RM Methods for Airline Fare Family Structures by Vincent B. Surges B.A., Mathematics, The College of St. Scholastica, 2011 Submitted to the Department of Aeronautics and Astronautics in partial fulfillment of the requirements for the degree of Master of Science at the MASSACHUSETTS INSTITUTE OF TECHNOLOGY June 2013 (cid:13)c Massachusetts Institute of Technology 2013. All rights reserved. Author .............................................................. Department of Aeronautics and Astronautics May 10, 2013 Certified by.......................................................... Peter P. Belobaba Principal Research Scientist, Aeronautics and Astronautics Thesis Supervisor Accepted by......................................................... Eytan H. Modiano Professor of Aeronautics and Astronautics Chair, Graduate Program Committee 2 RM Methods for Airline Fare Family Structures by Vincent B. Surges Submitted to the Department of Aeronautics and Astronautics on May 10, 2013, in partial fulfillment of the requirements for the degree of Master of Science Abstract The rapid growth of low cost carriers forced many legacy airlines to simplify their farestructuresanddevelopnewpricingstrategiestoremaincompetitive. Thestrategy of branded fares, or “fare families”, is an increasingly popular approach for airlines to differentiate their products and services from other competitors. This thesis provides a comprehensive overview of revenue management (RM) fore- casting and optimization methods developed specifically for fare family structures. These methods, collectively termed Q-Forecasting for Fare Families (QFF), provide airlines with the capability to manage branded fares from a RM perspective. The QFF methods are all constructed based on the assumed fare family passenger choice model, which accounts for both willingness-to-pay estimates as well as family pref- erence. Each formulation makes underlying assumptions regarding passenger sell-up and buy-across. The Passenger Origin Destination Simulator is used to test and compare the per- formance of each QFF formulation in a dual airline competitive environment, both with leg-based RM controls as well as network RM controls. The results from the simulations indicate that substantial gains in both revenue and yield over traditional RM methods can be achieved with appropriate RM in a fare family structure. Specif- ically, while Hybrid Forecasting (with leg RM controls) generates a 4.0% increase in revenue over Standard Forecasting, QFF is shown to increase revenues by more than 12.5%. The benefits of QFF are greater with network RM controls, with potential revenue increases of nearly 14.0% (over Standard Forecasting). The positive results obtained with each QFF formulation are dependent upon an appropriate estimate for passenger sell-up and family preference. Consequently, this research also illustrates the importance of the estimate for passenger willingness-to- pay and its relationship to forecasting and optimization in airline RM. Thesis Supervisor: Peter P. Belobaba Title: Principal Research Scientist, Aeronautics and Astronautics 3 4 Acknowledgments First and foremost, I would like to express my sincere gratitude to my research advisor, Dr. Peter Belobaba. He introduced me to the field of revenue management, and made the transition to MIT extremely smooth. He continually provided guidance and support to me, in addition to always making himself available for any questions I mighthavehad. Thisthesiswouldnotbepossiblewithouthiscontinualsupport; Iam truly grateful for everything he has provided for me, least of the all the opportunities I have after MIT. His vast knowledge of the airline industry and professional intuition towards revenue management has contributed tremendously to my research findings. I would also like to thank Craig Hopperstad, the developer of the Passenger Origin Destination Simulator (PODS), the simulation tool on which my results are based. He has provided extensive technical support on a wide variety of issues within PODS. His patience and guidance over the past two years, as well as his explanations of intricate concepts, all contributed to this thesis. I extend my appreciation to all the members of the PODS Consortium: United Airlines, Delta Airlines, Boeing, Air Canada, LAN, Lufthansa, and SAS. I had the opportunity to participate in five PODS conferences during my two years at MIT. Theseconferencesprovidedinvaluableadviceandsupportformyresults,andprovided me guidance on future research based on the members’ experiences. I am very grateful to have studied at MIT over the past two years. My time here has given me unparalleled research skills, while enhancing my skill sets in many other areas. I know that my tenure at MIT would not have been as exciting if not for all of the friendships that were formed both within my lab as well as with other graduate students throughout campus. I would like to especially thank Pierre-Olivier, Michael, Natasha, Tamas, Vishnu, and Alexander. Our time together was truly memorable, and I look forward to corresponding with each of you as we start the next chapter of our lives. And to the rest of the International Center for Air Transportation lab, including Eric, Mike, Karim, and Alyona, I want to thank you for making my time at MIT so enjoyable. 5 Most importantly, I must heartily thank my family and friends for their uncondi- tional love and support. In particular, my wife Amanda and my parents. Without the constant support and understanding from my wife, these last two years would not have been possible. Amanda, I am eternally grateful for your love, and for you always being there for me, even when MIT tried to interfere! And to my parents, I am very grateful for everything you have done to make this moment possible. You both have always supported me and encouraged me to achieve my highest educational goals. Lastly, I would like to extend an appreciation out to the rest of my family for their constant support. 6 Contents 1 Introduction 15 1.1 Overview of the Airline Industry . . . . . . . . . . . . . . . . . . . . . 16 1.2 Branded Fare Families . . . . . . . . . . . . . . . . . . . . . . . . . . 20 1.3 Objectives of the Thesis . . . . . . . . . . . . . . . . . . . . . . . . . 23 1.4 Structure of the Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . 23 2 Literature Review 25 2.1 Traditional Airline RM Methods . . . . . . . . . . . . . . . . . . . . . 25 2.1.1 Seat Allocation Models . . . . . . . . . . . . . . . . . . . . . . 27 2.1.2 Demand Forecasting Models . . . . . . . . . . . . . . . . . . . 30 2.2 RM Methods for Less-Restricted Fare Structures . . . . . . . . . . . . 31 2.2.1 Q-Forecasting . . . . . . . . . . . . . . . . . . . . . . . . . . . 33 2.2.2 Hybrid Forecasting . . . . . . . . . . . . . . . . . . . . . . . . 34 2.2.3 Fare Adjustment Theory . . . . . . . . . . . . . . . . . . . . . 35 2.3 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37 3 Revenue Management Methods for Fare Families 39 3.1 Introduction to Fare Family Structures . . . . . . . . . . . . . . . . . 39 3.2 Passenger Decision Process . . . . . . . . . . . . . . . . . . . . . . . . 40 3.3 Q-Forecasting for Fare Families . . . . . . . . . . . . . . . . . . . . . 42 3.3.1 QFF1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43 3.3.2 QFF2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52 3.3.3 QFF3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56 7 3.4 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60 4 Overview of the Passenger Origin Destination Simulator (PODS) 63 4.1 PODS Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63 4.1.1 Passenger Choice Model . . . . . . . . . . . . . . . . . . . . . 64 4.1.2 Revenue Management System . . . . . . . . . . . . . . . . . . 68 4.2 Simulation Environment . . . . . . . . . . . . . . . . . . . . . . . . . 70 4.2.1 Network D10 . . . . . . . . . . . . . . . . . . . . . . . . . . . 71 4.3 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74 5 PODS Simulation Results 77 5.1 Non-Overlapping Fare Family Structures . . . . . . . . . . . . . . . . 78 5.1.1 Leg-based Controls in Symmetric RM Experiments . . . . . . 79 5.1.2 OD Controls in Symmetric RM Experiments . . . . . . . . . . 89 5.1.3 Leg-based Controls in Competitive RM Experiments . . . . . 90 5.1.4 OD Controls in Competitive RM Experiments . . . . . . . . . 96 5.2 Overlapping Fare Family Structures . . . . . . . . . . . . . . . . . . . 100 5.2.1 Symmetric RM Experiments . . . . . . . . . . . . . . . . . . . 101 5.2.2 Leg-based Controls in Competitive RM Experiments . . . . . 104 5.2.3 OD Controls in Competitive RM Experiments . . . . . . . . . 108 5.3 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108 6 Conclusions 111 6.1 Summary of Thesis Objectives . . . . . . . . . . . . . . . . . . . . . . 112 6.2 Summary of Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 114 6.3 Potential Directions for Future Research . . . . . . . . . . . . . . . . 116 8 List of Figures 1-1 Airline Practicing Differential Pricing (d’Huart, 2010) . . . . . . . . . 17 1-2 Fare Family Structure Offered by ANZ for Sydney-Auckland Market. Data source: AirNewZealand.com . . . . . . . . . . . . . . . . . . . . 21 2-1 A Typical Third Generation RM System (Barnhart et al., 2003) . . . 26 2-2 Nested Bookings Limits and Class Protection Levels (Cleaz-Savoyen, 2005) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 2-3 Spiral-Down Effect (Tam, 2008) . . . . . . . . . . . . . . . . . . . . . 32 2-4 Process of Q-Forecasting (Belobaba, 2010) . . . . . . . . . . . . . . . 33 2-5 Process of Hybrid Forecasting (Belobaba, 2010) . . . . . . . . . . . . 35 2-6 Scatter Plot of Different Strategies Tracing out the Convex Hull (Fiig et al., 2010) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37 3-1 Assumed Passenger Decision Process in a Fare Family Structure . . . 42 3-2 QFF1 Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44 3-3 Typical FRAT5 Curve . . . . . . . . . . . . . . . . . . . . . . . . . . 46 3-4 Disutility Distribution of Random Passenger. Area of Shaded Region is Probability that Passenger Buys Up from f2 to f1 . . . . . . . . . 48 3-5 Scatter Plot of Total Revenue vs. Total Demand for all Policies . . . 51 3-6 QFF2 Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52 4-1 PODS Architecture (Belobaba, 2010) . . . . . . . . . . . . . . . . . . 65 4-2 Booking Curves in PODS by Passenger Type . . . . . . . . . . . . . . 66 4-3 WTP Curve by Passenger Type . . . . . . . . . . . . . . . . . . . . . 67 9 4-4 Different FRAT5 Curves used in PODS . . . . . . . . . . . . . . . . . 70 4-5 Map of Network D10 . . . . . . . . . . . . . . . . . . . . . . . . . . . 71 4-6 Fare Family Structure with Non-Overlapping Price Points . . . . . . 73 4-7 Fare Family Structure with Overlapping Fares in Network D10 . . . . 74 5-1 Baseline Revenues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80 5-2 Baseline Business/Leisure Bookings for Airline 1 . . . . . . . . . . . . 81 5-3 Revenues with QFF . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82 5-4 Revenues with QFF1 with different FRAT5 Curves and PBUP Values 83 5-5 Booking Class Mix with QFF1 (PBUP 3.5) with different FRAT5 Curves 84 5-6 Revenues with QFF2 with different FRAT5 Values by Fare Family . . 85 5-7 Booking Class Mix with QFF2 with different FRAT5 Inputs by Fare Family . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86 5-8 Class E1 Closure Rates with QFF2 with different Family 1 FRAT5 Values (Family 2 FRAT5 2.0) . . . . . . . . . . . . . . . . . . . . . . 87 5-9 Results with QFF3 (FRAT5 1.5|2.0) with different DUMLT Inputs . 88 5-10 Revenues with Standard Forecasting and QFF . . . . . . . . . . . . . 89 5-11 Revenues when Airline 1 uses Hybrid Forecasting and QFF . . . . . . 91 5-12 Booking Class Mix when Airline 1 uses HF and QFF . . . . . . . . . 92 5-13 Revenues when Airline 1 uses QFF1 (PBUP 2.0) with different FRAT5 Curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93 5-14 RevenueswhenAirline1usesQFF3(DUMLT2.0)withdifferentFRAT5 Values by Fare Family . . . . . . . . . . . . . . . . . . . . . . . . . . 94 5-15 Airline 1 Bookings with QFF3 (DUMLT 2.0) with different FRAT5 Values by Fare Family . . . . . . . . . . . . . . . . . . . . . . . . . . 95 5-16 Revenues with Standard Forecasting and QFF . . . . . . . . . . . . . 96 5-17 Business/LeisureBookingsforAirline1withStandardForecastingand QFF3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97 5-18 Cumulative Family 1 and Family 2 Bookings by Time Frame with Standard Forecasting and QFF3 . . . . . . . . . . . . . . . . . . . . . 98 10