relaively flexible a cerain exen (no really limied o manufacuring consrain). Lead ime consrain In pracice, airlines would ge an agreeable lead ime (he period beween placing and receiving an order) from aircraf manufacurer when hey place an order for new aircraf. This consrain should be considered as i indicaes when airlines are supposed o order new aircraf. For n ypes of aircraf, his consrain can be expressed as follows: for 1 ; 1 P RLT DLT,..., T i,..., n (3.36) i i In real life, here are chances ha he argeed lead ime would change (say, due o he echnical problems of he manufacurer), hus lead ime should be a random value ha could be represened by a cerain disribuion. In his research, he lead ime is assumed o be normally disribued wih mean and sandard deviaion LT. The consrain could be saed as follows: where 1 1 LT DLT F 1 1 for 1,...,T ; i 1,..., n (3.37) i LT LT F is he inverse cumulaive probabiliy of 1. 98

122 Selling ime consrain Aging aircraf which is considered as less economical and effecive a some exen migh be sold by airlines a a cerain operaing period. In such a case, airlines need o know he mos suiable ime o release heir aging aircraf for sales paricularly o look for prospecive buyers in advance. In real pracice, he real selling ime migh be longer han he desired selling ime. Therefore, his consrain is formed wih he aim o reduce he possibiliy of his inciden as leas as possible. This consrain could be defined as follows: for 1 ; 1 P RST DST,...,T i,..., n (3.38) i i I is assumed ha selling ime of aircraf has a normal disribuion wih mean and sandard deviaion : ST ST where 1 1 DST F 1 1 for 1,...,T ; i 1,..., n (3.39) i ST ST F implies he inverse cumulaive probabiliy of Objecive Funcion The objecive of aircraf acquisiion and leasing model is o maximize he operaional profi of airlines in deermining he quaniy and ype of aircraf ha should be purchased or leased o mee sochasic demand. The operaional profi of airlines could be derived by considering he subracion of he oal operaing cos from he oal revenue. For an airline, he oal revenue is generaed from he operaional income (i.e. he sales of fligh ickes) and he sales of aging aircraf while he oal operaing cos is formed by aircraf 99

123 operaional cos, purchase/lease cos, mainenance cos, depreciaion expenses, payable deposi of aircraf acquisiion and leasing, and fuel expenses. P L For operaing period, he oal revenue, TR( I I ) of airline can be expressed as follows: n m P L S S 1 TR I I E fare D sold resale for 1,..., T; S s,..., s iy iy k i1 y1 (3.40) The firs erm on he righ-hand side of Equaion (3.40) indicaes he expeced income from he sale of fligh ickes by considering sochasic demand, S D while he second erm signifies he revenue from he sales of aging aircraf. P L The oal operaing cos, TC( I I ) of operaing period can be formed as follows: n n n n m P L S S P L S i P P cos i i i i i, iy iy TC I I E D u purc x lease x hgf D A I dep n m n n n L L P L Iiy depiy dpi xi dli xi C fueli for 1,..., T; S s,..., 1 sk i1 i1 i1 i1 y1 i1 y1 i1 i1 i1 (3.41) The erms on he righ-hand side of Equaion (3.41) denoe he expeced operaional cos of airline, aircraf purchase cos, lease cos, mainenance cos, depreciaion expenses, payable deposi of aircraf acquisiion and leasing, and fuel expenses, respecively. 100

124 3.4.3 Problem Formulaion In summary, he aircraf acquisiion and leasing decision model can be presened mahemaically as follows: For = 1, 2,..., T P L max 1 P I I r X n m s1 s1 s1 s 1 E( fare ) D soldiyresaleiy E cos D i1 y1 n n n P L S i ui purci ( xi ) leasei xi hgf ( D, A ) i 1 i 1 i 1 p s... 1 n m n n P P L L P Iiy depiy Ii depi dpi xi i1 y1 i1 i1 n n L dli xi C fueli i1 i1 n m sk sk sk s k E( fare ) D soldiyresaleiy E cos D i1 y1 n n n P L S i ui purci ( xi ) leasei xi hgf ( D, A ) i1 i1 i1 P p s P k n m n n 1 I I P P L L P Iiy depiy Ii depi dpi xi i1 y1 i1 i1 n n L dli xi C fueli i1 i1 L (3.42) subjec o (3.31)-(3.35), (3.37) and (3.39) where S D, P X, L X, P I, L I, SOLD, O, R Z 0. The erm r 1 is used for discouned value across he planning horizon while k indicaes he k-h probable phenomenon for having P I and L I as iniial flee supply a he beginning of each operaing period. The opimal decision (oupu) of he developed model, i.e. he opimal quaniy of aircraf o be purchased and leased, could be used as he inpus in opimizing oher operaional decisions of airlines, such as opimizaion of flee rouing, fligh scheduling and crew assignmen (Barnhar e al., 2003). 101

125 3.4.4 Lower Bound and Opimal Soluions The soluion of decision variable in model (3.42) is found o be influenced by demand consrain (Equaion (3.32)). In case he change of demand is non-posiive (i.e. no incremen of demand), he lower bound of he soluion is 0. This is because he decision variable defined is nonnegaive, i.e. P L x, x 0 and oal of n ypes of aircraf o be purchased and leased is also i i P L nonnegaive, i.e. x x 0 n i1 i i for a paricular operaing period. In case if he change in demand is posiive (i.e. demand increases), he lower bound is governed by demand consrain. I is o ensure ha he supply of aircraf (via acquisiion or leasing) mus mee he level of demand a a cerain desired service level. Neverheless, he upper bound (UB ), i.e. he maximum of aircraf ha could be purchased (or leased) is subjec o aircraf availabiliy in he marke, ORDER, which is expressed in order delivery consrain (Equaion (3.35)). To summarize, he lower bound, LB, of he developed model follows he following equaion: LB P L S X 0, X 0 if D 0 S i S P S SEAT f D, A 1 D x ORDER if D 0 i i n i1 (3.43) where S D indicaes he change of demand from year o year, i.e.. S S S D D D 1 102

126 P L P L Le X X : LB X, X UB i, be he se of decision variable i i i for aircraf acquisiion and leasing decision model and he operaional profi (i.e. he objecive funcion o be maximized) of he developed model be P where P L I I I I P L P could be wrien as I L I. The opimal soluion of he developed model P where P is he opimum (maximum) profi of each operaing period for which P I and L I denoe he corresponding oal quaniy of aircraf (including he aircraf o be purchased and leased) ha P L maximizes I I P. As such, he opimal soluion (i.e. maximum operaional profi) of he developed model could be wrien as follows: P L P L * P I * I * Max P I I I (3.44) Soluion Mehod The developed model in he form of probabilisic dynamic programming model can be solved by decomposing i ino a series of simpler sub-problems. By using he backward working mehod, he sub-problem a he las period of he planning horizon, T is solved firs. The opimal soluion found for he saes a he curren sage leads o he problem solving a he period of T 1, T 2,..., 1. This procedure coninues unil all sub-problems have been solved opimally so ha he decision policy o purchase and/or lease aircraf can be deermined sraegically. For he developed opimizaion model, he ype of soluion mehod, i.e. linear programming model or non-linear 103

127 programming model can be idenified clearly based on he funcion of number S n S n of flighs, f D, A ; funcion of raveled mileage, D A gf, ; funcion of S n mainenance cos, hgf D, A ; funcion of fuel expenses, C fuel n and pracical consrains (3.31)-(3.35), (3.37) and (3.39). If hey are in he form of linear funcion in erms of he decision variables, hen model (3.42) can be solved as a linear programming model. Oherwise, i is solved as a non-linear programming model. The lineariy of hese componens is primarily based on he operaional daa of a paricular airline. I shall hen be validaed by using he regression es wih he aid of mahemaical sofware. For he illusraive case sudy as shown in he following secion, non-linear relaionship was adoped for he above-menioned componens as he regression relaionship obained from he published repors (Malaysia Airlines, 2010a; Air Asia Berhad, 2010a) show non-lineariy. Powell (2007) specified ha non-linear programming is one of he possible soluions for he dynamic programming model. Noneheless, i could no be solved direcly wih any available convenional mehods. The spreadshee funcionaliy of Excel 2007 coupled wih own developed algorihm was uilized o compue he opimal soluions. For a larger size of aircraf acquisiion and leasing decision model, he soluion mehod is sill feasible in generaing compuaional resuls. However, he compuaional efficiency reduces when he problem size ges larger due o addiional modeling parameers and variables. As such, more compuaional effor is necessary for he larger sae and sage spaces. Two major concerns ha could affec compuaional efficiency are airline s planning horizon and he 104

128 ype of aircraf. The exension of planning horizon, T would resul in an incremen raio of T 1, i.e. an addiional of 10-20% of compuaional effors for each incremen (in year). For each addiional ype of aircraf, here is ORDER 1 imes more compuaional ime required where ORDER refers o order delivery consrain. I is esimaed ha he compuaional ime required o generae an opimal soluion is abou seconds for each operaing period An Illusraive Case Sudy: Nonlinear Programming Model To examine he applicabiliy of he developed model, mos of he daa inpu are chosen based on publicly published repors and accessible websies of airlines in order o design a close o realiy case sudy Inpus for Sochasic Demand Modeling There are hree ypes of unexpeced evens, i.e. biological disaser (e.g. flu disease), economic recession and naural disaser (e.g. sorm) which are assumed o affec he demand level. The modeling of probabiliy disribuions o quanify hese unexpeced evens is carried ou based on some published repors. According o he daa obained from he Cenre for Research on he 105

129 Epidemiology of Disasers (2010), he occurrence of biological disaser was found o follow Poisson disribuion and has a mean, of 7, i.e. Prob bio- disaser ~ Pois 7. This indicaes ha he biological disaser happens 7 imes in average in a year. Based on he daa from Inernaional Moneary Fund (Global Recession, 2010; Franke and John, 2011), i was found ha he occurrence of economic downurn also follows Poisson disribuion in which Probecon - downurn ~ Pois 9 1. This shows ha he economic recession happens once in an average of 9 years. For a naural disaser o occur (Cenre for Research on he Epidemiology of Disasers, 2010; Franke and John, 2011), he probabiliy of occurrence has a normal disribuion, i.e. Prob naural disaser ~ N 64,8. For he ravel growh projecion, he hisorical daa shows ha he growh percenage ranges from 5% o 9% (Inernaional Civil Aviaion Organizaion, 2008; Malaysia Airpors Holding Berhad, 2008; Malaysia Airlines, 2010a; Inernaional Air Transpor Associaion, 2011). As such, an equal probabiliy for each uni of growh is assumed, i.e. he percenage growh of 5%, 6%, 7%, 8% and 9% has he probabiliy of 0.2 o happen hroughou he planning horizon. However, here is no resricion if uneven probabiliy is assumed. Besides, he forecased demand, D is esimaed o follow a normal f 7 12 disribuion, i.e. f ~ x10,9.04x10 D N according o he daa obained from he published repors from Malaysia Airlines (2010a). Wih he 106

130 aid of he convoluion algorihm, he projeced demand of base year, D is 0 hen deermined accordingly. Based on he above-menioned daa, he level of sochasic demand for each operaing period hroughou he planning horizon is obained by applying he 5-sep modeling framework of sochasic demand (as discussed in secion 3.2). The deailed oupu of sochasic demand is shown in Table 3.5. Table 3.5 reveals he fac ha he possible occurrence of unexpeced even and he prediced growh of ravel demand could affec he level of sochasic demand a varying degrees. Basically, he SDI value is greaer han 1 when unexpeced even does no exis. Conversely, he exisence of unexpeced even produces he SDI wih he value of a mos Inpus for Aircraf Acquisiion and Leasing Decision Model Two ypes of aircraf i.e. A ( n 1) and A ( n 2 ) are considered for a se of OD pairs. Only wo ypes of aircraf are considered as many low-cos carriers operae heir business wih few varieies of aircraf ypes, for example: AirAsia (A , A ), Jesar Airways (A , A , A ), JAL Express (B , B ) and Tiger Airways (A ) (more examples could be seen in O Connell and William (2005)). Furhermore, airlines end o operae aircraf from he same aircraf 107

131 manufacurer (mosly Airbus or Boeing). Therefore, he wo ypes of aircraf (boh Airbus) considered in he case sudy are pracical. A and A were chosen as he example as here is more available informaion for hese ypes of aircraf. However, he developed mehodology is no resriced o he quaniy and ype of aircraf used. In addiion, a planning horizon of eigh years is also jusified according o Malaysia Airlines (2010a) and AirAsia Berhad (2010a), on average, he acquisiion of new aircraf requires a period of five years o be compleely delivered. Besides, he desired lead ime is assumed o have a normal disribuion wih an average of hree years, and sandard deviaion of 1.5, i.e. DLT ~ N(3,1.5). As such, wo ypes of aircraf which are considered for a planning horizon of eigh years is reasonably pracical o reflec airline s real operaions. Tables 3.2 and 3.6 shows he inpu daa used in he model. Table 3.5: The Oupu of Sochasic Demand Operaing period Occurrence of unexpeced even (Y= i exiss, N=i does no exis) N N Y (biological disaser) N Y (biological disaser, recession) N N N Probabiliy of unexpeced evens, [1] Probabiliy of he possible incremen of forecased demand, [2] Toal of probabiliy, [1]+[2] Sochasic demand index, SDI Sochasic demand (number of ravelers in millions) The capaciy of A and A is assumed o be 180 (wih a oal size of m ) and 295 (wih a oal size of m ), respecively (Airbus, 2010a, 2010b). As menioned earlier, he expeced fligh fare and cos 108

132 as shown in Table 3.2 are generaed based on he available financial repors of Malaysia Airlines (Malaysia Airlines, 2010a). In addiion, aircraf purchase cos as shown in Table 3.6 are obained from he published daa of Airbus (Airbus, 2010c). Wih aircraf purchase price and he esimaed useful life of aircraf (i.e. five years), he depreciaion values of aircraf are calculaed by using sraigh-line depreciaion approach. By considering he residual value of AirAsia (AirAsia Berhad, 2010b), he aircraf resale price and depreciaion value (as shown in Table 3.6) are obained based on he assumed residual value (i.e. salvage cos) of aircraf, which is 10% of aircraf purchase cos. For aircraf leasing, he respecive lease cos, residual value and depreciaion value are obained based on he finance lease of MAS (Malaysia Airlines, 2010b). Table 3.6: Aircraf Resale Price, Depreciaion Value, Purchase Cos, Lease Cos and Residual Value ($ millions) y Average resale y resale 2y P dep y P 2 y dep P 1 purc P 2 purc L dep ny lease n Residual value A benchmark scenario is creaed o examine he applicabiliy of he developed mehodology. The daa inpu can be caegorized ino hree caegories, i.e. by definiion, by assumpion or by assumpion based on he real daa. They are shown as follows: 109

133 By definiion: Two probable phenomenon are considered, where k 2 Discoun rae, r 5% Significance level of demand consrain, 5% Significance level of lead ime consrain, 5% Significance level of selling ime consrain, 5% and 1 D D D D (3.45) s1 s2 s1 By assumpion: A 1, he probabiliy of aircraf possession is p 5 s 1 0. and p 5 s 2 0. P P A 1, iniial quaniy of aircraf o be hree years old is I113 I123 4 Seup cos, u 0 i By assumpion (based on real daa): P P A 1, iniial quaniy of aircraf is I I 50 and L L I I Allocaed budge, MAX $6,500,000, 000 budge ( ) Area of hangar, PARK 500,000 m 2 Order delivery consrain, ORDER 5 Salvage cos of aircraf = 10% of aircraf purchase cos Deposi of aircraf acquisiion, DP 10% of aircraf purchase cos Deposi of aircraf leasing, DL 10% of aircraf lease cos The funcion of number of flighs is n 2 2 n 4 2 f A 9.776x10 A 7.83x10 [R 0.97] (3.46) 110

134 The funcion of raveled mileage is g 2 2,066 f 2,875,383 [R 0.83] (3.47) The funcion of mainenance cos is h x x10 g [R 0.94] (3.48) The funcion of fuel expenses is C fuel 7.46 f 8.3x10 f 98,572 [R 0.88] (3.49) n The number of aircraf is NA NP 73.6 [R 0.92] (3.50) where NP is he number of ravelers. Based on he daa as repored by Malaysia Airlines (2010a) and AirAsia Berhad (20101b), Equaions (3.46)-(3.50) are obained by conducing polynomial regression analysis (Meyer and Krueger, 2005). Equaions (3.46)- (3.50) are anicipaed o be correlaed wih sochasic demand, S D and oal operaed aircraf, n A. The regression analysis shows ha Equaions (3.46)- (3.48) are fied fairly well as non-linear funcions in erms of n A. Similarly, he analysis reveals ha Equaion (3.49) is bes fied as a quadraic funcion in erms of he number of flighs, which could be consequenly, expressed as a non-linear funcion in erms of n A via Equaion (3.46). Besides, regression analysis exhibis ha Equaion (3.50) is bes fied as a linear funcion in erms of number of ravelers. Equaion (3.45) implies he proporion of sochasic demand, which corresponds o he phenomenon of s 1 and s 2. Equaion (3.46) indicaes ha he 111

135 number of flighs is affeced by oal operaed aircraf, which is gained from aircraf acquisiion and leasing. Equaion (3.47) denoes ha a fligh flies 2,066 kilomeres in average. Equaion (3.48) signifies ha a uni cos of is charged as mainenance cos for each addiional uni of mileage raveled. For his equaion, $5,177 indicaes an overall esimaed mainenance cos wihou considering an addiional raveled mileage. Equaion (3.49) shows ha oal of fuel expenses depends on number of flighs, which are operaed during he planning horizon. This implies ha fuel expenses associae closely wih oal operaed aircraf, n A which is grealy depending on aircraf acquisiion and leasing decision. Equaion (3.50) displays ha every addiion of 100,000 ravelers requires one addiional aircraf. In oher words, one raveler requires aircraf. According o Meyer and Krueger (2005), he inercep of regression equaion carries no pracical meaning if he range of independen variable does no include 0. The number of flighs, f, in Equaion (3.46) falls wihin he range of 67,460 f 79, 927 (based on he real daa). Accordingly, he consan in Equaion (3.46) has no pracical inerpreaion. In addiion, i can be shown ha he raveled mileage in Equaion (3.47) is always posiive. Such explanaion is also applicable o Equaions (3.49)-(3.50). For T 8, he developed opimizaion model could be simplified o model (3.51)-(3.59) as follows: 112

136 P( I I ) P L 8 8 max X p s1 p s2 s P 8 P 118D8 8.2x10 sold x10 sold x10 x x10 x82 7 L L x10 x81 x x x10 g 7 P 7 P 6 L L 1.476x10 I x10 I x10 I81 I82 6 P 7 P 7 L L x10 x x10 x x10 x81 x f 8.3x10 f 98,572 s P 8 P D8 8.2x10 sold x10 sold x10 x x10 x82 7 L L x10 x81 x x x10 g 7 P 7 P 6 L L 1.476x10 I x10 I x10 I81 I82 6 P 7 P 7 L L x10 x x10 x x10 x81 x f 8.3x10 f 98,572 (3.51) subjec o P P L L x 213x 26.7 x x 6,500 (3.52) A n A 11,321 0 (3.53) n 8 8 D 16, 744, 756, D 15,907,518 (3.54) s1 s2 8 8 P L P L P L P L I81 I81 x81 x81 I82 I82 x82 x82 1,282 3, ,000 (3.55) sold I, sold I (3.56) P P x 5, x 5 (3.57) P P DLT 32, DLT 32 (3.58) DST 24, DST 24 (3.59) where S P L P L D, X, X, I, I, SOLD, O, R Z n P P 0 for which A I I I I x x x x. Equaion (3.52) akes he budge consrain of L L P P L L $ 6.5x10 o purchase and/or o lease aircraf. By applying he simulaion approach as elaboraed earlier, he sochasic demand simulaed for 8 is 16,744,756. Wih a 95% confidence level, i is found ha he oal number of 113

137 aircraf ha should be operaed for his operaing period appears o be a nonlinear funcion, which is indicaed in Equaion (3.53). Equaion (3.54) indicaes ha he sochasic demand of = 8 is prediced o be 16,744,756 for he probable phenomenon of s 1 and 15,907,518 for he probable phenomenon of s 2, which is derived by Equaion (3.45). Equaion (3.55) is parking consrain as a geomery limiaion; Equaion (3.56) is sales of aircraf consrain, which is derived wih he assumpion ha an aircraf of five years old or more are P P P P considered o be sold, hus: sold I and sold I. Since I I and I P P I, hese expressions subsequenly resul in sold I and sold I P P as could be seen in Equaion (3.56). Equaion (3.57) indicaes order delivery consrain o purchase new aircraf. Wih he assumed normal disribuion of RLT n ~ 2, 0.4 and ~ 1.5, N RST n 8 N, Equaion (3.58) and Equaion (3.59) represen he lead ime and selling ime consrains respecively, for which he desired period o order new aircraf is a leas 32 monhs (i.e years 3 years) while he desired period o release aging aircraf for sales is a leas 24 monhs, i.e. 2 years in advance. For model (3.51), he funcions of number of flighs, raveled mileage, mainenance cos, and fuel expenses as depiced by Equaions (3.46)-(3.49) are found o be non-linear funcions in erms of oal operaed aircraf, n A. Hence, he developed model (3.51) is solved as a non-linear programming model. By using working backwards mechanism, he procedure can be repeaed o formulae he opimizaion model for he operaing period, 7, 6, 5, 4, 3, 2,

138 In order o invesigae he impac of changes of he inpus o he compuaional resuls, six scenarios wih variaions o some of he modeling parameers used in he benchmark scenario are developed. The following liss he oulined scenarios. Scenarios A and B have confidence level of 90% and 99%, respecively. Scenarios C and D have probable phenomena indicaor (i.e. probabiliy of aircraf possession) of 0.6:0.4 and 0.4:0.6, respecively. Scenarios E and F have order delivery consrain, ORDER 4 and ORDER 6, respecively Resuls and Discussions The compuaional resuls of benchmark scenario are shown in Table 3.7. Table 3.7 shows a consisenly increasing rend on he discouned annual profi of airline excep where here is a decrease in sochasic demand or when a cos is charged o purchase new aircraf, lease aircraf, or order new aircraf in advance. In paricular, he operaing period from 1 o 3, which involves aircraf leasing and higher demand, produce a higher operaional profi compared o subsequen operaing periods. For he operaing period wih aircraf acquisiion, i.e. operaing period from 4 o 8, he profi of airline increases gradually, mainly due o an incremen in sochasic demand. This shows ha he developed mehodology is capable of capuring demand uncerainy in real pracice in producing an opimal profi. Cerainly, his would 115

139 provide a beer insigh for airlines in making profiable decision o manage heir flee supply under sochasic demand. Table 3.7: The Resuls of Benchmark Scenario (Aircraf Acquisiion and Leasing Decision Model) Operaing period, Iniial quaniy of A aircraf owned A Iniial quaniy of A leased aircraf A Quaniy of aircraf A o be ordered A Quaniy of aircraf A o be received A Quaniy of aircraf A o be leased A Quaniy of aircraf o be A released for sales A Quaniy of aircraf A o be sold A Toal operaed aircraf Sochasic demand (million) Discouned annual profi ( $ millions) The graphical resuls of Scenarios A-F are illusraed in Figures The resuls of Scenarios A and B (in Figure 3.2) indicae ha when he confidence level changes, here is an impac on he operaional profi. The confidence level signifies he service level (i.e. level of demand) argeed by airlines, and hence airline's profi is affeced if he argeed service level changes. Apar from his, he resuls of Scenarios A and B esablished he fac ha a higher profi is gained when he value of confidence level increases i.e. when he level of service rises. The resuls also show ha here is a endency for airlines o purchase and/or o lease more aircraf o mee a higher level of 116

140 demand bu subjec o operaional consrains. In paricular, for operaing periods 5, 6 and 8, he operaional profi of he benchmark scenario is higher han Scenario B due o aircraf acquisiion decision o mee a higher level of demand. Overall, he findings show ha airlines have o make he flee planning decision wisely as well as o se heir arge properly in order o maximize operaional profi. Figure 3.2: The Resuls of Scenarios A and B Figure 3.3 shows he resuls in seing he probabiliy of probable phenomena (aircraf possession) for which Scenario C has he probabiliy of 0.6:0.4, Scenario D has he probabiliy of 0.4:0.6 and he benchmark scenario is 0.5:0.5. The resuls show ha Scenario C which has he highes probabiliy in meeing demand (i.e. highes level of service) could yield he highes operaional profi, which is in average 21% more han Scenario D and 11% more compared o he benchmark scenario. Comparaively, he benchmark 117

141 scenario generaes 12% more profi han Scenario D. As such, i is approximaed ha an incremen of 1% of sochasic demand would generae an addiional 1% of operaional profi. This could be explained by he fac ha he service level which is me a a higher chance (probabiliy) is likely o generae more revenue for airlines (from he sales of fligh ickes). Hence, i could be seen ha he probable phenomena and is probabiliy which associaes closely wih he level of sochasic demand could grealy affec he operaional profi of airlines. Figure 3.3: The Resuls of Scenarios C and D As displayed in Figure 3.4, he resuls of Scenarios E and F show ha he order delivery consrain could affec he opimal decision and operaional profi of airlines. The resuls illusrae ha he higher he value of order delivery consrain is, he lower would he profi be. For he operaing periods 118

142 1, 4, 5 and 6, Scenario E produces he lowes profi due o he aircraf acquisiion deposi and cos ha are incurred for aircraf acquisiion decisionmaking. Besides, he decision-making o purchase and/or o lease aircraf is also affeced by he consideraion of airlines in geing he leas number of aircraf as long as he oal quaniy of aircraf is adequae o provide he argeed level of service. Hence, i is imporan o noe ha i s no cerainly profiable o purchase or lease more aircraf. The decision o purchase (or lease) lesser aircraf probably conribues a higher profi level due o less charged coss. Figure 3.4: The Resuls of Scenarios E and F The consisency and sabiliy of resuls could be empirically confirmed by comparing he findings wih he acual operaional saisics of airlines (AirAsia Berhad, 2010a; Malaysia Airlines, 2010a). Table 3.8 summarizes he flee size of airlines (i.e. AirAsia and MAS) as compiled from heir annual 119

143 repors as well as flee planning decision of each operaing period as obained from he developed model. I could be observed ha, he flee size of AirAsia and MAS during he operaing years of 2006 o 2010 falls wihin he range of wo sandard deviaions from is average. The flee planning soluions obained from he benchmark problem and oher scenarios exhibi similar paern, i.e. he flee size for he operaing periods from 1 o 8 falls wihin he range of wo sandard deviaions from is average. Therefore, he soluions are coheren wih he operaing performance of airlines. As such, he findings are consisen wih he acual pracice and hence he sabiliy of he resuls (as well as he developed model) could be empirically confirmed. Concisely, i could be seen ha he resuls obained from he developed model are reasonable and sable when compared empirically wih airline's daa. The sensiiviy analysis shows ha he developed model and is soluions are sensiive o he modeling parameers. This implies ha he values of hese parameers need o be chosen wih care. In addiion, i is imporan o noe ha here is no ideal means o obain a supreme profi as opimal flee planning decision is affeced decisively by several facors, i.e. managemen policy of airlines (for insance, as repored by MAS (Malaysia Airlines, 2010c), a 100%- leased srucure is no opimal in he long-erm, MAS inends o shif o an opimal mix of leased/owned flee), he desired scenarios o be opimized and he occurrence of unpredicable unexpeced even. Therefore, in order o assure an opimal profi in flee planning, he aspecs as discussed earlier should be aken ino consideraion wisely. 120

144 Table 3.8: The Summary of Flee Planning Decision (Aircraf Acquisiion and Leasing Decision Model) Operaing Year Average (AG) Sandard Deviaion (SD) Empirical (from repors) Flee size Model Scenario Year AirAsia MAS Benchmark A B C D E F AG + 2SD AG 2SD Summary A new mehodology is developed o solve he flee planning decision model under uncerainy. To do his, a 5-sep modeling framework which is incorporaed wih he Sochasic Demand Index (SDI) is developed o quanify he demand level under uncerainy for each operaing period. To solve he flee planning problem, a probabilisic dynamic programming model is formulaed o deermine he opimal quaniy and ype of aircraf o be purchased and/or leased so ha he sochasic demand could be me profiably hroughou he planning horizon. Besides, a probable phenomena indicaor is defined necessarily o ensure ha he aircraf possession of airlines is sufficien adequaely o mee sochasic demand a a desired service level. The resuls obained from he illusraive case sudy demonsraed ha he developed 121

145 mehodology is well responsive o modeling parameers and i is viable in providing opimal soluion for flee planning decision model. In overall, he developed approach reflecs he acual siuaion of airline indusry, ranging from he challenge of uncerainy o he pracical issues in acquiring and leasing aircraf. Subjec o he occurrence of unexpeced even and operaional consrains, he developed mehodology could produce viable soluions for long-erm aircraf acquisiion and leasing decision model. For airlines, his is crucial o ensure economy susainabiliy (by maximizing profi) as well as managemen efficiency from he operaional aspec (by ensuring adequae flee supply of aircraf) o mee sochasic demand saisfacorily. 122

146 CHAPTER 4 STRATEGIC FLEET PLANNING MODELING FRAMEWORK 4.1 Supply-demand Ineracion In Flee Planning Wih he aim o make a sraegic flee planning decision o assure an adequae flee supply in meeing sochasic demand, his chaper is organized sysemaically in wo major secions. The firs secion deals wih he analysis of raveler's mode choice as an imporan elemen in demand managemen. Specifically, he raveler's mode choice for differen rip purpose (leisure and business) is modeled and analyzed specifically based on he ype of rip (local and rans-border rips). The resulan mode choice analysis is hen incorporaed in he Analyic Hierarchy Process (AHP) modeling framework o quanify he probabiliy of probable phenomena in flee planning. Noably, probable phenomena are he key aspecs (deerminan) of aircraf possession in he flee planning decision-making and hence is probabiliy (likelihood) in opimizing he flee planning decision needs o be quanified properly. To do his, he subjecive judgmen of airline's managemen (decision makers of flee planning) is ackled explicily. The developed framework enables airlines o capure he supply-demand ineracion in greaer deail. A numerical example, oulined wih airlines' operaional daa, is demonsraed o examine he applicabiliy 123

147 of he developed framework in deermining he probabiliy of probable phenomena wih regard o hree key aspecs, namely operaional, economy and environmenal aspecs. The resulan probabiliies are hen applied o solve a realisic flee planning problem. In erms of managerial and operaional pracices, he developed mehodology, incorporaed wih mode choice analysis, is useful o assure an adequae flee supply o mee demand flucuaion. More imporanly, sochasic demand could be me saisfacorily wih opimal profi. 4.2 Mode Choice Analysis Since he pas few decades, he rapid economy growh in he developed as well as he developing counries has fosered he brisk developmen of he ranspor nework. As a resul, various ravel modes are now available convenienly in moving ravelers from an origin o a paricular desinaion. Basically, ravelers make heir choices based on heir perceived uiliy on each ravel mode. Air ranspor was once considered as he mos elegan and expensive ranspor mode. I is used mosly for long disance ravel (especially o overseas) while ground ranspor is commonly used for shor disance (local or rans-border) ravel. 124

148 I is imporan for ranspor operaors o undersand how differen ranspor modes compee wih each oher in mulimode ranspor nework. This is a crucial sep o manage ravel demand efficienly and o predic he fuure ravel rend precisely. To ranspor operaors, a proper undersanding of ravelers mode choice behavior and he underlying conribuing facors could help hem improve heir services coninuously. Despie is imporance, here are limied sudies peraining o invesigae how air ranspor compees wih various ranspor modes in he mulimode nework. So far, mos of he exising sudies focus on he compeiion among air ranspor and also beween he air ranspor and he high-speed rain. Oher ravel modes, such as buses, rains, or privae vehicles, are no sudied. Those are imporan ravel modes of ranspor especially in he developing counries. Besides, mos sudies argeed he developed counries. However, here are known differences in ravel behavior and choices among he ravelers of he developing counries in comparison o he developed counries (Khoo and Ong, 2011; Khoo e al., 2012). Accordingly, he compeiion beween air ranspor as well as ground ranspor (i.e. buses, rains, and privae vehicles) in he developing counry conex should be invesigaed. Saed preference surveys were conduced in he Klang Valley region of Malaysia and raveler s mode choice are modeled o invesigae he facors affecing mode choice decision for boh local and ransborder rips by rip purposes (leisure and business). Nine aribues, namely ravel ime, ravel cos, safey, comfor, service frequency, faciliy, on-ime performance, booking/purchase mehod, and promoional package are sudied. 125

149 A sensiiviy analysis on he ravel cos and comfor is carried ou o inspec how he demand of ranspor operaors (including airlines) would change if hese aribues vary. In addiion, i is anicipaed ha mode choice decision would be influenced by he background of ravelers Local and Trans-border Trip A local rip is defined as a rip generaed beween wo ciies in he same counry. In his research, he ciies chosen for local rip are Kuala Lumpur and Penang. Kuala Lumpur, siuaed in he cenral region, is he capial ciy of Malaysia. I is he major governance, financial, and business cenre in he counry. Penang is a beauiful ciy in he norhern region of Malaysia. I is siuaed abou 330km from Kuala Lumpur. The ciy is declared by UNESCO as he Heriage Ciy in 2008 (UNESCO, 2008). I is famous among he ouriss for is food and hisorical buildings. A rans-border rip is defined as a rip generaed beween wo ciies across nearby counry. I requires ravelers o show heir passpor (eiher wih or wihou visa) when crossing he counry s boundary. For his research, Singapore is chosen as he abroad ciy for ransborder rip. Singapore is siuaed abou 350km, souh of Kuala Lumpur. Figure 4.1 shows he geographical locaions of hese ciies. For he case sudy, hese ciies are chosen because here are mulimode available beween hese ciies. One can choose ground ranspor (such as rain, bus, or privae vehicles) or air ranspor (LCC or FSC) o ravel beween hese ciies. In addiion, here is 126

150 considerable demand beween hese origin-desinaion (OD) pair. For each of hese rips (local and rans-border), wo rip purposes, i.e. leisure and business rips, are considered explicily. Figure 4.1: The Locaion of Klang Valley, Penang and Singapore Saed Preference Survey In his secion, he seing of quesionnaire and saed preference survey are explained in deail. 127

151 Experimenal Design Five ypes of ravel mode, i.e. LCC (AirAsia), FSC (MAS), bus (singledecker and double-decker bus), rain, and privae car are considered o model raveler s mode choice. Nine aribues, i.e. ravel ime (includes access and egress ime), ravel cos (includes access and egress cos), safey, comfor, service frequency, faciliy, on-ime performance, booking/purchase mehod and promoional package, were considered in he experimenal design of he quesionnaire. Two levels (i.e. low and high level) are considered for each aribues. This conribues o a oal of 512 (i.e. 9 2 ) possible ses of choice. However, i is impracical o presen all he aribues combinaion o he respondens in real pracice. Therefore, a fracional facorial design is adoped by considering 16 ses of choice o be presened o he ravelers. I involves four basic design aribues (i.e. ravel ime, ravel cos, safey, and comfor) and five independen generaors (i.e. service frequency, faciliy, on-ime performance, booking/purchase mehod, and promoional package). The basic designs are seleced based on he findings of a pilo survey. Subsequenly, he choices are divided ino wo blocks, each wih eigh ses of choice, by uilizing blocking approach (confounding facorial design). 128

152 Traveling Aribues Tables 4.1 and 4.2 respecively illusrae he raveling aribues, and he levels used (i.e. low and high levels) in designing he saed preference survey for local and rans-border rips. The seing of raveling aribues is described as follows: Travel ime The considered ravel ime in he survey is he oal ravel ime which akes ino accoun he access ime and egress ime of ravel mode. For air ranspor, he check-in ime is also considered. The values se for ravel ime are eiher obained or esimaed from available resources. For example, he average check-in ime for LCC and FSC are obained from he websies of airlines (Malaysia Airlines, 2010a; AirAsia Berhad, 2010a). In-vehicle ravel ime of buses and cars are esimaed based on ravel speed. For air ranspor and rain services, expeced in-vehicle ravel ime is obained from accessible websies (Malaysia Airlines, 2010a; AirAsia Berhad, 2010a; Kereapi Tanah Melayu Berhad, 2010) while access and egress ime are assumed for all ravel modes. A minimum value (such as access or egress ime) is esimaed for low level while a maximum value is esimaed for high level. Travel cos The oal ravel cos is he sum of journey cos, access and egress cos. The journey cos of air ranspor (LCC and FSC), wheher by rain or bus, is obained from he respecive websies. The ravel cos of privae 129

153 vehicles akes ino accoun he perol price (an average of RM0.15 (USD0.05) per kilomere) and oll charges. The perol price would change depending on he ypes of perol and vehicle. Accordingly, he minimum cos is defined for he low level while he maximum cos is defined for he high level. Table 4.1: The Aribues of KL-Penang Trip (Local Trip) Travel mode MAS AirAsia Bus Privae car Train Travel ime (minue) 120, , , , , 505 Travel cos (RM) 110, , , , , 170 Comfor 0, 9 0, 6 0, 13 0, , 4 Safey 5, 0 1, 0 129, , 0 28, 0 Service frequency 1, 9 1, 6 1, 13 1, , 4 Faciliy 0, 9 0, 6 0, 13 0, , 4 On-ime performance 0, 9 0, 6 0, 13 0, , 4 Booking/purchase mehod 1, 3 1, 3 1, 2 0, 1 1, 2 Promoional package 0, 9 0, 6 0, 13 0, , 4 Table 4.2: The Aribues of KL-Singapore Trip (Trans-border Trip) Travel mode MAS AirAsia Bus Bus Privae Train (single-decker) (double-decker) car Travel ime (minue) 185, , , , , , 575 Travel cos (RM) 315, , , , , , 190 Comfor 0, 6 0, 10 0, 13 0, 13 0, , 4 Safey 5, 0 1, 0 129, 0 129, , 0 28, 0 Service frequency 1, 6 1, 10 1, 13 1, 13 1, , 4 Faciliy 0, 6 0, 10 0, 13 0, 13 0, , 4 On-ime performance 0, 6 0, 10 0, 13 0, 13 0, , 4 Booking/purchase 1, 3 1, 3 1, 2 1, 2 0, 1 1, 2 mehod Promoional package 0, 6 0, 10 0, 13 0, 13 0, , 4 Safey The record of accidens of air ranspor (LCC and FSC) is exraced from Aviaion Safey Nework of Fligh Safey Foundaion (Fligh Safey Foundaion, 2010). For bus and car, a high level indicaes a safe journey wih no acciden while a low level akes ino accoun he pas year s acciden records involving each mode (MIROS, 2010). Due o unreadily accessible daa for he records of rain accidens in he conex of Malaysia, he relevan values 130

154 (as shown in Tables 4.1 and 4.2) were compiled based on he lis of rain accidens of some developed and developing counries (Wikipedia, 2012). Service frequency For privae cars, service frequency of 1 is used as he low level while 1440 is oulined as he high level wih he assumpion ha privae cars are available all he ime (24 hours). For oher ravel modes, he respecive service frequency is obained from accessible websies. Booking/purchase mehod For he booking/purchase mehod, he value of low level (i.e. he value of 1) across he alernaives (ravel mode) implies ha here is (minimum) one purchase or booking mehod in geing a place o ravel wih a paricular mode while he value of high level indicaes he oal number of ways in purchasing/booking a place o ravel wih seleced ravel mode. There are hree mehods for air ranspor (LCC or FSC) ravelers o buy fligh icke, i.e. websie booking, booking via ravel agen, and during ravel fair. For buses, ravelers could purchase heir ickes from he couner or hrough ravel agen. For rains, ravelers could purchase his/her icke from he websie or a he saions. For privae cars, ravelers drive alone (value of 0 ) or share wih heir friends/relaives (value of 1 ). Comfor, faciliy, on-ime performance and promoional package Cusomarily, he comfor level of ravel mode associaes closely wih he provided services 131

155 (or faciliies) in using a paricular ravel mode. The low level of comfor, which implies unsaisfacory condiion, signifies ha a raveler feels ha he ravel wih a paricular mode is no up o his or her perceived (comfor/saisfacory) level. This may occur due o numerous facors, for insance he condiion of he sea (spacious or no), he air-condiioner sysem (workable or no), check-in and on-board faciliies (convenien or no), ec. Conversely, a high level of comfor implies ha a paricular raveler is saisfied wih he exising condiion of he provided faciliies/services. Similarly, he elemens of faciliy, on-ime performance and promoional package could be oulined accordingly o reflec he percepion of ravelers owards on-board equipmens (for faciliy), delay or rescheduling issue (for on-ime performance) and special discoun or offer of ickes (for promoional package). Due o unavailable real daa from ranspor operaors, he value of high level of comfor, faciliy, on-ime performance and promoional package are considered in accordance o he provided service frequency while he low level considers he value of 0 for an unsaisfacory condiion. For insance, 0, 6 for he promoional package of MAS (in Table 4.2) denoes ha here is no promoional package for he low level and six promoional packages for he high level. For he values of high level of on-ime performance in Table 4.2, he values of 10 for AirAsia and 13 for bus signify ha he deparure of AirAsia and bus are all on-ime (no delay/rescheduling) i.e. 10 imes for AirAsia and 13 imes for bus in accordance wih he frequency of service provided. 132

156 4.2.3 Quesionnaire and Respondens The quesionnaire is divided ino wo secions for which he firs secion examines he socioeconomic characerisics of respondens while he second secion requires he respondens o choose heir preferred mode choice for local and rans-border rips. A group of five well-rained surveyors were sen ou o conduc he quesionnaire survey from 11 s January o 17 h February The argeed respondens for his research are he residens in he Klang Valley region which is he surrounding area of Kuala Lumpur. All ravelers are considered o have he same origin i.e. Kuala Lumpur. A oal of 552 respondens (i.e. 273 for local rip and 279 for rans-border rip) were inerviewed. The disribuions of respondens are displayed in Table 4.3. Table 4.3: The Characerisics of Respondens Origin-desinaion KL-Penang KL-Singapore Origin-desinaion KL-Penang KL-Singapore Variable Caegory Percenage Percenage Variable Caegory Percenage Percenage Gender Male RM Female RM RM Malay Monhly RM RM Race Chinese income RM RM Indian RM Household Age size Number of working SPM/STPM aduls Cerificae/ diploma/ Educaional level advanced diploma Degree Number of cars Maser & above Ohers Governmen servan Execuive/ Daily Privae car Occupaion adminisraor Professionals ravel mode Public ranspor Sudens Reired 2 7 Ohers

157 4.2.4 Modeling Approach: Mulinomial Logi Models By using he daa colleced from he saed preference survey, several models were esed wih Limdep/Nlogi sofware o model raveler's mode choice. Mulinomial logi model was found o be he bes fied model o reflec he mode choice decision for boh local and rans-border rips. The principle of a logi model is ha an individual is rying o opimize (i.e. maximize) his or her uiliy by selecing an opion which is he mos beneficial for a raveling siuaion. The higher uiliy denoes ha i is more likely ha an alernaive will be chosen. Based on he maximum likelihood esimaion echnique, mulinomial logi models (discree choice models) capure he influence of aribues and characerisics on decision makers preferences (Train, 2003). For he mode choice modeling of an individual i, he regression equaions which are he uiliy funcions of all ineresed alernaives (ravel modes), could be modeled in he form as below: Uij Uiliy funcion of ravel mode j, Uij βx (4.1) ij where β is he vecor of he esimaed parameers (corresponding o each ineresed aribue of x ), x is he vecor of he ineresed aribues (including raveling aribues and socioeconomic characerisics of ravelers) and ij is he error erm (i.e. he erm which is assumed o be Gumbel disribued). The choice probabiliy, P of individual i is hen expressed as follows for a oal of ij J alernaives. 134

158 exp x j ij Pij, j 1, 2,...,J J exp x q1 q iq (4.2) Noe ha here are a few limiaions of he logi model known in lieraure, such as he assumpion of independence from irrelevan alernaives (IIA), he unobserved facors are unrelaed o he choices (McFadden and Train, 2000), and ase variaions vary sysemaically wih respec o he observed variable (Train, 1998, 2003). Specifically for he insance wih nine raveling aribues and six socioeconomic characerisics (for which x : ravel ime, 1 x : ravel cos, 2 x 3 : comfor, x : safey, 4 x 5 : service frequency, x 6 : faciliy, x 7 : on-ime performance, x 8 : booking/purchase mehod, x 9 : promoional package, x 10 : gender, x : race, 11 x : age, 12 x 13 : income, x : household size, x : number of cars), he vecors of β and x could be expressed as follows: 1 x1 ravel ime 2 x2 ravel cos β, x 14 x14 household size 15 x15 number of cars (4.3) Noe ha he uiliy funcion U ij is he dependen variable of he mode choice analysis. By considering all ineresed aribues (wih esimaed parameers), he uiliy funcion could hen be expressed by: U x x x x (4.4) ij ij 135

159 For he insance o consider five ravel modes, i.e. he alernaive j MAS, AirAsia, Bus, Car, Train which respecively refers o he ravel of a rip wih Malaysia Airlines (full service carrier), AirAsia (low-cos carrier), bus, car and rain, he uiliy funcion of each specific ravel mode can be expressed, in general, as follows: U Time x Cos x Size x NumCars x (4.5) imas MAS 1 MAS 2 MAS 14 MAS 15 imas U Time x Cos x Size x NumCars x (4.6) iairasia AirAsia 1 AirAsia 2 AirAsia 14 AirAsia 15 iairasia U Time x Cos x Size x NumCars x (4.7) ibus Bus 1 Bus 2 Bus 14 Bus 15 ibus U Time x Cos x Size x NumCars x (4.8) icar Car 1 Car 2 Car 14 Car 15 icar U Time x Cos x Size x NumCars x (4.9) itrain Train 1 Train 2 Train 14 Train 15 itrain Noably, he respecive uiliy funcion (wih significan esimaed parameers) could be obained accordingly by performing mode choice analysis appropriaely. In order o perform mode choice analysis, all ineresed variables (wih he respecive measuremen uni) are described in Table 4.4 and he relevan sources of he modeling variables are presened in Appendix B Findings: Mode Share of Trips The probabiliy of choosing ravel mode for local and rans-border rips is presened in Table 4.5. For local leisure rip, he resuls show ha air ranspor is preferred. Abou 49% and 24% of ravelers choose FSC and LCC, respecively. Privae vehicle is he mos popular ravel mode among ground 136

160 ranspor. Abou 13% of ravelers choose o use privae vehicle, followed by bus (9%) and rain (5%). Local business ravelers show similar choice wih leisure ravelers. Neverheless, fewer road users (23%) are found for business rip. This demonsraes ha air ranspor gains more ineres from business ravelers han leisure ravelers. Table 4.4: The Descripion of he Ineresed Variables Variable name Measuremen uni Descripion Dependen variable Uiliy funcion, U ij Traveling aribues Travel ime, x 1 Travel cos, x 2 Comfor, x 3 Numerical value (poin form) Hour and/or minue Ringgi Malaysia (RM) Number of imes (associaed wih service frequency) 137 An indicaor which implies how likely an alernaive (ravel mode) is chosen for raveling purposes. A higher value of uiliy indicaes ha i is more likely ha a specific ravel mode will be chosen. Toal ravel ime which akes ino accoun he in-vehicle ime, access ime and egress ime of a specific ravel mode. For air ranspor, he check-in ime is also considered. Toal ravel cos which akes ino accoun he sum of journey cos, access and egress cos. An indicaor which depics he saisfacory level of a raveler owards a specific ravel mode. Safey, x Number of accidens An indicaor which demonsraes how secure 4 (safe) a ravel mode is. Service frequency, x 5 Faciliy, x 6 On-ime performance, x 7 Booking/purchase mehod, x 8 Promoional package, x 9 Number of imes (associaed wih service frequency) Number of imes (associaed wih service frequency) Number of imes (associaed wih service frequency) Number of available ways Number of imes (associaed wih service frequency) An indicaor which shows how frequen a ravel mode is available. An indicaor which indicaes he adequacy of he supporing services (in erms of faciliies/equipmen) provided by he ranspor operaors. An indicaor which signifies he puncualiy of a ravel mode. An indicaor which reveals how a raveler purchases/orders a sea in order o ravel wih a paricular ravel mode. An indicaor which shows he possibiliy of a raveler o ravel wih a cheaper cos (e.g. purchase fligh icke during he promoional period). Socioeconomic characerisics Gender, x 10 Caegorical Sexual caegory (male or female). Race, x Caegorical Ehnic group (Malay, Chinese, Indian or 11 ohers). Age, x Year The lengh of living ime. 12 Income, x Ringgi Malaysia The earnings level (monhly) of a paricular 13 (RM) raveler. Household size, x Numerical value The number of family members. 14 Number of cars, x Numerical value The quaniy of privae cars possessed by a 15 raveler.

161 Trans-border leisure ravelers are found o have he highes endency (36%) in using air ranspor for which LCC (20%) is preferred han FSC (16%). Among he ground ranspor, privae vehicle is he mos likely opion wih he choice probabiliy of 32%, followed by single-decker bus (20%), double-decker bus (5%) and rain (4%). Besides, air ranspor dominaes he marke of rans-border business rip for which FSC (53%) is favored compared o LCC (15%). The privae vehicle again dominaes he ground ranspor wih is choice probabiliy of 15%. Bus and rain are no preferred by business ravelers. Table 4.5: The Choice Probabiliy of Local and Trans-border Trips (%) Local rip (KL-Penang) Trans-border rip (KL-Singapore) Travel mode Leisure rip Business rip Leisure rip Business rip FSC (MAS) LCC (AirAsia) Bus (single-decker) Bus (double-decker) Privae car Train Analysis of LCC's Impacs on Mode Choice Decision Tables 4.6 and 4.7 presen mulinomial logi models developed from he saed preference survey of local and rans-border rips, respecively. The models are saisically significan a 95% confidence level from he perspecives of socioeconomic facors as well as he raveling aribues of ranspor operaors. All models are examined wih he same socioeconomic backgrounds of ravelers and also on nine raveling aribues as described 138

162 earlier. The resuls show ha he models are saisically significan owards differen raveling facors. Based on hese models, he impacs of LCC on mode choice decision is invesigaed and presened in he following subsecions Impac of LCC on FSC The resuls show ha he major facor ha encourages rans-border leisure rip makers o choose LCC over FSC is low ravel cos. This is similar o he findings of O Connell and William (2006) and Mohd Suki (2014). However, for local leisure rip, ravelers prefer FSC more han LCC (as he uiliy consan of FSC is higher han LCC). This is because FSC could offer a beer comfor level compared o LCC. For rans-border business rip, he ravel cos remains as he major facor ha encourages ravelers o choose LCC. However, mos of he rans-border business ravelers op for FSC han LCC mos probably owing o he company s policies (normally larger firms) in aking FSC due o he frequen flyer iniiaives. This fac could be suppored by he occupaion of respondens for which more han 50% of he respondens (as shown in Table 4.3) are holding he posiion as governmen servan, execuive and professional, which are mos likely working in large-scaled companies which prefer FSC han LCC. This finding is in accordance wih he resuls of Evangelho e al. (2005) and O'Connell and Williams (2005, 2006). Besides, i was found ha comfor is he major facor for local business rip. The esimaed 139

163 parameer shows ha LCC is able o increase heir mode share if he comfor level could be improved (more significan han FSC since he value is higher). For rans-border leisure rip, a sensiiviy analysis is performed o invesigae he impac of he changes of ravel cos on he mode share. The resuls are displayed in Table 4.8. The resuls show ha if he ravel cos of LCC decreases by 10% o 50%, he choice probabiliy of LCC is esimaed o increase 2.71%-18.39%, wih an average of 10.12%. In addiion, his would arac an addiional of 2.22% of FSC ravelers o LCC. For local business rip, he resuls of sensiiviy analysis as illusraed in Table 4.9 signify ha if he comfor level of LCC improves gradually (from 10% o 50%), he marke share of LCC could increase a an average of 4.24%. Specifically, he improvemen in comfor means ha ranspor operaors aim o ensure ha heir ravelers would use and enjoy more faciliies/services. In such a case, ranspor operaors may provide some on-board faciliies so ha ravelers could feel more relax and comforable during heir ravel. For insance, elecric rain services (ETS) rain is equipped wih LCD TV paricularly for he informaive and relaxaion purposes. Accordingly, he higher level of improvemen of comfor level refers o more faciliies/services ha can be enjoyed by ravelers during heir ravel. Similarly, he reducion of ravel cos, he adjusmen (i.e. improvemen) of he comfor level a a higher level (i.e. from 10% o 50%) would increase he marke share of he LCC a a greaer exen (i.e. from 1.26% o 7.45%). Besides, i could be seen ha he improvemen of LCC's comfor level aracs an average of 3% of FSC ravelers. 140

164 Table 4.6: The Modeling Resuls of KL-Penang Trip (Local Trip) Leisure rip Business rip Parameer Coefficien -saisic Coefficien -saisic Consan MAS Consan AirAsia Consan Bus Gender MAS Gender AirAsia Gender Bus Gender Car Gender Train Race MAS Race Car Time Car Comfor MAS Comfor AirAsia Comfor Bus Safey Bus Log-likelihood funcion p-value Number of observaions 2,216 2,216 Table 4.7: The Modeling Resuls of KL-Singapore Trip (Trans-border Trip) Leisure rip Business rip Parameer Coefficien -saisic Coefficien -saisic Consan MAS Consan Single - dec ker bus Income MAS Income AirAsia Income Single - dec ker bus Income Double - dec ker bus Income Car Income Train Cos AirAsia Cos Double - dec ker bus Cos Train Safey AirAsia Safey Single - dec ker bus Log-likelihood funcion p-value Number of observaions 2,222 2,

165 Table 4.8: The Sensiiviy Analysis of KL-Singapore Trip (Trans-border Leisure Trip) Travel cos of AirAsia -10% -20% -30% -40% -50% Average MAS AirAsia Choice probabiliy (%) Single-decker bus Double-decker bus Car Train Table 4.9: The Sensiiviy Analysis of KL-Penang Trip (Local Business Trip) Comfor level of AirAsia +10% +20% +30% +40% +50% Average MAS Choice probabiliy (%) AirAsia Bus Car Train Impac of LCC on Ground Transpor For local business rip, he comfor of LCC is found o be significan. This reveals ha local business ravelers end o fly wih LCC, which is anicipaed o have a beer comfor level (probably due o shorer ravel ime) compared o he ground ranspor (i.e. privae car, bus, rain). For local rip (boh leisure and business purposes), he fac ha LCC is preferred is also suppored by he consan of LCC which has a higher value compared o he consan of bus. For rans-border rip (boh leisure and business purposes), ravel cos emerges as he major cause ha encourages ravelers o ravel wih LCC. In addiion o ravel cos, safey appears as he significan deerminan for 142

166 rans-border business ravelers in choosing LCC. LCC is preferred mos possibly owing o he leas acciden occurrence compared o ground ranspor (as shown in Table 4.2). Furhermore, here migh be a negaive percepion of ravelers owards he safey of group ranspor, especially car and bus due o increasingly road accidens from year o year (MIROS, 2012). This suppors he decision-making of business ravelers who have a concern for safey. The findings of sensiiviy analysis for rans-border leisure rip and local business rip (as displayed in Tables 4.8 and 4.9), respecively show ha he adjusmen of ravel cos and comfor level could increase he mode share of LCC. If he ravel cos of LCC decreases (10%-50%), he choice probabiliy of LCC is esimaed o increase gradually wih an average of 10.12%, by shifing abou 8% of he users of ground ranspor (i.e. 5% of car users, 2% of bus users and abou 1% of rain users). Besides, he findings show ha if he comfor level of LCC improves gradually (10%-50%), he marke share of LCC could increase a an average of 4.24% for which he improvemen of he comfor level a a higher level would increase he marke share of he LCC wih a greaer proporion. However, he impac of he improvemen of comfor level in mode shifing is lesser compared o he reducion of ravel cos, i.e. his sraegy (improvemen of comfor level) would only arac abou 1.15% of he users of ground ranspor. 143

167 Effec of Socioeconomic Background From he mode choice analysis, i was found ha one of he significan facors ha could influence ravelers mode choice is heir socioeconomic background. Table 4.6 shows ha gender is a significan facor ha affec leisure and business ravelers of local rips. The resuls show ha male ravelers prefer o ravel wih air ranspor and bus while car and rain are he likely opions among female ravelers. Besides, mode choice decision is also affeced by he race facor. Malay and Chinese ravelers show a high endency in raveling wih FSC and car. From Table 4.6, i could be seen ha respondens income level is anoher significan facor for leisure and business rips. I shows ha hose wih higher income end o ravel wih air ranspor or privae vehicles. On he oher hand, hose wih lower income show a endency o ravel wih bus, mos probably due o cheaper ravel cos Implicaions for Managerial Pracices For airlines, mode choice analysis is paricularly crucial from hree major managerial perspecives, i.e. demand, supply and susainabiliy as lised as follows: 144

168 From he aspec of demand: o manage ravel demand properly o predic fuure ravel rend precisely o develop and implemen appropriae markeing sraegy considerably o arac new ravelers (o increase mode share) effecively From he aspec of supply: for operaions and performance enhancemen (o improve exising services and sysems) for services planning (o mee ravelers' demand and expecaions) for flee planning decision-making (o suppor operaing neworks) From he aspec of susainabiliy: o reain he loyaly of ravelers o ouperform compeiors (under mulimode ransporaion sysem) o assure profiable operaions and marke shares Summary A proper undersanding of ravelers mode choice decision in mulimode ranspor nework is imporan for an efficien planning. I is a crucial sep o predic he usage of ranspor faciliies and o manage ravel demand effecively. While mos of he sudies focus on he analysis for he 145

169 developed counries, his research invesigaes ravelers mode choice decision in he developing counries. Besides sudying he choices beween LCC and FSC, his research also aims o deliberae he compeiion beween LCC and ground ranspor. A saed preference survey is carried ou o invesigae ravelers choice for local and rans-border rips. Mulinomial logi models are developed o idenify he underlying facors ha conribue o mode choice. I was found ha a few influenial facors ha could affec mode choice decision are rip purpose, raveling desinaion, ravel cos, comfor, safey, and ravelers background. I was also found ha LCC is only preferred by hose on rans-border leisure rip. For hose who choose LCC, low ravel cos is heir main concern. Oher ravelers choose FSC due o is excellen comfor level. Mos of he ravelers who choose o use ground ranspor prefer o ravel wih heir own privae vehicles for more flexibiliy and comfor. However, for hose who choose o use buses, ravel safey has become heir main concern. A sensiiviy analysis is carried ou o sudy he impac of changes of ravel cos and comfor level on LCC. The resuls show ha he LCC is able o arac a subsanial amoun of ravelers from FSC and ground ranspor if he icke price is reduced while he comfor level is increased. As such, i could be seen ha mode choice analysis could provide informaive highlighs o airlines for services enhancemen (including how o provide a sraegic flee planning as discussed in he following secion). Therefore, in order o capure he mode choice of ravelers properly, hose aspecs as discussed earlier should be handled wisely. 146

170 4.3 Analyic Hierarchy Process (AHP) Modeling Framework The Role of Analyic Hierarchy Process (AHP) In Flee Planning Decision-Making While providing an adequae flee supply, i is imporan o capure mode choice analysis (raveler s response) in view of he fac ha air ravelers (passengers) are he main users of airline's services which consiues he marke share and main income o airlines. Furhermore, raveler s behavior was changing wih exensive growh of mulimode ransporaion neworks. As such, how airlines make an opimal flee planning decision, i.e. a muliple crieria decision-making, hroughou he planning horizon is imporan o mee ravel demand profiably a a desired service level. Therefore, he flee planning decision-making of airlines which is, in fac, uncerain (primarily due o sochasic demand) and grealy governed by various key aspecs (muli-crieria) could be solved sraegically wih he aid of he AHP. By allowing he respecive judgmens o vary over he values of a fundamenal scale of 1-9, AHP possess he capabiliy o capure he fuzziness (uncerainy) in making a muli-crieria decision (Saay and Tran, 2007). Specifically, he judgmens made wih AHP, in he form of pair-wise comparison, by using judgmen scale 1-9 are fuzzy (uncerain). As such, he 147

171 flee planning decision-making of airlines which is, in fac, uncerain could be solved by making use AHP suiably. Comparaively, AHP is widely used due o is ease of use as well as is sraighforward scalable and undersandable manner han any oher muli-crieria decision making mehods (Velasquez and Heser, 2013). More imporanly, AHP which is able o capure he fuzziness and vagueness (uncerainy) explicily could reflec realisically he flee planning problem of airlines in a beer manner. To he bes of he auhors' knowledge, his research is he firs ha inegraes AHP and mode choice modeling in solving he flee planning problem explicily. The developed mehodology is capable o show how various key aspecs affec flee planning decision-making, i.e. how flee planning (a muli-crieria decision) is made o mee sochasic demand Modeling Framework The proposed modeling framework of AHP basically involves hree major sages as displayed in Figure 4.2. Sage 1 involves he judgmen and comparison among decisional crieria while Sage 2 focuses on he judgmen and comparison among he key aspecs for each decisional crieria. Finally, Sage 3 compues he end resul which is he probabiliy of he key aspec ha influence he flee planning decision-making. For Sage 1, he modeling framework commences by evaluaing he relaive comparison of n decisional crieria wih a comparison scale of 1-9 (Saay, 1980, 1994). Typically, he 148

172 relaive comparison is presened in he form of a marix, i.e. a square marix wih a size of n x n. In oher words, he marix is playing he role o reflec he subjecive judgmen of decision makers owards he relaive comparison of n decisional crieria. For n x n marix, he diagonal elemen which is he comparison of n decisional crieria agains iself is always equal o 1 while oher elemens in he marix signify he relaive comparison of n decisional crieria. For n x n marix, i is observable ha a ij 1 for which a ij defines a ji he elemen a row i and column j of he marix. This shows ha he elemen a and ij a are he reciprocal of each oher. As such, he ji known as reciprocal marix. n x n marix is also I is imporan o noe ha some degree of inconsisency is expeced due o he fac ha he decision-making is made based on he subjecive judgmen of decision makers. Therefore, he consisency of he marix needs o be examined accordingly. This can be done by conducing a consisency es. Mahemaically, he marix is said o be consisen if a x a a i, j, k. ij jk ik Generally, here are hree componens, namely consisency raio (CR), consisency index (CI) and random consisency index (RI) ha are required o carry ou he consisency es. Basically, CR evaluaes he raio of CI and RI of he marix in such a way ha CI measures he consisency of he marix by making use of he deviaion of he eigenvalue and marix size while RI is he average CI of a large sample of randomly generaed marices. The marix is said o be consisen if CR < 0.1. To examine he consisency, he larges 149

173 eigenvalue of he marix, also acs as a consisency indicaor for which he max marix is said o be more consisen if he value of marix size (Saay, 1990). is geing closer o max Figure 4.2: The Modeling Framework o Quanify he Probabiliy of Key Aspec A sage 2, a similar procedure (as in sage 1) is carried ou o form he judgmen marix ha reflecs he relaive comparison among he key aspec for each decisional crieria. As addressed earlier, he key aspec refers o a paricular perspecive (concern) ha could affec he flee planning decisionmaking. By validaing he consisency of marix, subsequenly he oupu of AHP approach is compued (a sage 3) o quanify he probabiliy of he respecive key aspec. The aspec wih a higher probabiliy is inerpreed o be more essenial han oher aspec wih a lower value. Noe ha he oal of probabiliy is one, i.e. 100% as he full decision of airlines. The decisionmaking in flee planning shall hen be driven by he resulan probabiliy of key 150

174 aspecs. The modeling framework of AHP (embedded wih mode choice modeling) as oulined in Figure 4.2 can be carried ou as follows: For sage 1: I. Deermine decisional crieria, C i To make flee planning decision, decisional crieria, C, i 1,..., n can be idenified appropriaely by idenifying he relevan elemen ha could affec he decision-making of airlines. Generally, he elemens ha are found o affec airlines are decision policy of airlines, consulancy of expers/consulans, pas performance of airlines and ravelers response in view of heir influenial impacs in flee planning (AirAsia Berhad, 2004; KPMG, 2007; Lessard, 2012; Malaysia Airlines, 2010a; Ryanair, 2012). i II. Esablish judgmen marix (for n decisional crieria) A pair-wise comparison marix, A involving n decisional crieria (as deermined from sep I) can be expressed as follows (Saay, 1980): 1 a a 12 1n 1 1 a2n a 12 A a a 1n 2n nxn (4.10) Generally, marix A is governed by 1 a >0, a 1, a for i, j. To assure ij ii ij a ji 151

175 consisency, noe ha a x a a i, j, k. Specifically, a ij jk ik ij implies he relaive comparison of crieria i over j based on judgmen scale 1-9 (for which 1:equal imporance, 3:weak imporance of one over he oher; 5:srong imporance; 7:demonsraed imporance; 9:absolue imporance while 2, 4, 6 and 8 signify he corresponding inermediae values beween wo adjacen judgmen (Saay, 1977, 1980, 1990)). In real pracice, decision makers are likely o make an inconsisen comparison (judgmen). To handle his issue, he following acions could be done (Saay and Tran, 2007): (1) Idenify he mos inconsisen judgmen in he marix and deermine he range of values for which he inconsisency could be improved. (2) Reques he decision maker o consider if he/she can aler his/her judgmen o a possible value in ha range. Oherwise, he decision is posponed unil a beer undersanding is obained. (3) Same procedure could be repeaed by examining he second mos inconsisen judgmen and so on. III. Calculae he larges eigenvalue As an indicaor for consisency, he larges eigenvalue, of marix A max can be deermined as follows (Saay, 1990): w n j max a (4.11) ij j1 w i 152

176 for which a is he elemen of marix A while ij w and i w respecively j represen he average of row i and j of marix A. Noe ha a marix is said o be more consisen if he value of he larges eigenvalue is geing closer o marix size. IV. Perform consisency es (for marix A) Consisency es is needed o assure he consisency of marix A (wih size n). This es can be conduced based on he consisency index, CI and random consisency index, RI which are oulined as follows (Saay, 1977): n 1.98 n 2 max CI, RI n1 n (4.12) Saay (1977) showed ha n if he marix does no include any max inconsisency. This implies ha he closer he value of max o n, he marix is more consisen. By using he measuremen of CI and RI, he consisency raio, CR can be evaluaed as follows: CI CR (4.13) RI As shown in Equaion (4.13), CR compares he consisency index, CI of he marix and a purely random marix, RI. The judgmen marix is said o be consisen if CR <

177 For sage 2: V. Esablish judgmen marix of key aspec (for each decisional crieria) For airlines, key aspec s k signifies he relevan aspec (concern) ha could affec flee planning decision-making. Specifically, he probabiliy of key aspec reflecs he likelihood or degree of each perspecive in making an opimal flee planning decision. As such, how o capure hese key aspecs for decision-making is vial. I is imporan o noe ha he number of key aspec o be capured may vary among airlines. For insance, some airlines claim ha he aspecs of operaional and economy are wo major deerminans in flee planning (AirAsia Berhad, 2010a; Malaysia Airlines, 2010a). In addiion, he environmenal aspec should be aken inconsideraion due o is increasing concern and impacs on airline s operaions. In such a case, hree key aspecs namely s, s 1 2 and s 3 could be defined accordingly o capure he aspec of operaional, economy and environmenal, respecively. For airlines, operaional aspec ( s 1 ) paricularly refers o relevan perspecives such as he abiliy o secure raffic righs and operaing difficulies of aircraf ype while he economy aspec ( s 2 ) covers he financial benefis of shareholders, economic benefis of new aircraf and so on. For he environmenal aspec ( s 3 ), he fuel efficiency of airlines is included in view of he fac ha lesser fuel consumpion produces fewer emission (Williams e al., 2002). In fac, hese aspecs (operaional, economy and environmenal) are closely relaed o one anoher owing o he fac ha aircraf operaions of airlines in supporing he operaing neworks would grealy affec no only he financial gains of airlines bu also heir green (environmenal) performance in erms of fuel efficiency, aircraf 154

178 emission and noise. By considering k key aspecs, i.e. s,..., s 1 k and decisional crieria C i, he pair-wise comparison marix of he key aspec (for each decisional crieria), B C i can be formed as follows: B C i 1 s s 12 1n 1 1 s2n s s s 1n 2n kxk (4.14) for which marix B is a square marix wih size k x k while s reflecs he Ci ij relaive comparison of key aspec s over s. i j In order o obain he pair-wise comparison marix of he key aspec s,..., s 1 k for he decisional crieria of ravelers response, ravel survey and mode choice analysis can be done as follows: Sep 1: Conduc ravel survey Travel survey can be underaken necessarily by airlines o examine he preference of ravelers as well as is impacs in flee planning. Specifically, ravel survey for differen rip purpose (e.g. leisure or business) can be carried ou for differen desinaion (e.g. local or rans-border). Traveler s response via survey could reveal heir preferences and ravel behavior in a beer manner. 155

179 Sep 2: Conduc mode choice analysis The response of ravelers via ravel surveys (inpu) could be uilized o generae he mode choice modeling models (oupu). The esimaed parameers of respecive rip (e.g. local leisure rip, local business rip, rans-border leisure rip and rans-border business rip) which consiue he mode share of respecive operaing nework can be obained accordingly from he mode choice modeling analysis. Sep 3: Evaluae he raio of key aspec o form judgmen marix, For he decisional crieria of ravelers' response, he pair-wise comparison marix of key aspec s,..., s 1 k can be evaluaed by aking ino accoun he mode share analysis (from sep 2). Subsequenly, he mode share of operaing nework which corresponds o he respecive key aspec can be evaluaed accordingly o obain he relaive comparison of key aspec. To do his, he B Ci relevan componens of each key aspec, he relaionship of he operaing nework, w F S k are aken ino accoun o assess Ne d and corresponding key aspec, s k. This is necessary o examine he impacs beween operaing nework and key aspec (o work ou he raio of key aspec). The deail framework o evaluae he raio of key aspec is shown in Figure 4.3. By considering W relevan componens of key aspec and also P operaing neworks, he respecive raio of key aspec (o form judgmen marix 156

180 B C i ) can be compued as follows: w Ne F d S s : s i, i, j 1,..., k; w 1,..., W; d 1,..., P i j w Ne F d S j (4.15) where Ne refers o operaing nework and d w F S k denoes he relevan componen of key aspec, s k. Figure 4.3: The Evaluaion of he Raio of Key Aspec VI. Perform consisency es (for marix B ) Ci The consisency of marix, BC i can be confirmed by adoping a similar procedure as described in sage 1, i.e. he marix is said o be consisen if CR

181 For sage 3: VII. Compue he probabiliy of key aspec The probabiliy of key aspec can be evaluaed as follows (Saay, 1980): Probabiliy i r * * AB (4.16) where A represens he average of row i 1, 2,..., n (i.e. decisional crieria) of * i normalized marix A while B denoes he average of row r 1, 2,..., k * r (i.e. key aspec wih regard o each decisional crieria) of normalized marix B. Ci Numerical Example This secion illusraes he applicabiliy of he developed framework (wih mode choice modeling) o quanify he probabiliy of he respecive key aspec in making an opimal flee planning decision. I. Deermine decisional crieria, C i Based on publicly accessible published repors (AirAsia Berhad, 2004; KPMG, 2007; Lessard, 2012; Malaysia Airlines, 2010a; Ryanair 2012), four decisional crieria, namely he decision policy of airline (DP), consulancy of expers (CE), pas performance of airline (PP) and ravelers response (TR) are idenified as he major key aspec ha could affec he flee planning decision of airlines. The elemens of DP, CE, PP and TR are denoed as crieria 158

182 C, C, C and C 4 respecively. Specifically, he decision policy of airlines (DP), C 1 refers o a paricular course of acion for airlines o make operaional and managerial decision (including flee planning decision). For insance, airlines may sandardize heir flee choice in erms of aircraf ype primarily due o financial and operaional concerns. For airlines, he consulancy of expers (CE), C 2 refers o advices or judgmens of consulans/panels owards airline's operaing performance, financial managemen as well as decisionmaking. In supporing he servicing neworks, he pas performance of airline (PP), C 3 includes he perspecives of demand and supply for which demand perspecive akes ino accoun he ravel rend while he perspecive of supply considers he flee performance in servicing available operaing neworks. From he perspecive of he users of air ranspor sysem, he ravelers response (TR), C 4 focuses on he mode choice modeling of ravelers which reveals he behavior and percepion of ravelers owards he services of airlines. II. Esablish judgmen marix (for n decisional crieria) Based on he decisional crieria of C, C, C and C 4 as idenified in Sep I, judgmen marix A can be formed as follows: A x 4 4 To obain marix A, accessible published informaion is compiled wih he aid of a simulaion approach for which he simulaed daa (in Table 4.10) 159

183 represens he relaive comparison of 10 managerial expers owards he decisional crieria C i over C. As shown in Table 4.10, he judgmens of j expers are compiled suiably as geomeric mean (Aczel and Saay, 1983). III. Calculae he larges eigenvalue The larges eigenvalue, of marix A can be deermined as follows: max max n j1 w j aij w i The value of he larges eigenvalue is coheren wih he fac of Saay (1977) for which n. max IV. Perform consisency es (for marix A) follows: Since n 4 (i.e. size of marix A), he elemen of CR is compued as 5 CI 4.1 x 10 CR 4.2 x 10 RI 0.99 Thus, he consisency is accepable because CR 0.1. This signifies ha he marix A is consisen (reliable) in erms of he judgmen (evaluaion) of decision makers

184 Table 4.10: The Evaluaion of Relaive Comparison Exper, k Geomeric mean a a a a a a V. Esablish judgmen marix of key aspec (for each decisional crieria) To form he marix of raveler s response (TR), he following procedure can be conduced: Sep 1: Conduc ravel survey Four saed preference ravel surveys had been underaken in 2011 (as discussed earlier) in order o model he mode choice decision of ravelers. These surveys aim o idenify and analyze he preference of ravelers owards he ground ranspor (bus, car and rain) and air ranspor (FSC and LCC). These surveys are denoed as y1, y2, y 3 and y 4 for local leisure rip, local business rip, rans-border leisure rip and rans-border business rip, respecively. Generally, he operaing neworks of airlines can be caegorized as shor-haul and medium/long-haul neworks. In such a case, local leisure rip ( y 1 ) and local business rip ( y 2 ) as domesic flighs are classified as shor-haul nework ( Ne 1 ) while rans-border leisure rip ( y 3 ) and rans-border business rip ( y 4 ) are included for medium/long-haul nework ( Ne 2 ). These neworks are hen uilized o evaluae he raio of key aspec, s k (as described furher in 161

185 Sep 3). Sep 2: Conduc mode choice modeling The response of ravelers via ravel surveys (inpu) are hen adoped o generae he mode choice modeling models (oupu). The resuls of mode choice analysis are summarized in Table Table 4.11: The Modeling Resuls of Travel Survey Nework Shor-haul nework, Ne 1 Medium/long-haul nework, Ne 2 Aribue Local leisure rip, y 1 Local business rip, y 2 Trans-border leisure rip, y 3 Trans-border business rip, y 4 Mode choice of rip Average mode choice Sep 3: Evaluae he raio of key aspec o form judgmen marix, As shown in Table 4.12, he mode share of respecive nework which correspond o he aspec of operaional ( s 1 ), economy ( s 2 ) and environmenal ( s 3 ) of airlines are evaluaed accordingly o compare wih he respecive key B TR aspec. To do his, he relevan componens of each key aspec, F are aken w Sk ino accoun o assess he relaionship of he operaing neworks, Ne d and corresponding aspec, s k. As menioned earlier, he relevan componens of operaional aspec ( s 1 ) of airlines may refer o several perspecives, including he number of passengers carried and he load facor in servicing operaing neworks. On he oher hand, he economy aspec of airlines ( s 2 ) may cover 162

186 numerous componens, including he operaing revenue and available capaciy (seas) while he environmenal aspec ( s 3 ) capures he fuel consumpion of airline in response o fuel efficiency of he operaing neworks. The daa of hese aspecs are compiled based on he operaing performance of Malaysia Airlines (Malaysia Airlines, 2010a). The evaluaion of he raio of respecive key aspec is shown in Table Operaing nework Table 4.12: The Evaluaion of he Raio of Key Aspec (For Traveler's Response) Average mode choice Key aspec s 1 (operaional) Load facor, F 1 S1 Passengers carried, F 2 S1 Revenue passenger kilomeers (RPK), F Key aspec s 2 (economy) 1 S2 Available seas kilomeers (ASK), F 2 S2 Key aspec s 3 (environmenal) Fuel efficiency, 1 F Shor-haul, Ne % 45% 10% 10% 68% Medium/longhaul, Ne % 55% 90% 90% 32% w Ne F d Sk Raio of key aspec s : s 1.15, s : s 1.86, s : s S3 By having he raio of key aspec, he judgmen marix of ravelers response, B TR could be formed as follows: B TR x3 As menioned earlier, here are hree more decisional crieria, i.e. decision policy of airline (DP), consulancy of expers (CE) and pas performance of airline (PP). The judgmen marices of hese decisional crieria are assumed o be as follows (due o he lack of accessible daa): 163

187 B 1 1 2, B 1 1 2, B DP CE PP x3 3x3 3x3 By carrying ou consisency es, he consisency of hese marices were confirmed because CR 0.1 for all marices. A he final sage (sage 3), he respecive probabiliy of key aspec is summarized in Table Table 4.13 shows ha he probabiliy of he key aspec are p , p and p for he aspecs of s s operaional, economy and environmenal, respecively. Pracically, p 44% s1 signifies he likelihood of aircraf possession (via acquisiion or leasing) in accordance o he operaional aspec of airlines ( s 1 ) while is complemen, i.e. ps 2 38% and p 18% s refer o he probabiliy of aircraf possession by 3 aking ino accoun he aspec of economy ( s 2 ) and environmenal ( s 3 ). Noe ha p 1 S i and his implies 100% of he full (complee) decision-making of i airlines in flee planning. s Table 4.13: The Evaluaion of Key Aspec in Flee Planning Decisional crieria Key aspec DP (0.1977) CE (0.2630) PP (0.3396) TR (0.1997) Probabiliy Operaional, Economy, s Environmenal, s

188 4.3.4 An Applicaion in Solving Flee Planning Problem In such a complicaed air ransporaion sysem, airlines encouner many challenging unexpeced evens which are unpredicable in naure. In accordance o he occurrence of unexpeced evens (risks), an efficien flee planning is necessary. As such, he probable phenomena, s,..., s 1 k for a oal of k phenomena are defined o describe he possible key aspec of aircraf possession in meeing sochasic demand under uncerainy. The probabiliy of probable phenomena, p,..., p quanifies he likelihood (probabiliy) of s1 sk respecive key aspec (deerminan) in making flee planning decision via aircraf acquisiion/leasing. In oher words, hey define how he flee supply (aircraf composiion) is made o mee ravel demand. Preferably, he quaniy of aircraf should be available adequaely (a a righ ime) for a sraegic flee planning decision. If probable phenomena and is probabiliy are no defined, i means ha airlines only deals wih one possible aspec o mee sochasic demand, i.e. hey have perfec confidence ha a cerain level of sochasic demand will be me by considering a single aspec only. However, his should no be he case because he acual decision-making process is subjec o muliple crieria (aspecs). Furhermore, he decision-making may vary from ime o ime under uncerainy. As such, his indicaor is necessary in flee planning. The number of probable phenomenon varies depending on he percepion and consideraion of airlines. In his research, hree key aspecs (probable phenomena), namely operaional, economy and environmenal are considered. 165

189 4.3.5 Flee Planning Decision Model In general, he airline's profi is conribued by he oal revenue and he P L oal operaing cos. For operaing period, he oal revenue, TR( I I ) can be expressed as follows: n m P L S S 1 TR I I E fare D sold resale for 1,..., T; S s,..., s iy iy k i1 y1 (4.17) For Equaion (4.17), he firs erm on he righ-hand side indicaes he expeced income from he sales of fligh ickes while he second erm denoes he revenue from he sales of aging aircraf. P L On he oher hand, he oal operaing cos, TC( I I ) of airlines can be formed as follows: n n n n m P L S S P L S i P P cos, i i i i i iy iy TC I I E D u purc x lease x hgf D A I dep i1 i1 i1 i1 y1 n m n n n L L P L I dep for 1,..., ; iy iy dp x dl x C fuel T S i i i i s,..., s i 1 k i1 y1 i1 i1 i1 (4.18) The erms on he righ-hand side of Equaion (4.18) signify he expeced cos of fligh, aircraf purchase cos, lease cos, mainenance cos, depreciaion expenses, payable deposi of aircraf acquisiion/leasing and fuel expenses, respecively. 166

190 In summary, flee planning model of airlines of operaing period 1,..., T can be presened as follows: P I P L P L p TR I I TC I I s1 L I max X 1 P L P L P L r p TR I I TC I I P I I sk P (4.19) subjec o: n P L Budge consrain: purc x lease x MAX i i i i budge () i1 i1 n (4.20), 1 n Demand consrain: S i S SEAT f D A D (4.21) i i1 P Sales of aircraf consrain: sold I iy ( 1) i( y 1) (4.22) Lead ime consrain: 1 1 DLT F (4.23) i LT LT Selling ime consrain: 1 1 DST F (4.24) i ST ST where S P L P L D, X, X, I, I, SOLD, O, R Z 0. For model (4.19), budge consrain ascerains wheher if he soluion is financially feasible for airlines while demand consrain ensures ha ravelers demand could be me saisfacorily. The consrain of sales of aircraf ensures ha he quaniy of aircraf sold is no more han he aircraf owned by airlines. Lead ime consrain and selling ime consrain respecively indicae when airlines are supposed o order new aircraf and release aging aircraf for sales. The erm r 1 is used for discouned value across he planning horizon while p sk indicaes he probabiliy of k-h probable phenomenon for having P I and L I as iniial flee supply (aircraf possession). Specifically, he elemen of p, p s1 s2 and p respecively signifies he probabiliy of operaional, economy and s 3 environmenal aspec of airlines in making sraegic flee planning decision. 167

191 Mahemaically, he developed opimizaion model, in he form of probabilisic dynamic programming model, can be solved by decomposing i ino a series of simpler sub-problems by using backward workings mechanism Daa Descripion A case sudy consising of hree ypes of aircraf, i.e. A , A and B are considered for a se of OD pairs for a planning horizon of eigh years. Mos of he daa are compiled based on he available repors (AirAsia Berhad, 2010a; Malaysia Airlines, 2010a; Airbus, 2010a, 2010b, 2010c; Boeing, 2012). For each operaing period, he level of sochasic demand is obained by applying he 5-sep modeling framework of sochasic demand. Based on he AHP modeling framework, he probabiliy of key aspecs (for he benchmark scenario) are p , p and 1 2 ps for he aspec of operaional, economy and environmenal, respecively. The daa inpu of benchmark scenario are lised as follows: Three probable phenomena are considered, i.e. k 3 P P P A = 1, quaniy of aircraf is I I I and I L I L I L A = 1, quaniy of aircraf o be hree years old is I P I P I P Budge, MAX $6,500 million budge () Discoun rae, r 5% Significance level of demand consrain, 5% s s 168

192 Significance level of lead ime consrain, 5% Significance level of selling ime consrain, 5% Deposi of aircraf acquisiion, DP 10% purc Deposi of aircraf leasing, DL 10% lease Seup cos, u 0 i s si si1 1 D D, D 1 α D, i 2, 3,.., k (4.25) The funcion of number of flighs is n 2 2 n 4 2 f A 9.776x10 A 7.83x10 [R 0.97] (4.26) The funcion of raveled mileage is n n g 2 2,066 f 2,875,383 [R 0.83] (4.27) The funcion of mainenance cos is h x x10 g [R 0.94] (4.28) The funcion of fuel expenses is The quaniy of aircraf is C fuel 7.46 f 8.3x10 f 98,572 [R 0.88] (4.29) n NA NP 73.6 [R 0.92] (4.30) where NP is he number of ravelers. In addiion o benchmark scenario, wo more scenario (as shown in Table 4.14) are examined for furher analysis o inspec relevan influenial inpu in generaing sraegic flee planning decision. 169

193 Scenario P Q Table 4.14: Furher Analysis in Solving Flee Planning Problem Decisional crieria Decision policy (DP) Travelers response (TR) Descripion The adjusmen on he relaive comparison of key aspec is done in he form as follows: environmenal operaional economy The change of ravelers response is invesigaed owards ravel cos reducion sraegy. According o mode choice analysis, he mode share was found o increase 18.39% in response o he sraegy of airline (AirAsia) in reducing 50% of ravel cos (airfare). B TR B DP Judgmen marix x Noe: For he relaive comparison of key aspec, he benchmark scenario is oulined in such a way as follows: operaional economy environmenal 3x3 Specifically, scenario P focuses on he changes of decision policy (DP) of airline, i.e. from he aspec of supply. By allocaing differen weigh (prioriy) on operaional, economy and environmenal aspec in such a way ha he environmenal aspec gains he highes concern (prioriy) in erms of he decision policy of airline, followed by operaional and economy aspecs, scenario P inspecs no only he possible variaion on he probabiliy of he respecive key aspec (operaional, economy and environmenal) bu also he possible changes of decision-making in flee planning. I is anicipaed ha he resulan oupus are driven by he weigh allocaion of airline based on he relevan decision policy (i.e. an influenial decisional crieria in flee planning). Scenario Q inspecs he perspecive of demand in erms of he changes of ravelers' response (TR). By capuring he possible changes of mode choice decision of ravelers owards he services of airlines (i.e. ravel cos reducion), scenario Q inspecs he impacs of demand level (in erms of choice probabiliy) in flee planning, i.e. by quanifying he possible probabiliy of operaional, economy and environmenal aspecs ha could affec flee 170

194 planning decision grealy. Specifically, raveler's response could be compiled by underaking various ravel survey on he respecive operaing neworks (shor-haul and medium/long-haul). I is anicipaed ha scenarios P and Q would capure he supply-demand ineracion in a greaer and beer manner Resuls and Discussion Benchmark Problem versus Scenario P The resuls of he case sudy are shown in Table 4.15 and 4.16 (wih he graphical resuls as displayed in Figure 4.4). The resuls imply ha he decisional crieria could affec he probabiliy of he key aspecs (operaional, economy and environmenal), opimal profi of airline as well as flee planning decision-making. From he resuls of he benchmark scenario, i could be seen ha he relaive comparison of key aspec for decisional crieria ends o produce he probabiliy of key aspec in abou he same way. This could be seen in Tables 4.14 and 4.15 for which he judgmen marix of benchmark scenario which has he relaive comparison of he key aspec in he form of operaional economy environmenal would produce he probabiliy of key aspec in similar way, i.e. probabiliy of operaional aspec probabiliy of economy aspec probabiliy of environmenal aspec. Some changes o he probabiliy of key aspec could be seen in scenario P for which he adjusmen 171

195 of he relaive comparison of key aspec has been done wih regard o he decision policy of airline (while oher decisional crieria remain unchanged). Scenario P was oulined by puing more weigh (relaive comparison) on environmenal aspec insead of operaional and economy aspec. Subsequenly, he resuls in Table 4.15 show ha he probabiliy of environmenal aspec increases abou 30% while he probabiliy of operaional and economy aspecs decreases (compared o benchmark scenario). However, he probabiliies of operaional and economy aspecs are sill higher compared o environmenal aspec. This happens because he resulan probabiliy of key aspec is affeced no only by he decision policy of airlines bu also oher decisional crieria. This signifies ha he decisional crieria of flee planning have a direc and influenial impac on he resulan probabiliy of he key aspec which would subsequenly consiue an opimal soluion of airline. Generally, he resuls confirm ha here s a linkage beween he decisional crieria and he probabiliy of key aspec as well as he opimal soluion of he flee planning model. Therefore, he key aspec in flee planning has o be quanified wisely. Table 4.15: The Resuls of Flee Planning Model Scenario Probabiliy of key aspec Average profi Operaional Economy Environmenal ($ millions) Benchmark P (-3%) (-11%) (+30%) 302 (-2.3%) Q (-1%) (+2%) (-2%) 395 (+27.8%) Noe: The value in bracke denoes he improvemen level compared o he benchmark scenario. 172

196 Table 4.16: The Resuls of Flee Size Aircraf Purchase/lease Benchmark scenario Scenario P Scenario Q A320 Purchase Lease A330 Purchase Lease B747 Purchase Lease Toal flee size (by year 8) 122 (purchase:104, lease:18) 122 (purchase:103, lease:19) 130 (purchase:107, lease:23) Figure 4.4: The Graphical Resuls of he Probabiliy of Key Aspecs In erms of he profi level of airline, Table 4.15 shows ha he benchmark scenario produces a higher profi han scenario P (in average). I could be seen ha he average profi of airline decreases 2.3% (abou $7 millions) when he probabiliy of operaional and economy aspecs decreases (for scenario P). Thus, i could be deduced ha a higher concern (or relaive comparison) on operaional and economy aspecs ends o produce a higher profi. This could be explained by he fac ha he operaional aspec of airline plays a vial role o generae income and opimal profi in meeing ravel demand. Therefore, his aspec is more revenue-sensiive (from he economy 173