Operationalizing Dynamic Pricing Models: Bayesian Demand Forecasting and Customer Choice Modeling for Low Cost Carriers

362 Pages·2011·4.372 MB·English

by  Steffen Christ (auth.)

Steffen Christ
Operationalizing Dynamic Pricing Models
GABLER RESEARCH
Steffen Christ

Operationalizing
Dynamic Pricing Models
Bayesian Demand Forecasting
and Customer Choice Modeling
for Low Cost Carriers
With a foreword by Prof. Dr. Robert Klein
RESEARCH
Bibliographic information published by the Deutsche Nationalbibliothek
The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliografi e;
detailed bibliographic data are available in the Internet at http://dnb.d-nb.de.
Dissertation University of Augsburg, 2009

1st Edition 2011
All rights reserved
© Gabler Verlag | Springer Fachmedien Wiesbaden GmbH 2011
Editorial Offi ce: Stefanie Brich | Nicole Schweitzer
Gabler Verlag is a brand of Springer Fachmedien.
Springer Fachmedien is part of Springer Science+Business Media.
www.gabler.de
No part of this publication may be reproduced, stored in a retrieval system
or transmitted, in any form or by any means, electronic, mechanical, photo-
copying, recording, or otherwise, without the prior written permission of the
copyright holder.
Registered and/or industrial names, trade names, trade descriptions etc. cited in this publica-
tion are part of the law for trade-mark protection and may not be used free in any form or by
any means even if this is not specifi cally marked.
Cover design: KünkelLopka Medienentwicklung, Heidelberg
Printed on acid-free paper
Printed in Germany
ISBN 978-3-8349-2749-1
“It is very difficult to make an accurate prediction,
especially about the future.”
Niels Bohr1
1 Niels (Henrik David) Bohr (October 7, 1885 – November 18, 1962) was a Danish
physicist who made fundamental contributions to understanding atomic structure
andquantummechanics. Bohriswidelyconsideredoneofthegreatestphysicistsof
thetwentiethcentury.
Foreword
Followingthesuccessoflowcostcarriers,dynamicpricinghasbecomeoneof
the most popular fields of research at the interface of Marketing and Opera-
tions Management. However, analyzing the available literature reveals that
most publications concentrate on the development of optimization models
for pricevariation. Themajor challenge offorecasting demandis most often
ignored. Withhisdissertation,SteffenChristaimstoclosethecorresponding
gap by examining the applicability of existing optimization models for dy-
namic pricing on the specifics of highly volatile consumer markets using the
example of low cost air carriers. The explicit objective is the operational-
ization of theoretically sound standing dynamic pricing models using only
realistic input assumptions and retracting to factually available data.
The approach chosen by Steffen Christ is rooted in the development of
selflearningdemandmodelsthatcalibratetheirparametersasdatabecomes
available yielding the option of using the returned results as input to con-
ventional dynamic pricing models. It claims that the developed models can
provide parts of the necessary input data in a merely plug-and-play fash-
ion. Generally, the work of Steffen Christ targets the understanding of both
relevantinputvaluesfordynamicpricingmodels, forecastingoflatentprice-
independent demand and models that estimate the price sensitivity of such
latent demand.
Forforecastingoflatentdemand,SteffenChristdevelopsamethod,which
isbasedonaBayesianinterpretationoflinearregressionmodeling. Here,the
parameters of a linear function are not considered deterministic, unknown
and to be estimated based on a stochastic data set, but are considered ran-
dom numbers each with a stochastic distribution that is iteratively learned
basedoncollecteddataitselfseenasdeterministic. Themethodislessprone
to distortions if few data is available and explicitly allows the inclusion of
subjective or expert knowledge into the estimation process.
The concluding evaluation of different self-learning models is done using
a proprietary software environment. The forecast error for latent demand
VIII FOREWORD
ranges between 16.1 – 17.4% based on single values and 11.2 – 12.7% based
on the total demand for a single flight leg.
Furthermore,SteffenChristconsiderstheestimationofpurchasebehavior
forindividualsexpressingsuchlatentdemand. ThechosenmethodCustomer
Choice Modeling uses disaggregate observations of discrete and individual
customer behavior that depend on the different price points for air travel
found in the market and their individual attributes (e.g., flight schedule).
Through the combination of multiple databases (the demand protocol from
theairline’sonlinechannel,itscomputerreservationsystemandpricingdata
collectedthroughwebcrawlers)SteffenChristisabletoconstructacompre-
hensive data field providing revealed preferences as basis for the estimation
of individual purchase behavior.
That later model yields a forecast error of 14 – 27% on the completed
bookings for the outbound direction and 26 – 39% for the inbound direc-
tions, what is considered satisfactory based on the data limitations concern-
ing bookings received by the considered airline’s competition.
Insummary,SteffenChristshowswithhisworkthatitisindeedpossible
todevelopforecastingmodelsforboth,latentdemandandpurchasebehavior,
even in highly dynamic and volatile markets. His excellent work is of high
relevance for both, researchers and industry experts in the field of dynamic
pricing. I hope, that his work will find many readers and will receive the
recognition it deserves.
Prof. Dr. Robert Klein
Acknowledgements
It is an honor for me to thank my adviser and primary reviewer Prof. Dr.
Robert Klein for the valuable guidance and continuous support throughout
thecompletionofthiswork. DuringnumerousdiscussionsIreceivedprecious
advice with regards to both, methodology and subject matter. Furthermore
he has always been a truly inspiring and motivational mentor.
I would also like to show my gratitude to Prof. Dr. Michael Krapp for
kindly undertaking the second review and for his important methodical ad-
vice reflected in all statistic sections of this work.
I am indebted to all colleagues at the chair in “Analytics & Optimiza-
tion” for their support of many kinds, particularly Dr. Jochen G¨onsch and
Dr. Claudius Steinhardt for their ample structural remarks.
It is a pleasure to thank Achim Lameyer and Michael Stellwagen, who
foremost made this work possible by supplying not only real-life data but
also disclosing lots of their practical experience and knowledge.
SpecialthanksgotocolleaguesandfriendsatMcKinsey&Company,Inc.
forsharedtimesofbalefulnessandcontinuoussupplywithcoffeeanddistrac-
tion–withaspecialmentiontoDr.AnjaLindauwhoisthepickofthebunch.
This work would not have been possible without the unfailing support,
encouragement and love of my fianc´ee Sarah Hein. The completion of this
dissertation is eventually due to her constant motivation and belief.
Ultimately I owe my deepest gratitude to my beloved parents Marianne
and Manfred Christ, who have continuously supported and encouraged me
in all my endeavors – peaking in the completion of this dissertation.
Steffen Christ
Contents
List of Figures XV
List of Tables XXI
Nomenclature XXIII
Mathematical Nomenclature XXV
Mathematical Notation XXVII
I Dynamic Pricing in the Airline Industry 1
1 Introduction 3
1.1 The Passenger Airline Industry. . . . . . . . . . . . . . . . . . . 3
1.2 The Low Cost Revolution . . . . . . . . . . . . . . . . . . . . . . 6
1.3 The Advent of Dynamic Pricing . . . . . . . . . . . . . . . . . . 11
2 Motivation and Structure 15
2.1 Relevance of the Topic . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2 Focus on the Airline Industry. . . . . . . . . . . . . . . . . . . . 18
2.3 Objective and Differentiation . . . . . . . . . . . . . . . . . . . . 19
2.4 Structure of Work . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3 Dynamic Pricing 23
3.1 Definition and Scope . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.1.1 Introduction to Pricing . . . . . . . . . . . . . . . . . . . 23
3.1.2 Dynamic Pricing and Revenue Optimization . . . . . . 25
3.2 Literature Overview . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.2.1 Demand Learning Models. . . . . . . . . . . . . . . . . . 31
3.2.2 Non-learning Pricing Models . . . . . . . . . . . . . . . . 42
3.2.2.1 Dynamic Pricing with Myopic Customers. . . 45
XII CONTENTS
3.2.2.2 Dynamic Pricing with Strategic Customers. . 51
3.2.2.3 Customer Choice Models . . . . . . . . . . . . 53
3.3 Limitations and Shortcomings . . . . . . . . . . . . . . . . . . . 54
3.3.1 Dynamic Pricing Models . . . . . . . . . . . . . . . . . . 54
3.3.2 Demand Learning Models. . . . . . . . . . . . . . . . . . 56
3.4 Proposed Approach. . . . . . . . . . . . . . . . . . . . . . . . . . 58
II Forecasting Latent Demand 63
Part II Objective 65
4 Self-Learning Linear Models 67
4.1 Linear Regression Models . . . . . . . . . . . . . . . . . . . . . . 68
4.2 Bayesian Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . 79
4.2.1 Bayesian Probabilities . . . . . . . . . . . . . . . . . . . . 80
4.2.2 Bayesian Inference . . . . . . . . . . . . . . . . . . . . . . 83
4.3 Bayesian Linear Regression . . . . . . . . . . . . . . . . . . . . . 85
4.3.1 Parameter Distribution . . . . . . . . . . . . . . . . . . . 85
4.3.2 Predictive Distribution . . . . . . . . . . . . . . . . . . . 89
4.4 Critique and Limitations . . . . . . . . . . . . . . . . . . . . . . 92
5 Demand in Low Cost Markets 97
5.1 Experimental Data Set. . . . . . . . . . . . . . . . . . . . . . . . 97
5.1.1 Data Collection. . . . . . . . . . . . . . . . . . . . . . . . 98
5.1.2 Data Cleansing . . . . . . . . . . . . . . . . . . . . . . . . 101
5.2 Overarching Long-term Characteristics . . . . . . . . . . . . . . 103
5.2.1 Log-linear Demand Structure . . . . . . . . . . . . . . . 104
5.2.2 Macro-Seasonalities and Trends . . . . . . . . . . . . . . 110
5.2.3 Similarities of Adjacent Flights . . . . . . . . . . . . . . 113
5.3 Short-term Characteristics . . . . . . . . . . . . . . . . . . . . . 115
5.3.1 Time Series Disruption Through Outliers . . . . . . . . 116
5.3.2 Patterns Based on Departure Weekdays . . . . . . . . . 121
5.3.3 Micro-Seasonalities along Observation Weekdays . . . . 125
5.3.4 Cross-Effects of Departure and Observation Weekdays 128
5.4 Implications for Forecasting Model . . . . . . . . . . . . . . . . 129
6 The Demand Forecasting Model 131
6.1 Linear Basis Function Model . . . . . . . . . . . . . . . . . . . . 131
6.1.1 Indexing and Data Transformation . . . . . . . . . . . . 132
6.1.2 Driving Model Parameters . . . . . . . . . . . . . . . . . 134