Technical Program
RM
Presentations ( Based on submitted
abstracts to date* )
Customer
Centric Dynamic Pricing and Availability
Steve
Pinchuk
SAS
Institute
The topic
shows how RM, pricing, CRM and marketing need to be automated based on
future expected interactions and value from individual guests...
Optimization of Mixed Fare
Structures: Theory and Applications
Thomas Fiig, Karl
Isler, Craig
Hopperstad and Peter
Belobaba
Scandinavian
Airlines System, Swiss
International Air Lines Ltd, Hopperstad
Consulting Inc. and MIT International Center for Air Transportation
This
paper develops a theory for optimizing revenue that can be applied in a
variety of airline fare structures, including the less restricted and fully
undifferentiated fare products that have become more common in the recent
past. We describe an approach to transform the fares and the demand
of a general discrete choice model to an equivalent independent demand
model. The transformation and resulting fare adjustment approach is valid
for both static and dynamic optimization and extends to network revenue
management applications. This transformation allows the continued use
of the optimization algorithms and control mechanisms of traditional
revenue management systems, developed more than two decades ago under the
assumption of independent demands for fare classes.
Fare
Families: RM Forecasting and Optimization Models
Peter Belobaba
and Craig Hopperstad
MIT International Center for Air
Transportation and Hopperstad Consulting
Fare families represent an alternative strategy for fare product
differentiation -- two or more distinct product families are offered.
Within each family, there are multiple price levels with identical
attributes. The RM challenge is then to incorporate both passenger
choice of fare family and willingness to sell-up within a family. We
develop a forecasting and optimization approach for fare families, and
present PODS simulation results of the revenue impacts of this approach
relative to traditional and hybrid forecasting methods. The
simulation results also illustrate some of the fundamental structural
differences between fare families and other fare structures.
Demand
Arrival Order and RM Controls
Guillermo
Gallego, Lin Li
and Richard Ratliff
Columbia
University and Sabre Holding Inc.
Most
airline revenue management systems assume a "low-before-high
(LBH)" fare class demand arrival order (i.e. lower valued classes book
strictly before higher valued ones). Many published academic papers assume
a pure mixed demand arrival process in which all fare classes have the same
shape booking profile. Empirical analysis shows that actual demand arrival
order is typically in between these two extremes. We discuss the revenue
impact of incorrect arrival order assumptions and provide methods for
computing inventory controls considering demand arrival order and the type
of nesting controls used in the airline reservations system.
An Overview
of RM Optimization with Dependent Demands
Larry Weatherford and Richard Ratliff
University of Wyoming and Sabre Holding Inc.
This presentation will provide
an overview of the latest techniques for handling the difficult problem of
dependent demands in airline revenue management optimization. Comparisons
of new approaches being used in practice are provided including
methodology, indicative revenue performance and implementation
considerations.
Using
uncertainty products in the airline industry to fill the gaps and increase
revenues
David Post
SigmaZen
Airlines,
whether they be network carriers or low-cost
carriers, are faced with high fixed costs, low marginal costs and
perishable inventory. Hence any method that enables an airline to sell a proportion
of its distressed inventory in a way that does not dilute revenue damage
brand equity or start a price war should be worth considering.
This
paper presents a successful method that has been implemented at a large
European LCC that has significantly enhanced revenues by marketing some of
the airline's distressed inventory to a previously untapped customer base:
those customers prepared to accept a high degree of uncertainty in their
travel destination. Customers can also self-segment into various
destination pools (e.g. warm-water destinations, cultural destinations, or
east European destinations) and for an increase in price remove various
unattractive destinations from the pool.
Availability
Calculation Based on Robust Optimization
Benoit Lardeux and Oriana Goyons
AMADEUS
The presentation describes a novel method of dynamic airline seat inventory
control that is robust to uncertainties in the O&D demand forecast. The
method is based on the computation of the opportunity cost of selling
airline seats. The determination of the opportunity cost relies on a robust
formulation that consists in minimizing the maximal regret between
solutions from several scenarii of demand. A fast
solution approach was implemented and tested on large network instances in
simulation. The response times that were obtained make it suitable for
airline IT systems. Moreover, these controls lead to an increase in revenue
compared to state-of-the-art revenue maximization methods in our simulation
test. Finally, we discuss the implications of the integration of the
algorithms in a current revenue management system of an airline and why
this process can be a valuable alternative to bid-price control.
A
risk ratio procedure for estimation of market size and parameters
Kalyan Talluri
Universitat Pompeu Fabra
Estimation of market-size when no-purchases are unobservable has rarely
been attempted in the marketing or revenue management literature. Indeed,
we point out that it is akin to the classical statistical problem of
estimating the parameters of a binomial distribution with unknown
population size and success probability, and hence likely to be
challenging. However, when the purchase probabilities are given by a
functional form such as a multinomial-logit
model, we propose an estimation heuristic that exploits the speci
cation of the functional form, the variety of the
oer sets in a typical RM setting, and qualitative
knowledge of arrival rates. Finally we perform simulations to show that the
estimator is very promising in obtaining unbiased estimates of population
size and the model parameters.
On
bounds for Network Revenue Management
Kalyan
Talluri
Universitat Pompeu Fabra
Recently Adelman [1] and Topaloglu [18] have
proposed new upper bounds, the affine relaxation
(AR) bound and the Lagrangian relaxation (LR)
bound respectively, and showed that their bounds are tighter than the DLP
bound. Tight bounds are of great interest as it appears from empirical
studies and practical experience that models that give tighter bounds also
lead to better controls (better in the sense that they lead to more
revenue). In this paper we give tightened versions of three known bounds,
calling them sAR (strong Affine Relaxation), sLR (strong Lagrangian
Relaxation) and sPHLP (strong Perfect Hindsight
LP), and show relations between them.
Impacts
of Matching Competitor RM Seat Availability
Olivier d'Huart and Peter
Belobaba
MIT International Center for Air Transportation
Matching the availability of the lowest open competitor’s class has been used
as a competitive revenue management strategy by various airlines. We
present PODS simulation results of the impacts on an airline and its
competition of such a strategy implemented by RM controllers after the
usual RM optimization. The results show that such a strategy can be
counter-effective from a RM point-of-view. Although it results in lower
load factors for the competitor, in terms of revenue it consistently hurts
the matching airline while yielding better results for the targeted
competitor.
Computing
Bid-prices for Revenue Management under Customer Choice Behavior
Juan Manuel Chaneton and Gustavo Vulcano
Universidad de Buenos Aires and New
York University
We develop a stochastic
approximation algorithm to compute bid prices for network revenue
management, accounting explicitly for choice behavior effects. One of the
main practical advantages of this proposal is that it can be built-in
as an extra layer on current RM systems that implement bid-price
controls. Our numerical experiments show that the approach has
interesting potential from a revenue performance perspective.
An
Improved Dynamic Programming Decomposition Approach for Network Revenue
Dan Zhang
McGill
University
We
introduce a variant of the resource-based dynamic programming decomposition
for network revenue management, which is based on a nonlinear non-separable
functional approximation to the value function of a dynamic programming
formulation of the problem. We propose a parallel dynamic programming
approach to solve the resulting problem which is a nonlinear optimization
problem with nonlinear constraints. Our approach leads to a tighter upper
bound on revenue than the decomposition bound recently introduced in Zhang
and Adelman (2008). The computational cost of this new decomposition
approach is only slightly higher than the classical version. We report
encouraging numerical results.
Stationary
of Demand Arrival Rate Process and Nesting Booking
Jean
Michel Chapuis and Lindsay Temaiana
University
of French Polynesia
Revenue
Managers usually nest the booking limits to avoid the situation in which
high-fare bookings are rejected in favor of low-fare class. To date, there
are both net nesting and threshold nesting methods. However
the consequence on revenues of each one is not clearly understood. This
research investigates their underlying assumptions and supports that the
stationary of the demand process is the key point. This paper also suggests
a co-integration test and an event study
methodology to know what is appropriate in practice.
Forecast
Evaluation and Simulation Based Key Performance Indicators
Catherine
Cleophas
University
of Paderborn
This presentation features results
from the implementation of a simulation-based approach including a
flexible, hybrid demand model. The
knowledge of the customer demand model available in such a system allows
for the generation of so-called psychic forecasts. Ways of evaluation forecast methods for
revenue management with the help of psychic forecasts are described. Alternative approaches to defining and
generating a psychic forecast are listed and illustrated. The risks and benefits of benchmarking
forecast methods in this way are demonstrated with simulation experiments.
Integrating
choice-based models with capacity-based RM routines
Mark Ferguson and Laurie Garrow
Georgia Institute of Technology
This paper investigates
choice-based control policies for RM for multi-product industries.
The addition of multiple product types complicates the choice-based RM
problem. Although the literature provides promising insights for
applications of choice-based models in single-product RM frameworks, our
preliminary results show that neither the methodology nor the findings can
be directly applied to multi-product industries. This paper develops
several new methodologies and extensions to the single-product choice-based
RM case and compares the performance of these extensions against traditional
RM techniques using actual industry data.
Two of the key (inter-related)
methodological extensions developed and tested in the paper relate to
demand unconstraining (or representation of
customers who search for a particular product but ultimately do not
purchase) and choice set composition (or defining the set of products in a
way that accurately reflects how customers make trade-offs among different
products). A MNL model is used to investigate how consumers make trade-offs
among several product attributes, including price, restrictions, and
type. Different utility functions that incorporate these trade-offs
are investigated. A particular emphasis is placed on developing
alternative formulations that help overcome potential forecasting errors
introduced when including a “no purchase” alternative in the universal
choice set that is assumed to have zero utility.
Multivariate
Demand Modeling and Estimation from Censored Sales
Catalina Stefanescu
London Business School
Current practice in demand modeling focuses on univariate
demand forecasting, where models are built separately for each product.
However, there is empirical evidence of correlated product demand. Also,
demand is usually observed in several periods during a selling horizon, and
it may be truncated due to inventory constraints so that only censored
sales are recorded. In this paper we propose a class of models for
multi-product multiperiod aggregate demand
forecasting. We develop an approach for model estimation from censored
sales data in a maximum likelihood framework. Through a simulation study,
we show that the algorithm is computationally attractive under different
demand and censoring scenarios. We exemplify the methodology with the
analysis of booking data from the entertainment and the airline industries,
and show that the use of these models for airline revenue management
increases revenues by up to 11% relative to alternative forecasting
methods.
On
the robustness of the network based Revenue Opportunity Model
Christian Temath, Michael Frank and Stefan Poelt
University
of Paderborn
The
Revenue Opportunity Model (ROM) is a widely known method for measuring
Revenue Management performance. While adapting the ROM to the developments
in Revenue Management science, the question of applicability and in
particular the validity of the ROM became increasingly important. In our
presentation, we will introduce a simulation-based method for measuring the
robustness of the network-based ROM against errors in the input data and
present computational results. We will furthermore present an extension of
the ROM to consider for dependent demand structures in the model.
Sandbox
Testing a New Revenue Management Concept
Fernando Castejon Solanas,
Kalyan Talluri, Begoña Codina and
Juan Magaz
Iberia
LAE
In this
paper we describe a ``live" testing experiment on a set of flights at
an airline. A set of competing algorithms control a set of flights, and
their behavior and results are observed over a relatively long period of
time (9 months). In parallel, a group of control flights were run using the
traditional mix of manual and algorithmic control. Such ``sand-box"
testing, while common at many large internet search and e-commerce
companies is relatively rare in the revenue management area. Sand-box
testing has an undisputable model of customer behavior but the experimental
design and analysis of results is less clear. In this paper our goal is to
present (i) a rigorous design and live testing
framework (ii) present the problems and pitfalls we faced during the
experiment (iii) describe the econometric analysis of the results. Our
focus is not so much on the results, the revenue performance of the
algorithms, but on how we went about justifying and analyzing the results,
that we hope will serve as guidelines for such future live testing.
Application of a Revenue Management Simulator to
drive Strategy
Debraj Basumallick and Manmeet Singh
United
Airlines
This presentation
will provide an overview of an internally built simulator of our Revenue
Management systems. This simulator not only replicates our forecasting,
optimization and reservation systems but also has capabilities to model
some aspects of the customer behavior. This has been widely used tool to
build business cases, evaluate what-if scenarios and test hypotheses. We
will also be presenting a few of recent case studies that demonstrate the
use of the simulator.
Practical
Aspects of the Hybrid Approach to Revenue Management
Maarten Oosten and Darius Walczak
PROS
For many years the need for revenue management solutions that handle a mix
of priceable and yieldable demand has been
recognized. Since then various solutions to the forecasting and optimization
challenges have been studied. Yet it has taken quite some time for the
hybrid approaches to revenue management to take a foothold as an industry
practice. In this presentation we will discuss various practical hurdles
that had to be cleared for a successful implementation of a hybrid
approach, as well as some remaining challenges.
Revenue
Increase by O and D Optimization at Lufthansa
Jutta Rockmann and
Christopher Alder
Deutsche
Lufthansa AG
Many airlines operate in a network environment where passengers are buying
tickets for O&Ds which often consist of more than one leg. Still for a long time revenue optimization was only done on a leg
basis. Lufthansa switched to O&D optimization at the end of 2005.
Revenue and flight data was collected before and after cutover and a
statistical analysis was done to calculate the revenue contribution of the
new optimization logic. The results were compared to the results of PODS’
research.
Cargo Presentations ( Based
on submitted abstracts to date* )
Cargo Revenue Optimization at American Airlines
Laura
Freeland
American
Airlines Cargo
This presentation
will describe AA Cargo's overall roadmap for RM, from first adoption to the
decision to move to a new generation.
The emphasis will be the organization of AA’s next generation IT inititiave, “Project Everest,” including incremental
releases and the corresponding change management strategy.
Cargo Revenue Management: Immediate Application
Larry Hou
Delta Air
Lines Cargo
There are
many differences between passenger and cargo business: cargo demand is more
volatile, booking window is much shorter, and freight movement pattern is
more complex, etc. However, this does not mean that some revenue
management technique developed for the passenger business cannot be applied
in the cargo world. This presentation illustrates how Delta/Northwest
introduced linear programming into day to day cargo revenue management
practice. It describes a network optimization model developed using
Solver, and explains how it helped improve O and D mix for DL/NW Pacific
network, and how it repudiated some long held notion about best ways to
manage capacity. The presentation will not attempt to discuss any new
revenue management theories; rather it will be developed like a case study,
with the focus being directed to demonstrating the possibility of improving
cargo revenue management with existing technology and minimal investment.
Maximizing Network Contribution: An Insight into
Managing Complexities of Allotments and Free Sale
Demand
Sandeep Parmekar
Sabre Airline Solutions
An effective
treatise using optimization techniques and demand forecasting techniques to
solve the complex problem of managing the mix of Allotment and Free Sale
demand to maximize the network contribution. It not only deals with the
contribution aspect of the business problem but also solves the
multi-dimensional problem of managing service guarantee levels and
maintaining recur usage levels.
Variable Costs in Cargo Revenue Management: Challenges and
Opportunities
Kenneth Fuhrmann and Steven Balleby
SAS Cargo
and RevMan Consulting
This is a presentation of the challenges in identifying and collecting
quality cost data to be used in cargo revenue management. The focus will be on variable costs
elements such as handling, trucking and fuel cost, and the challenges
in identifying or calculating these.
The importance of accurate cost data will be addressed and examples
of the challenges will be presented.
The speakers represent more than 30 years of combined cargo industry
experience, with a special focus on cargo revenue management.
Modeling Multidimensionality in
Air Cargo Demand Forecasting
Jamison Graff
JDA Software Group
One
well-known aspect of the air cargo demand forecasting problem is
multidimensionality: to be useful
for operations and control, cargo demand forecasts must address such
correlated dimenesions as revenue, weight, and
volume. The modeling approaches most
commonly used in practice are naïve (independent) forecasting, use of
static conversion factors, and rate-density clustering. This study uses anonymized
cargo data from a passenger carrier to compare the results of these
techniques, and also assesses two more rigorous suggestions from the
academic literature (dynamic conversion factors and principle components
regression).
|