Professional Documents
Culture Documents
Mcgill 1999
Mcgill 1999
Transportation Science
Publication details, including instructions for authors and subscription information:
http://pubsonline.informs.org
This article may be used only for the purposes of research, teaching, and/or private study. Commercial use
or systematic downloading (by robots or other automatic processes) is prohibited without explicit Publisher
approval, unless otherwise noted. For more information, contact permissions@informs.org.
The Publisher does not warrant or guarantee the article’s accuracy, completeness, merchantability, fitness
for a particular purpose, or non-infringement. Descriptions of, or references to, products or publications, or
inclusion of an advertisement in this article, neither constitutes nor implies a guarantee, endorsement, or
support of claims made of that product, publication, or service.
© 1999 INFORMS
INFORMS is the largest professional society in the world for professionals in the fields of operations research, management
science, and analytics.
For more information on INFORMS, its publications, membership, or meetings visit http://www.informs.org
Revenue Management: Research Overview
and Prospects
Downloaded from informs.org by [129.127.145.240] on 08 October 2017, at 09:50 . For personal use only, all rights reserved.
JEFFREY I. MCGILL
Columbia University, Graduate School of Business, 412 Uris Hall, New York 10027
This survey reviews the forty-year history of research on transportation revenue management
(also known as yield management). We cover developments in forecasting, overbooking, seat
inventory control, and pricing, as they relate to revenue management, and suggest future
research directions. The survey includes a glossary of revenue management terminology and a
bibliography of over 190 references.
TABLE I
Revenue Management Research in Non-Airline Service Sectors
Sector References
Automobile rental Geraghty and Johnson (1997), Carol and Grimes (1995)
Broadcasting Cross (1998)
Cruise lines Kimes (1989), Ladany and Arbel (1991), Gallego and van Ryzin (1994)
Internet service provision Nair, Bapna, and Brine (1997), Paschalidis and Tsitiklis (1998)
Downloaded from informs.org by [129.127.145.240] on 08 October 2017, at 09:50 . For personal use only, all rights reserved.
Lodging and hospitality Rothstein (1974), Ladany (1977), Liberman and Yechiali (1977, 1978), Kimes (1989), Bitran and
Mondschein (1995), Feng and Gallego (1995), Bitran and Gilbert (1996)
Nonprofit sector Metters and Vargas (1999)
Passenger railways Kimes (1989), Strasser (1996), Ciancimino et al. (1999)
count and higher fare passengers in the same air- (1995)]. In 1999, most of the world’s major air car-
craft compartments. For example, BOAC (now Brit- riers and many smaller airlines have some level of
ish Airways) offered earlybird bookings that charged revenue management capability. Other small air-
lower fares to passengers who booked at least twen- lines and international airlines in newly deregu-
ty-one days in advance of flight departure. This in- lated markets are beginning the development pro-
novation offered the airline the potential of gaining cess.
revenue from seats that would otherwise fly empty; The success of airline revenue management was
however, it presented them with the problem of de- widely reported, and this stimulated development of
termining the number of seats that should be pro- revenue management systems for other transporta-
tected for late booking, full fare passengers. If too tion sectors and in other areas of the service sector.
few seats were protected, the airline would spill full A sample of related literature is listed in Table I.
fare passengers; if too many were protected, flights
would depart with empty seats. No simple rule, like 1.2 The Airline Revenue Management
protecting a fixed percentage of capacity, could be Problem
applied across all flights because passenger booking The objective in revenue management is to maxi-
behavior varied widely with relative fares, itinerar- mize profits; however, airline short-term costs are
ies, season, day of week, time of day, and other largely fixed, and variable costs per passenger are
factors. small; thus, in most situations, it is sufficient to seek
It was evident that effective control of discount booking policies that maximize revenues. Also, al-
seats would require detailed tracking of booking his- though there is lower risk in accepting a current
tories, expansion of information system capabilities, booking request than in waiting for later possible
and careful research and development of seat inven- bookings, booking decisions are repeated millions of
tory control rules. LITTLEWOOD (1972) of BOAC pro- times per year; therefore, a risk-neutral approach is
posed that discount fare bookings should be ac- justified. All of our discussion in this paper will
cepted as long as their revenue value exceeded the assume risk-neutral maximization of expected reve-
expected revenue of future full fare bookings. This nues as the objective.
simple, two fare, seat inventory control rule (hence- Consider the arrival of a booking request that
forth, Littlewood’s rule) marked the beginning of requires seats in an itinerary— one or more flights
what came to be called yield management and, later, departing and arriving at specified times, within a
revenue management. In North America, the begin- specific booking class, at a given fare. The funda-
ning of intensive development of revenue manage- mental revenue management decision is whether or
ment techniques dates from the launch of American not to accept or reject this booking. DURHAM (1995)
Airlines’ Super Saver fares in April of 1977, shortly reports that a large computer reservations systems
before the deregulation of U.S. domestic and inter- must handle five thousand such transactions per
national airlines. second at peak times, thus the decision must be
Over the last twenty years, development of reve- reached within milliseconds of the request’s arrival.
nue management systems has progressed from sim- Not surprisingly, no current revenue management
ple single leg control, through segment control, and system attempts full assessment of each booking
finally to origin– destination control. Each of these request in real time. Instead, precomputed aggre-
advances has required investment in more sophisti- gate control limits are set that will close the system
cated information systems, but the return on these for further bookings of specific types while leaving it
investments has been excellent [see, for example, open for others. The reservations system can quickly
Smith, Leimkuhler, and Darrow (1992), Cross determine the open or closed status of a booking
RESEARCH OVERVIEW AND PROSPECTS / 235
TABLE II
Elements of Airline Revenue Management
category and report back to an agent or customer ities of revenue management are daunting—we do
without actually evaluating the request. not have space here to discuss all of them. As is
The accept–reject decision can be restated as a always true, modeling, theoretical analyses, and im-
question of valuation: What is the expected displace- plementation rely on assuming away many of these
ment cost of closing the incremental seats in the complicating factors and approximating others. It is
requested itinerary? To maximize expected reve- important to remember that such approximations
nues, the request should be satisfied only if the fare have yielded enormous revenue benefits for airlines
value of the requested itinerary equals or exceeds and other enterprises.
the expected displacement cost. [See TALLURI and The performance of a given revenue management
VAN RYZIN (1999b) for an analysis of such displace- system depends, in large part, on the frequency and
ment cost controls.] accuracy of updates to control limits and the number
The apparent simplicity of this valuation problem of distinct booking classes that can be controlled.
is deceptive—a complete assessment must allow for The determination of suitable control limits and
all possible future realizations of the reservations characterization of their structural properties over
process that could be influenced by the availability time has been the principal focus of academic re-
of any of the seats on any of the legs in the booking. search, whereas the need for practical and imple-
Fully traced, this influence propagates across the mentable approximations to optimal limits has
entire airline network because a booking can dis- driven much of the practitioner research.
place potential bookings that will have subsequent A readable account of the practical challenges in
impacts of their own. This influence also propagates airline revenue management and related informa-
forward in time because many affected itineraries tion systems developments can be found in Sections
will terminate later than the booking being consid- 6 and 7 of JENKINS (1995). Previous research sur-
ered. Also, a booking will normally have a return veys of airline operations research and revenue
component at a later date with its own set of con- management are available in BELOBABA (1987b) and
current and downstream effects. Many other factors ETSCHMAIER and ROTHSTEIN (1974). A categoriza-
increase the complexity of the evaluation process. tion of revenue management and more general per-
Table II lists some of them. ishable asset revenue management problems is pro-
As can be seen in Table II, the practical complex- vided in WEATHERFORD and BODILY (1992).
236 / J. I. MCGILL AND G. J. VAN RYZIN
TABLE III
Airline Forecasting Research 1958 –1999
Year Reference Year Reference
Graduate theses often contain thorough reviews of influence forecasts have on the booking limits that
past research literature. We are aware of five mas- determine airline profits. Not surprisingly, publica-
ter’s theses and twelve doctoral dissertations in rev- tion of approaches to airline forecasting are concur-
enue management or related areas. The master’s rent with the literature on overbooking because
theses are those of SA (1987), WILLIAMSON (1988), overbooking calculations depend on predictions of
BOHUTINSKY (1990), SUN (1992), SHAYKEVICH ultimate demand, cancellations, and no-shows. Ta-
(1994). The doctoral dissertations are those of ROTH- ble III lists references from the bibliography.
STEIN (1968), BELOBABA (1987a), MCGILL (1989), Unfortunately, the disaggregate forecasting re-
LEE (1990), WONG (1990), WEATHERFORD (1991), quired for both overbooking and revenue manage-
WILLIAMSON (1992), CHATWIN (1993), NOONAN ment is extremely difficult. The list of passenger
(1993), BOTIMER (1994), LI (1994) and ZHAO (1999). behaviors and other complicating factors contained
In the following sections, we review revenue man- in Table II should make clear the reasons for this
agement research in four key areas—forecasting, difficulty. Simply accounting for the effects of price
overbooking, seat inventory control, and pricing. volatility is a significant challenge in itself—in 1989,
Each section contains a listing of references that are there were a reported 30,000 daily price changes in
relevant to the topic of the subsection and included the U.S. domestic airline industry alone [see Wil-
in the bibliography. For completeness, we have in- liamson (1992, p. 42)].
cluded published articles, conference proceedings, In the following subsections, we review models for
working papers, industrial technical reports, and demand distributions, models for arrival processes,
graduate theses. We cannot discuss all of the listed uncensoring of demand data, aggregate versus dis-
works, so limit ourselves to reviewing a subset that aggregate forecasting, and current practices.
are illustrative of the type of work that has been
done. 2.1 Models for Demand Distributions
We make no claim to having identified all revenue Early descriptions of statistical models of passenger
management publications and regret any omissions. booking, cancellation, and no-show behavior di-
The authors would be delighted to receive copies of rected toward overbooking calculations can be found
or references to any work that has not been in- in BECKMANN and BOBKOWSKI (1958). In that paper,
cluded. the authors compare Poisson, Negative Binomial,
and Gamma models of total passenger arrivals and
offer evidence of a reasonable fit for the Gamma
2. FORECASTING
distribution to airline data. BECKMANN (1958) uses
FORECASTING IS AN important component of plan- Gamma distributions to model the components of
ning in any enterprise; but it is particularly critical show-ups and develops an approximate optimality
in airline revenue management because of the direct condition for the overbooking level. TAYLOR (1962)
RESEARCH OVERVIEW AND PROSPECTS / 237
TABLE IV
Stochastic Process Models for Airline Arrivals 1958 –1999
Arrivals Process Reference
Homogeneous, nonhomogeneous and/or Gerchak, Parlar, and Yee (1985), Alstrup et al. (1986), Lee (1990), Virtamo and Aalto
compound Poisson processes (1991), Stone and Diamond (1992), Sun (1992), Lee and Hersh (1993), Weatherford,
Bodily, and Pfeifer (1993), Gallego and van Ryzin (1994), Lautenbacher and Stidham
(1999), Subramanian, Lautenbacher, and Stidham (1999), Zhao (1999)
Downloaded from informs.org by [129.127.145.240] on 08 October 2017, at 09:50 . For personal use only, all rights reserved.
Stuttering Poisson processes Beckmann and Bobkowski (1958), Rothstein (1968, 1971a)
Censored Poisson processes Lee (1990)
General point processes Gallego and van Ryzin (1994, 1997)
determines empirical probability-generating func- distribution; however, both the homogeneous and
tions for booking behaviors that determine show- nonhomogeneous Poisson processes lead to Poisson
ups. Allowance is made for single and batch book- cumulative arrival distributions. This is problematic
ings, cancellations, and go-shows. The generating for total demand modeling because the coefficient of
function is then used to estimate the parameters of variation of the Poisson distribution is the reciprocal
a distribution for final show-ups. LYLE (1970) mod- of the square root of the mean. Thus, for example,
els demand as composed of a Gamma systematic the coefficient of variation for a booking class with
component with Poisson random errors. This model mean demand of 100 would be 0.10 —much lower
leads to a negative binomial distribution for total than the 0.25 to over 1.0 encountered in practice.
demand, as in Beckmann and Bobkowski (1958), Fortunately, cumulative arrivals from compound
which is then truncated for demand censorship. Poisson processes can provide a reasonable fit to the
MARTINEZ and SANCHEZ (1970) give a detailed anal- coefficients of variation of real world arrival data.
ysis of booking and cancellation data from Iberia For example, the stuttering Poisson process used by
Airlines and discuss a convolution methodology sim- ROTHSTEIN (1968, 1971a) and Beckmann and
ilar, in part, to Taylor’s approach for obtaining em- Bobkowski (1958) is a compound process that allows
pirical demand and cancellation probability distri- for batch arrivals at each occurrence of a Poisson
butions. arrival event and achieves more realistic variances.
Empirical studies have shown that the normal
probability distribution gives a good continuous ap-
proximation to aggregate airline demand distribu- 2.3 Uncensoring Demand Data
tions [see, for example, Belobaba (1987a) and Data contained in historical booking records are
SHLIFER and VARDI (1975)]. Given the central limit censored by the presence of booking and capacity
theorem and the role of the normal distribution as limits on past demands. SWAN (1990) addresses the
the limiting distribution for both binomial and Pois- downward bias of censoring on late booking data
son distributions, this is not surprising. However, and suggests simple statistical remedial measures.
many researchers have pointed out that the normal An earlier spill formula developed by Swan has been
distribution becomes increasingly inappropriate at used for many years by practitioners to unconstrain
greater levels of disaggregation. Section 2.4 dis- demand. Lee (1990) presents a detailed stochastic
cusses this more fully. model of passenger arrivals based on a censored
Poisson process and develops maximum likelihood
2.2 Models for Arrival Processes methods for estimating the parameters of these
Many of the references cited above use a model for models. MCGILL (1995) develops a multivariate mul-
the stochastic arrival process of individual booking tiple regression methodology for removing the ef-
requests to construct distributions of total flight de- fects of censorship in multiple booking classes, and
mand. In other work that seeks dynamic booking describes a bootstrapping approach to testing for
rules, specification of the arrivals process is an es- correlations between fare class demands. A treat-
sential starting point. Examples of stochastic pro- ment of censored data in general inventory contexts
cesses that have been used to model arrivals and the can be found in NAHMIAS (1994). Smith, Leimkuhler,
related references are listed in Table IV. and Darrow (1992) mention experiments with cen-
The use of the Poisson process, when appropriate, sored regression at American Airlines, but, as far as
is useful in dynamic treatments because of the we know, there have been no follow-up reports on
memoryless property of the exponential interarrival these experiments.
238 / J. I. MCGILL AND G. J. VAN RYZIN
2.4 Aggregate and Disaggregate Forecasting GALLEGO (1996) presents a deterministic model of
Traditional regression techniques for aggregate demand behavior under price changes that incorpo-
airline forecasting are described in a book by TANEJA rates diversion and recapture of passengers in dif-
(1978). A more recent account of regression experi- ferent booking classes. WEATHERFORD, BODILY, and
ments with airline data can be found in the master’s PFEIFER (1993) incorporate diversion in a stochastic
thesis of Sa (1987). Sa concludes that use of regres- model of booking arrivals for two classes. They use
sion techniques can improve the performance of rev- beta functions for the intensity functions of nonho-
Downloaded from informs.org by [129.127.145.240] on 08 October 2017, at 09:50 . For personal use only, all rights reserved.
In this setting, an early, nondynamic optimization of overbooking for a two-class, nonstop flight and
model for overbooking is that of Beckmann (1958). describe computational experience with the ap-
Statistical models of various levels of sophistication proach at Scandinavian Airlines. A DP analysis sim-
are described by THOMPSON (1961), Taylor (1962), ilar to Rothstein’s but developed for the hotel/motel
ROTHSTEIN and STONE (1967), Martinez and industry and extended to two fare classes is de-
Sanchez (1970), and Littlewood (1972). An overbook- scribed in LADANY (1976, 1977) and LADANY and
ing model extended to allow for two fare classes and ARBEL (1991). A control-limit type structural solu-
a two-leg flight is described by Shlifer and Vardi tion to the (one class) hotel overbooking problem is
(1975). In part of his Ph.D. dissertation, Belobaba described in LIBERMAN and YECHIALI (1977, 1978).
(1987a) discusses the problem of overbooking in The dissertation of Chatwin (1993) deals exclu-
multiple fare classes and suggests a heuristic ap- sively with the overbooking problem and provides a
proach to solving the problem. BRUMELLE and number of new structural results. Chatwin’s paper
MCGILL (1989) present a static formulation of the in this issue and CHATWIN (1999b) contain excerpts
overbooking problem and show that it is a special and refinements based on his dissertation. KARAES-
case of a general model of the two fare class seat MAN and VAN RYZIN (1998) address the problem of
allocation problem. None of these studies allow for jointly setting overbooking levels when there are
the dynamics of the passenger cancellation and res- multiple inventory classes that can serve as substi-
ervation process subsequent to the overbooking de- tutes for one another; for example, first class and
cision. coach travel service, or compact and full size rental
A number of researchers have developed dynamic cars.
optimization approaches to the airline overbooking
problem and the related problem in the hotel/motel
industry. The usual objective in these formulations 4. SEAT INVENTORY CONTROL
is to determine a booking limit for each time period THE PROBLEM OF seat inventory control across mul-
before flight departure that maximizes expected rev- tiple fare classes has been studied by many re-
enue, where allowance is made for the dynamics of searchers since 1972. There is a progression from
cancellations and reservations in subsequent time Littlewood’s rule for two fare classes, to expected
periods and for penalties for oversold seats. KOSTEN marginal seat revenue (EMSR) control for multiple
(1960) develops a continuous time approach to this classes, to optimal booking limits for single-leg
problem, but this approach requires solution of a set flights, to segment control and, more recently, to
of simultaneous differential equations that make ODF control. Each step in this progression can be
implementation impractical. Rothstein (1968), in his viewed as a refinement of the displacement cost
Ph.D. thesis, describes the first dynamic program- valuation discussed in Section 1.2. We review re-
ming (DP) model for overbooking and reviews the search on the single leg and network problems sep-
results of test runs of the model at American Air- arately below. Table VI list publications on single-
lines. ALSTRUP et al. (1986) describe a DP treatment leg control.
240 / J. I. MCGILL AND G. J. VAN RYZIN
TABLE VI
Single-leg Seat Inventory Control 1972–1999
Year Reference Year Reference
4.1 Single-Leg Seat Inventory Control nish evidence that EMSR provides reasonable ap-
Most early seat inventory control research re- proximations with typical airline demand distribu-
quired most or all of the following simplifying as- tions. ROBINSON (1995) shows that, for more general
sumptions: 1) sequential booking classes; 2) low-be- demand distributions, the EMSR method can pro-
fore-high fare booking arrival pattern; 3) statistical duce arbitrarily poor results. A later refinement of
independence of demands between booking classes; EMSR, called EMSRb, apparently produces better
4) no cancellations or no-shows (hence, no overbook- approximations to optimal booking limits and has
ing); 5) single flight leg with no consideration of been widely implemented. For a description of EM-
network effects; and, 6) no batch booking. SRb, see VAN RYZIN and MCGILL (1998).
Littlewood’s rule can be viewed as an early expres- Methods for obtaining optimal booking limits for
sion of the displacement cost rule for two booking single-leg flights are provided in McGill (1989),
classes under all six assumptions. Derivations of CURRY (1990), Wollmer (1992), BRUMELLE and
Littlewood’s rule are given by BHATIA and PAREKH MCGILL (1993). All of these results require assump-
(1973) and, by a different method, RICHTER (1982). tions 1 through 6. Curry also proposes an approxi-
MAYER (1976) describes a simulation study of the mation for the network problem, a relaxation of as-
performance of Littlewood’s rule and offers evidence sumption 5. Brumelle and McGill show that, under
that, if it is used more than once before flight depar- all six assumptions, the seat inventory control prob-
ture, the rule can perform as well as a more complex lem is a monotone optimal stopping problem (CHOW,
DP model in which the low-before-high fare arrival ROBBINS, and SIEGMUND, 1971), and, consequently,
assumption is relaxed. He also suggests that the that static control limit policies are optimal over the
seat allotment and overbooking analyses can be class of all control policies for this restricted prob-
done independently with little revenue loss. TITZE lem, including dynamic ones. They also characterize
and GRIESSHABER (1983) offer additional simulation optimal booking limits with a set of probability con-
evidence that Littlewood’s rule is robust to modest ditions that are the same as the EMSR conditions
departures from the low-before-high fare assump- for the first two fare classes but include joint prob-
tion. abilities that are lacking from the EMSR method for
Belobaba (1987a) extends Littlewood’s rule to additional fare classes. Robinson (1995) generalizes
multiple fare classes and introduces the term EMSR the probability conditions to relax the low-before-
for the general approach. The EMSR method does high fare assumption.
not produce optimal booking limits except in the Van Ryzin and McGill (1998) show that, when
two-fare case; however, it is particularly easy to demand distributions are stationary across multiple
implement. McGill (1989) and WOLLMER (1992) fur- flights, the optimality conditions of Brumelle and
RESEARCH OVERVIEW AND PROSPECTS / 241
McGill (1993) can be exploited in an adaptive sto- lems. Both address nonmonotonicity of bid-prices
chastic approximation method that requires no sep- (see below) under batch booking.
arate forecasting or uncensoring of demands and no
direct reoptimization of protection levels. 4.2 Segment and Origin–Destination Control
LI and OUM (1998) provide a preliminary analysis Since the 1980s, network effects in revenue man-
of the seat allocation problem for two airlines com- agement have become increasingly significant be-
peting with identical aircraft and fares on a single- cause the expansion of hub-and-spoke networks has
Downloaded from informs.org by [129.127.145.240] on 08 October 2017, at 09:50 . For personal use only, all rights reserved.
leg route. Their game theoretic formulation demon- dramatically increased the number of passenger
strates the existence of equilibrium seat allocations itineraries that involve connections to different
under certain assumptions on the demand distribu- flights. It has been recognized for some time that
tions. revenue management should account for these net-
Belobaba (1987a) proposes an optimality condi- work effects but that this cannot be accomplished
tion for the two fare class problem that allows for the effectively with single-leg control. Progress in this
possibility of upgrades from the lower fare class to area has been impeded not by the lack of approaches
the higher in the event that the lower class is closed to network inventory control so much as by the
for bookings. PFEIFER (1989) presents a proof of the limitations of older reservations systems. Even if
result. McGill (1989) and BRUMELLE et al. (1990) optimal solutions existed for the many hundreds of
show that a variant of Littlewood’s rule is optimal itineraries that traversed a single leg, those solu-
for the two fare case when discount and full fare tions had to be mapped into a much smaller number
demands are statistically dependent, subject to a of controllable booking classes. This situation is
mild monotonicity assumption on the nature of the changing—the most advanced reservations systems
dependency. The upgrade result is obtained as a are now capable of incorporating network informa-
corollary. They further show that, with the same tion, and the emergence of seamless availability will
monotonicity assumption, this problem is a mono- allow for much finer control of seat availability.
tone optimal stopping problem. We review, below, four approaches that have been
No optimal static control limit policies exist when taken to network revenue management. Table VII
any assumption other than the low-before-high or contains relevant references.
independent demand assumptions is relaxed. DP
treatments of the single leg problem are presented 4.2.1 Mathematical Programming Formulations
in Mayer (1976), LADANY and BEDI (1977), STONE GLOVER et al. (1982) describe a minimum cost
and DIAMOND (1992), Sun (1992), LEE and HERSH network flow formulation for the passenger O–D
(1993), Shaykevich (1994), YOUNG and VAN SLYKE problem and an implementation at Frontier airlines.
(1994), BRUMELLE and WALCZAK (1997, 1998b), Passenger demands are assumed deterministic in
ZHAO and ZHENG (1998a), LAUTENBACHER and this formulation, so the focus of the formulation is on
STIDHAM (1999), SUBRAMANIAN, LAUTENBACHER, network effects rather than the stochastic elements.
and STIDHAM (1999), and Zhao (1999). Space does DROR, TRUDEAU, and LADANY (1988) propose a sim-
not permit a review of all of these contributions, and ilar deterministic network minimum cost flow for-
there is some overlap among them. GERCHAK, PAR- mulation that allows for cancellations as determin-
LAR, and YEE (1985) contains the earliest dynamic istic losses on arcs in the network. Wong (1990), in
structural results for this type of problem, but this his Ph.D. dissertation, develops a network formula-
work is often overlooked because it deals with opti- tion for the single fare class, multiple itinerary prob-
mal discounting of unsold bagels—a classic example lem and extends it to approximations for the multi-
of the generality of revenue management problems. ple booking class case. The single fare case and some
Ladany and Bedi (1977) and HERSH and LADANY comparisons of different cabin assignment methods
(1978) deal with a two-leg flight; however, we in- are discussed in WONG, KOPPELMAN, and DASKIN
clude it as a single-leg model here because their (1993). WOLLMER (1986c) proposes a linear program-
models assume no boarding of passengers at the ming (LP) network formulation that allows for sto-
intermediate stop. chastic demand by incorporating expected marginal
Lee and Hersh (1993) report a generalization of seat values as coefficients in the objective function.
monotonicity results to the batch booking case; how- Each ODF generates a set of zero– one decision vari-
ever, these results are shown to be in error in Bru- ables for each flight, with a corresponding set of
melle and Walczak (1997). Both KLEYWEGT and PA- monotonically decreasing objective function coeffi-
PASTAVROU (1998) and VAN SLYKE and YOUNG cients determined by the marginal expected values.
(1994) characterize seat inventory control problems Wollmer shows that this formulation can be con-
as special cases of certain stochastic knapsack prob- verted to a minimum cost network formulation for
242 / J. I. MCGILL AND G. J. VAN RYZIN
TABLE VII
Origin–Destination Control 1982–1999
Year Reference Year Reference
approximate displacement cost, or bid-price, for that management practice, is in a unique competitive
itinerary. A booking request for a passenger itiner- equilibrium. BORENSTEIN and ROSE (1994) provide
ary is rejected if the bid price exceeds the fare for the empirical tests of airline competition and its rela-
itinerary, and is accepted otherwise. Thus, this ap- tionship to the degree of price dispersion observed in
proach attempts to directly incorporate the esti- fares. They also address the question of whether
mated displacement cost as a cutoff value for accept- price dispersion is the result of monopoly or second-
able fares. degree price discrimination. OUM, ZHANG, and
The disadvantages and advantages of this ap- ZHANG (1993) and OUM (1995) deal with aspects of
proach are discussed in Williamson (1992) and pricing in deregulated airline markets and the in-
Smith and Penn (1988). A more recent account from fluence of code-sharing agreements on international
a managerial viewpoint can be found in Vinod fares.
(1995). Despite potential theoretical drawbacks to It is now common for airline practitioners to view
the bid-price approach, it has a very convincing ad- pricing as part of the revenue management process.
vantage—it replaces multiple booking limits and The reason for this is clear—the existence of differ-
complex nesting schemes with a single bid-price ential pricing for airline seats is the starting point
value for each flight leg and a simple rule for reject- for revenue management, and price is generally the
ing or accepting itinerary requests. most important determinant of passenger demand
behavior. There is also a natural duality between
5. PRICING price and seat allocation decisions, as pointed out in
THERE IS AN EXTENSIVE literature on airline pricing GALLEGO and VAN RYZIN (1997). If price is viewed as
from an economic perspective that we cannot review a variable that can be controlled on a continuous
here. For the most part, that literature deals with basis, a booking class can be shut down by raising
pricing and price competition at an industry level the price sufficiently high. Also, when there are
rather than the operational, revenue management many booking classes available, shutting down a
decision level. This literature is nonetheless rele- booking class can be viewed as changing the price
vant to strategic and marketing decisions that are structure faced by the customer. Treatments of rev-
important in revenue management. We discuss here enue management as a dynamic pricing problem can
a few examples. Table VIII contains a listing of some be found in Ladany and Arbel (1991), GALLEGO and
relevant articles. VAN RYZIN (1994, 1997), FENG and GALLEGO (1995),
KRETSCH (1995) describes fare management pol- and YOU (1999).
icy from a managerial standpoint. DANA (1998) ex- The single-leg revenue management problem with
plains price dispersion as a competitive market re- two fare classes is essentially equivalent to the
action to consumers’ uncertainty about travel and much-studied single period inventory or newsvendor
the risk of rationing due to capacity constraints. The problem. Thus, research on pricing in the single-
work explains how airline pricing, which looks like period inventory setting has some relevance to rev-
classic second-degree monopoly price discrimina- enue management. LAU and LAU (1988) provide a
tion, can, in fact, be the result of a perfectly compet- joint pricing/inventory control analysis for the news-
itive market. DANA (1996) shows that a firm that vendor problem. A more general treatment of price,
offers high and low prices and then rations the ca- capacity, and technology decisions for profit maxi-
pacity sold at the low price, as is done in yield mizing firms is provided in GAIMON (1988).
244 / J. I. MCGILL AND G. J. VAN RYZIN
There has been very little published research on diction of cancellation and no-show behavior in dif-
joint capacity allocation/pricing decisions in the rev- ferent passenger categories. Smith, Leimkuhler,
enue management context. Botimer (1994), in his and Darrow (1992) discuss the potential of discrete
doctoral dissertation, uses a model for unrestricted choice modeling [see, for example, BEN-AKIVA and
demand for airline seats to derive the demand func- LERMAN (1985), Lee (1990)]. Much work remains to
tions for restricted fare products by incorporating be done—the potential benefits of sharper forecasts
the cost of restrictions to passengers. He also con- certainly justify substantial investments in forecast-
Downloaded from informs.org by [129.127.145.240] on 08 October 2017, at 09:50 . For personal use only, all rights reserved.
siders modifications to allow for passenger diver- ing methodology and market analysis.
sions. Li (1994) proves that it is optimal to offer
relatively small numbers of fare classes (three in one 6.2 Implementable Dynamic Programming
case, four in another) when the restrictions applied Approaches
to differentiate the fare classes satisfy certain regu- Dynamic formulations of the revenue manage-
larity conditions. WEATHERFORD (1994) presents a ment problem are required to properly model real-
formulation of the simultaneous pricing/allocation world factors like cancellations, overbooking, batch
decision that assumes normally distributed de- bookings, and interspersed arrivals. Unfortunately,
mands, and models mean demand as a linear func- DP formulations, particularly stochastic ones, are
tion of price. The corresponding expressions for total well known for their unmanageable growth in size
revenue as a function of both price and allocation are when real-world implementations are attempted.
extremely complex, and no structural results are Usually, the only hope for dynamic optimization in
obtained. Computational results from a variety of these settings lies in identification and exploitation
test problems are supplied along with general com- of structural properties of optimal or near optimal
ments on when inclusion of prices as decision vari- solutions. Knowledge that an optimal solution must
ables justifies the greater computational effort re- be of a control-limit type or be represented by a
quired. monotonic threshold curve can be invaluable in de-
Recent work by Gallego (1996) uses a simple de- velopment of implementable systems. The existing
terministic model to examine pricing and market literature has already identified such structures in
segmentation decisions. His model takes into ac- special cases of the revenue management problem;
count both demand diversion and demand recap- however, there are difficult areas still requiring
ture. He gives precise conditions to guarantee the work. The inclusion of batch bookings, which influ-
optimality of low to high pricing and lower and up- ences the regularity properties required for existing
per bounds on the optimal revenue. structural results, is a particularly critical area for
research— batch bookings are common in airline
6. CONCLUSION AND RESEARCH PROSPECTS
reservations. The work of Young and Van Slyke
(1994), Brumelle and Walczak (1997), and Kleywegt
RESEARCH AND DEVELOPMENT of revenue manage- and Papastavrou (1998) provide reference points for
ment systems is far from over. We close our survey this line of research.
with some suggested directions that future research It seems unrealistic at this time to imagine that
may take in four areas—forecasting, dynamic pro- exact DP algorithms could be used for real-time
gramming, ODF revenue management, and systems revenue management in the airline industry. How-
integration. ever, DP can be used in two important ways: first, on
an exception basis for flight or network segments
6.1 Forecasting that are identified as being particularly critical, and
It is difficult to be optimistic about breakthroughs second, as a calibration tool for checking the perfor-
in airline disaggregate forecasting because of the mance of less accurate but more efficient solution
slow progress of effective forecasting technology in methods.
less complex areas elsewhere in industry. Nonethe- Recently developed approximation methods for
less, a need for reliable airline demand forecasts at DP and stochastic programming may be useful in
increasingly disaggregate levels will parallel in- revenue management. Good references for these ap-
creases in sophistication of ODF revenue manage- proximation methods are the books by BERTSEKAS
ment systems. The most promising direction for im- and TSITSIKLIS (1996) and BIRGE and LOUVEAUX
provement of airline forecast accuracy is in detailed (1997).
empirical studies of the behaviors of different pas-
senger types in response to changes in fare product 6.3 ODF Revenue Management
offerings. Tracking of individual behavior of passen- Bid-price methods appear to be the most promis-
gers who fly frequently could lead to improved pre- ing method for the ODF revenue management prob-
RESEARCH OVERVIEW AND PROSPECTS / 245
lem because they are simple to implement, and airline planning; in particular, pricing, fare product
there is empirical evidence that they lead to in- design, fleet assignment, and route planning. For
creased revenues for airlines. However, at present, example, although pricing and fare restriction deci-
it is not known how close the total dual prices are to sions occur at a much slower rate than seat inven-
the true expected displacement costs of booked seats tory control decisions, the revenue impact of a pric-
in an itinerary in a large airline network. One sur- ing decision is ultimately determined at the seat
prising finding is how well the deterministic dual allocation level, thus there is a clear need for inte-
Downloaded from informs.org by [129.127.145.240] on 08 October 2017, at 09:50 . For personal use only, all rights reserved.
price approach appears to work despite its obvious gration of these two decision processes. Also, as dis-
drawbacks. Talluri and van Ryzin (1999) provide a cussed above, dual prices at the seat inventory level
number of findings on the accuracy (and potential are relevant to fleet assignment decisions. They are
inaccuracies) of the bid-price technique and furnish also relevant to longer-term route planning and
some simple counterexamples to its optimality. market development decisions. [See, for example,
There is room for further analysis of this approach. DOBSON and LEDERER (1993).] Each of these differ-
A particularly interesting question is how bid ent decision levels present significant theoretical
prices can be applied to the common batch and group and computational challenges. Mathematical pro-
booking problems. For example, even if we assume gramming and other optimization techniques have
that the sum of the dual prices along an itinerary been developed for each of them, but these formula-
are approximately equal to the displacement cost for tions typically treat the problems in isolation.
one seat, it is unlikely that the displacement cost of There is emerging the prospect of an integrated
a batch booking for the same itinerary will be a hierarchy of decision systems in which decisions and
simple multiple of the individual displacement costs. information obtained at one level are smoothly
Assuming that the suitability of bid-prices can be available to other levels. There are a number of
verified, there is an interesting computational chal- interesting associated research questions. Is there a
lenge in increasing the speed with which the dual natural problem/subproblem structure for such an
prices can be updated. Is it possible to accomplish integrated planning system that can exploit decom-
near real time reoptimizations of the LP model so position techniques from network optimization and
that dual prices reflect the latest information on mathematical programming? What information
fares and bookings in the network? should be exchanged? How frequently? What is the
As mentioned in Williamson (1992), one useful reliability and stability of such a system? There are
by-product of the dual-price calculation is the iden- other areas of interest to revenue management prac-
tification of flights with unusually high dual prices. titioners and researchers; for example, the long-
Such flights correspond to bottlenecks in the airline term implications of ticket sales or auctioning
network. There is a direct analogy here with the through the Internet.
concept of bottleneck in the manufacturing indus- It is clear that revenue management will continue
try— bottleneck flights can constrain flows through to generate interesting applications and research
major segments of the airline network and should be questions for years to come. We hope that this sur-
managed intensely. In the industrial setting, the vey will serve as a stimulus for future research in
recommended treatment for bottlenecks is: 1) to ex- this challenging and important area.
pand their capacity if possible, and 2) ensure maxi-
mum use of the bottleneck by placing buffer inven- APPENDIX
tory in front (and sometimes behind) the bottleneck.
In the airline setting, the bottleneck capacity can be REVENUE MANAGEMENT GLOSSARY
increased by assignment of more or larger aircraft,
WE PROVIDE HERE a glossary of the sometimes-con-
and the buffer inventory can be achieved with over-
fusing terminology of revenue management. Our
booking. The interesting research question is how to
aim in supplying a separate glossary is to avoid
systematically incorporate bottleneck dual price in-
needless definitions for readers familiar with reve-
formation into overbooking and fleet assignment
nue management while assisting others who are
processes. BERGE and HOPPERSTAD (1993) offer evi-
new to the field. Many of the terms described here
dence that there is significant revenue potential in
have different meanings in more general contexts
dynamic fleet assignment.
but are presented here with their usual meanings in
6.4 Integration with Other Planning revenue management.
Functions Aggregation of demand: The level of summariza-
Revenue management decisions are highly inter- tion of passenger demand data. The trend has been
dependent with decisions made in other key areas of toward increasing levels of disaggregation in seat
246 / J. I. MCGILL AND G. J. VAN RYZIN
inventory optimization; however, pricing, forecast- sively to bookings in that fare class. This method
ing, and booking control processes often operate at simplifies reservations control but is clearly unde-
different levels of aggregation. Possible dimensions sirable from a revenue standpoint because seats
for disaggregation include: market, season, month, could fly empty in a discount bucket even if there is
week, section of week (e.g., midweek versus week- higher fare demand available to fill them. Second,
end), day of week, time of day, flight number, book- buckets also refer to clusters of different fare classes
ing class, fare, flight leg, segment, and itinerary. or ODFs that are grouped together for control pur-
Downloaded from informs.org by [129.127.145.240] on 08 October 2017, at 09:50 . For personal use only, all rights reserved.
Arrival pattern: The pattern of arrivals of booking poses in a virtual nesting system. A single booking
requests. In the airline context, some possible ar- limit is set for all classes in the bucket or lower value
rival patterns are: sequential booking classes, low- buckets.
before-high fares, or interspersed arrivals. Bulk arrival: See batch booking.
Batch booking: (also multiple booking, or bulk ar- Bumping: See denied boarding.
rival) A booking request that arrives through nor-
mal reservation channels for two or more seats to be Cabin: The physical compartment of an aircraft
booked for the same itinerary. Contrast with group containing a particular type of seating. For example,
bookings. an aircraft may be equipped with a first class cabin
and a coach cabin, each with different seating and
Bid price: A net value (bid-price) for an incremental separated by a partition. Multiple fare classes are
seat on a particular flight leg in the airline network. usually available in each cabin of the aircraft.
Also referred to as minimum acceptable fare, hurdle
price, probabilistic shadow price, displacement cost, Cancellations: Returns or changes in bookings
or probabilistic dual cost. that occur early enough in the booking period to
permit subsequent rebooking through the reserva-
Bid price control: A method of network seat in-
tions system.
ventory control that assesses the value of an ODF
itinerary as the sum of the bid-prices assigned to Censorship of demand data: Typically, no record
individual legs in the itinerary. Typically, an ODF can be kept of booking requests that occur after a
request is accepted if its fare exceeds the total bid- fare class is closed down. Thus, in booking histories,
prices. Also referred to as continuous nesting. the number of flights on which demand reached a
booking limit can be determined but not the amount
Booking class: A category of bookings that share
by which demand exceeded the limit. Formally, this
common features (e.g., similar revenue values or
condition is known as multiple Type I censorship of
restrictions) and are controlled as one class. This
term is often used interchangeably with fare class or the data—the censorship points are known (booking
bucket. limits), but may vary between observations (flights).
Booking limit: The maximum number of seats that Code-sharing: It is now relatively common for
can be sold to a particular booking class. In nested small groups of domestic and international airlines
booking systems, booking limits apply to the total to form alliances in which the members interlist
number of seats sold to a particular booking class some or all of their flights. The two character airline
and any lower fare booking classes. designator code from one airline is applied to flight
numbers of other alliance airlines so that there is an
Booking policy: A booking policy is a set of rules apparent expansion of participating airlines’ net-
that specify at any point during the booking process works.
whether a booking class should be open. In general,
such policies may depend on the pattern of prior Coefficient of variation: The standard deviation
demands or be randomized in some manner and expressed as a proportion of the mean of a probabil-
must be generated dynamically as the booking pro- ity or relative frequency distribution. Thus a de-
cess unfolds for each flight. In some circumstances, mand distribution with mean demand 100 and stan-
optimal or approximately optimal booking policies dard deviation 40 would exhibit a coefficient of
can be defined by a set of fixed protection levels or variation of 0.40. Airline demand data typically dis-
threshold curves. play coefficients of variation in the range 0.25 to
over 1.0, depending on the level of aggregation of the
Buckets: This term is used in two related ways.
data.
First, in older reservations systems, seats for differ-
ent fare classes or groups of classes are pre-assigned Connectivity (in reservations systems): The de-
to distinct buckets. These seats are available exclu- gree to which the elements of the reservations sys-
RESEARCH OVERVIEW AND PROSPECTS / 247
tem are electronically interconnected. See seamless cludes all future revenues that may be lost if the
availability. booking is accepted. Taken to the extreme, these
include the revenue value of potential displaced fu-
Continuous nesting: (see bid-price control)
ture bookings anywhere in the airline network and
Controllable booking classes: All early reserva- goodwill costs from those displacements. Assess-
tions systems and many existing systems offer only ment of the costs and probabilities of such displace-
a small number of distinct booking categories (five to ments should allow for the dynamics of cancellations
Downloaded from informs.org by [129.127.145.240] on 08 October 2017, at 09:50 . For personal use only, all rights reserved.
ten) that can actually be controlled at booking out- and overbooking and the expected costs of oversold
lets. Thus, regardless of the number of booking conditions.
classes or distinct passenger itineraries that can be
Diversion: The booking of a customer at a fare level
handled by the revenue management optimization
lower than one they would have been prepared to
process, the controls in such systems can only be
pay. This occurs, for example, when a business trav-
applied to a small number of aggregate booking
eler has sufficient advance notice of a trip to book in
classes or buckets.
a discount class intended primarily for leisure trav-
Control limit policy: A structural solution that elers. Restrictions are designed to inhibit such diver-
specifies an upper bound (limit) on the number of sion.
seats sold in each fare class (or collection of fare
Dual prices (also shadow prices): The marginal
classes) for each time before flight departure.
value of one additional unit of a constrained re-
CRS: Computer reservations system. source, as determined by a mathematical program-
ming solution to an optimization model. Dual prices
Defections: It can occur that a confirmed passenger
are one source of the marginal seat values used in
who shows up for a flight switches to a flight with
bid-price control.
another airline (usually because of a delay in the
original flight departure). Defections constitute a Dynamic models: Models that take into account
relatively small component of lost passengers and future possible booking decisions in assessing cur-
are normally counted as part of no-shows. However, rent decisions. Most revenue management problems
they are distinct from no-shows, and any attempt to are properly modeled as dynamic programming
predict their occurrence requires an estimation of problems.
the probability distribution for departure delays.
Expected marginal seat revenue (EMSR): The
Demand distribution: An assignment of probabil- expected revenue of an incremental seat if held
ities (probability distribution) to each possible level open. This is a similar concept to that of bid-price
of demand for a flight or booking class. A prelimi- but generally used in a simpler context.
nary estimate of such a demand distribution can be
Expected revenue: The statistical expected reve-
obtained by calculating the proportion of each de-
nue; that is, the sum of possible revenue values
mand level seen on comparable past flights; i.e., a
weighted by their probabilities of occurrence.
relative frequency distribution.
Fare basis code: An alphanumeric encryption of
Demand factor: The ratio of demand over capacity
the conditions and restrictions associated with a
for a flight or booking class. (Contrast with load
given fare. Usually several fare basis codes are con-
factor.)
tained in a single fare class.
Denied boarding: Turning away ticketed passen-
Fare class: A category of booking with a (relatively)
gers when more passengers show-up at flight time
common fare. Typical labels for such classes [see
than there are seats available on the flight, usually
Vinod (1995)] are: F for first class (separate com-
as a result of overbooking practices. Denied board-
partment); J for business class, U for business class
ings can be either voluntary, when passengers ac-
frequent flyer redemption (often separate compart-
cept compensation for waiting for a later flight, or
ment); Y for full fare coach; B, M, Q, V for progres-
involuntary, when an insufficient number of pas-
sively more discounted coach bookings; and T for
sengers agree to accept compensation. In the latter
frequent flyer coach cabin redemptions. Often other
case, the airline will be required to provide compen-
fare products (such as travel agent or company trav-
sation in a form mandated by civil aviation law.
elers) are categorized under one of these designa-
Disaggregate: See aggregation of demand. tions for control purposes.
Displacement cost: In revenue management, the Fare product: The full set of attributes associated
displacement (or opportunity) cost of a booking in- with a specific transportation service. The set in-
248 / J. I. MCGILL AND G. J. VAN RYZIN
cludes the fare as well as any restrictions or benefits Hub-and-spoke network: A configuration of an
that apply to that service at that fare. airline’s network around one or more major hubs
that serve as switching points in passengers’ itiner-
Fences: See restrictions.
aries to spokes connected to smaller centers. The
Fleet assignment: Most airlines have a variety of proliferation of these networks has greatly increased
aircraft types and sizes in their fleets. The fleet the number of passenger itineraries that include
assignment process attempts to allocate aircraft to connections to different flights.
Downloaded from informs.org by [129.127.145.240] on 08 October 2017, at 09:50 . For personal use only, all rights reserved.
routes in the airline network to maximize contribu- Hub bank: A collection of inbound and outbound
tion to profit. There are strong potential linkages flights that are scheduled to arrive or depart within
between fleet assignment and revenue management a time span that enables convenient passenger con-
processes because aircraft assignments determine nections among flights. An airline hub will typically
leg capacities in the network. operate with several hub banks throughout the day.
Flight leg: A section of a flight involving a single Incremental seat: One additional seat, given the
takeoff and landing (or no boarding or deplaning of number of seats already booked.
passengers at any intermediate stops). Also leg.
Independence of demands: The assumption that
Flight number: A numeric or alphanumeric label demands in one customer category (e.g., booking
for a flight service that involves (generally) a single class or ODF) are statistically independent of de-
aircraft departing from an origin airport, possibly mands in other categories. It is widely believed that
making additional scheduled stops at one or more this assumption is not satisfied in practice. See, for
intermediate airports, and terminating at a destina- example, Hopperstad (1994).
tion airport.
Indexing: The process of assigning individual ODF
Full Nesting: See nested booking. categories to virtual nesting buckets. Smith, Le-
imkuhler, and Darrow (1992) provide details.
Global distribution system (GDS): Computer
and communications systems for linking booking lo- Interspersed arrivals: Characteristic of an arriv-
cations with the computer reservation systems of als process in which booking requests in different
different airlines. Examples are SABRE, Galileo, booking classes do not arrive in any particular order.
and Amadeus. (Compare with sequential booking classes.)
Goodwill costs: An airline’s rejection of a booking Itinerary: For purposes of this paper, an itinerary
request can affect a customer’s propensity to seek is a trip from an origin to a destination across one or
future bookings from that airline. This cost is diffi- more airline networks. A complete specification of
cult to assess but is considered particularly acute in an itinerary includes departure and arrival times,
competitive markets and with customers who are flight numbers, and booking classes. The term is
frequent air travelers. An approximate assessment used ambiguously to include both one-way and
of the cost of a permanently lost customer is the round-trip travel. That is, used in the first way, a
expected net present value of all future bookings round-trip involves two itineraries and, in the sec-
from the customer minus the opportunity costs of ond way, one itinerary.
those bookings. Leg: See flight leg.
Go-show: Passengers who appear at the time of Leg based control: An older, but still common,
flight departure with a valid ticket for the flight but method of reservations control and revenue manage-
for whom there is no record in the reservation sys- ment in which limits are set at the flight leg level on
tem. This no-record situation can occur when there the number of passengers flying in each booking
are significant time lags in transferring booking in- class. Such systems are unable to properly control
formation from reservations sources (e.g., travel multileg traffic, although virtual nesting provides a
agent’s offices) to the CRS or when there are trans- partial solution.
mission breakdowns.
Littlewood’s rule: This simple two-fare allocation
Group bookings: Bookings for groups of passen- rule was proposed by Littlewood (1972). Given aver-
gers that are negotiated with sales representatives age high fare f 1 , average discount fare f 2 , random
of airlines; for example, for a large group from one full fare demand Y, and s seats remaining, Little-
company travelling to a trade show. These should be wood’s rule stipulates that a discount seat should be
distinguished from batch bookings. sold as long as the discount fare equals or exceeds
RESEARCH OVERVIEW AND PROSPECTS / 249
the expected marginal return from a full fare book- among full fare passengers whose tickets are fully
ing of the last remaining seat; that is, discount de- refundable in the event of cancellation or no-show.
mand should be satisfied as long as f 2 $ f 1 Pr(Y .
OBL: See optimal booking limits.
s). This is essentially equivalent to the classic opti-
mal stocking rule for single period stochastic inven- ODF control (O–D problem): Origin– destination
tory (newsvendor) problems. fare control. An approach to revenue management
that accounts for all possible passenger itineraries
Downloaded from informs.org by [129.127.145.240] on 08 October 2017, at 09:50 . For personal use only, all rights reserved.
set of flights with an airline onto alternative flights sibly high revenue value of a booking that includes,
with the same airline. for example, the segment AB in its itinerary but
switches to a different flight at B.
Reservation system controls: The internal logic
used by the reservation system for controlling the Sequential booking classes: The assumption that
availability of seats. This logic is usually difficult to requests for bookings in particular classes are not
change and is often a significant constraint when interleaved; for example, all B-class requests arrive
Downloaded from informs.org by [129.127.145.240] on 08 October 2017, at 09:50 . For personal use only, all rights reserved.
implementing a yield management system. See con- before any Y-class requests. This assumption is
trollable booking classes. rarely satisfied in practice; however, it is close
enough to permit significant revenue gains from
Restrictions: Sets of requirements that are applied
methods based on the assumption. Also, early book-
to discount fare classes to differentiate them as fare
ing restrictions on many discount booking classes
products and discourage diversion. Examples are
ensure a degree of compliance.
fourteen-day advance booking requirements, cancel-
lation penalties, Saturday night stayover, and mid- Sequential fares: See low-before-high fares.
week departure requirements. Also referred to as
Serial nesting: See nested booking.
booking fences.
Show-ups: Passengers who appear for boarding at
Revenue management: The practice of controlling
the time of flight departure. The number of show-
the availability and/or pricing of travel seats in dif-
ups is (final bookings 1 go-shows 1 standbys 2
ferent booking classes with the goal of maximizing
no-shows).
expected revenues or profits. This term has largely
replaced the original term yield management. Single-leg control: See leg based control.
Rules: See restrictions. Space control: See seat inventory control.
Seamless availability: A capability of reservation Spill: Unsatisfied demand that occurs because a
and information systems that allows for direct capacity or booking limit has been reached. See cen-
transmission of availability requests from ticket sorship of demand data.
agents to airlines. With this capability, airlines may
Spill formula: A formula or algorithm that esti-
be able to provide unrestricted origin– destination
mates the amount of spill that has occurred on past
control of their seat inventory.
flights.
Seat allocation: See seat inventory control.
Spoilage: Seats that travel empty despite the pres-
Seat inventory control: The component of a reve- ence of sufficient demand to fill them. This will
nue management system that controls the availabil- occur, for example, if discount booking classes are
ity of seats for different booking classes. closed too early, and full fare demands do not fill the
remaining seats. This should be distinguished from
Segment: One or more flight legs covered by a sin-
excess capacity—seats that are empty because of
gle flight number. Thus, if a flight originates at
insufficient total demand.
airport A, makes an intermediate stop at B, and
terminates at C; the possible flight segments are Standby fares: Some airlines will sell last minute
AB, BC, and ABC. discount seats to certain categories of travelers (e.g.,
youth or military service personnel) who are willing
Segment closed indicator (SCI): A flag in reser-
to wait for a flight that would otherwise depart with
vations control systems that indicates that a book-
empty seats.
ing class is closed to bookings over a particular seg-
ment. The same booking class may be open for Static models: Models that set current seat protec-
bookings over other segments of the same flight. tion policies without consideration of the possibility
This allows for O–D control at the segment level. of adjustments to the protection levels later in the
booking process. (Compare with dynamic models.)
Segment control: A level of itinerary seat inven-
tory control that accounts for the revenue value of Structural solution: A solution to an optimization
flight segments, but does not account for itineraries problem in the form of specifications (frequently
that involve other flight segments. In the case of a equations) that reveal the pattern of behavior of
two leg flight A to B to C, segment control would optimal solutions. These are important because they
permit closing the AB segment to a discount booking lead to a deeper understanding of the nature of
class but leaving the ABC segment open for the optimal solutions and can lead to development of
same class. This system fails to account for the pos- efficient solution algorithms.
RESEARCH OVERVIEW AND PROSPECTS / 251
Threshold curves: Threshold curves are functions tion Laboratory, Massachusetts Institute of Technol-
that return time-dependent booking limits for over- ogy, Cambridge, MA, 1987a.
booking or seat inventory control. P. P. BELOBABA, “Airline Yield Management: An Overview
of Seat Inventory Control,” Transp. Sci. 21, 63–73
Unconstrained demand: An estimate of the de- (1987b).
mand for a past flight or fare class that has been P. P. BELOBABA, “Application of a Probabilistic Decision
corrected for censorship. Model to Airline Seat Inventory Control,” Opns. Res.
37, 183–197 (1989).
Downloaded from informs.org by [129.127.145.240] on 08 October 2017, at 09:50 . For personal use only, all rights reserved.
Upgrade: This term is used in two ways. First, it P. P. BELOBABA AND A. FARKAS, “Yield Management Im-
refers to an offer to a passenger to fly in a higher pacts on Airline Spill Estimation,” Transp. Sci. 33,
service class without additional charge (e.g., in ex- 217–212 (1999).
change for frequent flyer points, or to avoid a denied P. P. BELOBABA AND L. R. WEATHERFORD, “Comparing De-
boarding). Second, it refers to a decision by a cus- cision Rules that Incorporate Customer Diversion in
tomer to book in a higher fare class than originally Perishable Asset Revenue Management Situations,”
intended when he or she is advised that no seats are Decision Sci. 27, 343 (1996).
available at their preferred fare. P. P. BELOBABA AND J. L. WILSON, “Impacts of Yield Man-
agement in Competitive Airline Markets,” J. Air
Virtual nesting/virtual classes: This is one ap- Transp. Mgmt. 3, p. 3 ff (1997).
proach to incorporating origin– destination informa- M. BEN-AKIVA, “Improving Airline Passenger Forecasting
tion into leg or segment based control systems. Mul- Using Reservation Data,” in ORSA/TIMS Proc., St.
tiple ODFs are grouped into virtual buckets on the Louis, MO, 1987.
basis of similar revenue characteristics (e.g., compa- M. BEN-AKIVA AND S. LERMAN, Discrete Choice Analysis:
Theory and Application to Travel Demand, MIT Press,
rable total fare values, or similar total bid prices).
Cambridge, MA, 1985.
The virtual buckets may easily contain a mixture of M. E. BERGE AND C. A. HOPPERSTAD, “Demand Driven Dis-
traditional fare classes. The buckets are then nested patch: A Method for Dynamic Aircraft Assignment,
and assigned to traditional booking classes for con- Models and Algorithms,” Opns. Res. 41, 153–168
trol in a leg based reservation system. (1993).
D. P. BERTSEKAS AND J. N. TSITSIKLIS, Neuro-Dynamic
Yield management: The early term used for what
Programming, Athena Scientific, Belmont, MA, 1996.
is now more commonly called revenue management. A. V. BHATIA AND S. C. PAREKH, “Optimal Allocation of
Cross (1995) attributes the original term to Robert Seats by Fare,” Presentation to AGIFORS Reserva-
L. Crandall when he was Senior Vice President for tions Study Group, Trans World Airlines, 1973.
Marketing (later CEO) at American Airlines. H. BIERMAN, JR. AND J. THOMAS, “Airline Overbooking
Weatherford and Bodily (1992) introduced the gen- Strategies and Bumping Procedures,” Public Policy 21,
eral term, perishable asset revenue management, 601– 606 (1975).
for the general class of inventory control problems of J. R. BIRGE AND F. LOUVEAUX, Introduction to Stochastic
which airline revenue management is an example. Programming, Springer-Verlag, Berlin, 1997.
G. R. BITRAN AND S. M. GILBERT, “Managing Hotel Reser-
vations with Uncertain Arrivals,” Opns. Res. 44, 35– 49
(1996).
BIBLIOGRAPHY
G. R. BITRAN AND S. V. MONDSCHEIN, “An Application of
W. ADAMS AND V. MICHAEL, “Short-Term Forecasting of Yield Management to the Hotel Industry Considering
Passenger Demand and Some Application in Quantas,” Multiple Day Stays,” Opns. Res. 43, 427– 443 (1995).
in AGIFORS Symposium Proc. 27, Sydney, Australia, S. E. BODILY AND L. R. WEATHERFORD, “Perishable-Asset
1987. Revenue Management: Generic and Multiple-Price
J. ALSTRUP, “Booking Control Increases Profit at Scandi- Yield Management with Diversion,” Omega 23, 173–
navian Airlines,” Interfaces 19, 10 –19 (1989). 185 (1995).
J. ALSTRUP, S. BOAS, O. B. G. MADSEN, R. VIDAL, AND V. C. H. BOHUTINSKY, The Sell-Up Potential of Airline De-
VICTOR, “Booking Policy for Flights with Two Types of mand, Master’s thesis, MIT Flight Transportation Lab,
Passengers,” Eur. J. Oper. Res. 27, 274 –288 (1986). Massachusetts Institute of Technology, Cambridge,
S. E. ANDERSSON, “Fares and Bookings,” in AGIFORS MA, 1990.
Symposium Proc. 12, Nathanya, Israel, 1972. S. BORENSTEIN AND N. L. ROSE, “Competition and Price
J. M. BECKMANN, “Decision and Team Problems in Airline Dispersion in the U.S. Airline Industry,” J. Political
Reservations,” Econometrica 26, 134 –145 (1958). Economy 102, 653– 683 (1994).
M. J. BECKMANN AND F. BOBKOWSKI, “Airline Demand: An T. C. BOTIMER, Airline Pricing and Fare Product Differen-
Analysis of Some Frequency Distributions,” Naval Res. tiation, Ph.D. thesis, Department of Civil and Environ-
Logistics Q. 5, 43–51 (1958). mental Engineering, Massachusetts Institute of Tech-
P. P. BELOBABA, Air Travel Demand and Airline Seat In- nology, Cambridge, MA, 1994.
ventory Management, Ph.D. thesis, Flight Transporta- T. C. BOTIMER, Select Ideas on Forecasting with Sales
252 / J. I. MCGILL AND G. J. VAN RYZIN
Relative to Bucketing and ‘Seasonality,’ Unpublished R. G. CROSS, Revenue Management, Broadway Books
Company Report, Continental Airlines, Inc., 1997. (Bantam, Doubleday, Dell Publishing Group), New
A. BOYD, “Airline Alliances,” OR/MS Today 25, 28 –31 York, NY, 276, 1998.
(1998). R. E. CURRY, “Optimal Airline Seat Allocation with Fare
M. A. BRENNER, “The Significance of Airline Passenger Classes Nested by Origins and Destinations,” Transp.
Load Factors,” in Airline Economics, G. W. James (ed.), Sci. 24, 193–204 (1990).
Lexington Books, Lexington, MA, 1982. J. D. DANA, Peak-Load Pricing when the Peak Time is
S. BRUMELLE AND D. WALCZAK, “Dynamic Allocation of Unknown, Working Paper 96-84, General Motors Re-
Downloaded from informs.org by [129.127.145.240] on 08 October 2017, at 09:50 . For personal use only, all rights reserved.
Airline Seat Inventory with Batch Arrivals,” in 1997 search Center for Strategy in Management, Kellogg
Air Transport Research Group of the WCTR Society School, Northwestern University, Evanston, IL, 1996.
Proc. 3, Faculty of Commerce and Business Adminis- J. D. DANA, “Advance-Purchase Discounts and Price Dis-
tration, University of British Columbia, Vancouver, crimination in Competitive Markets,” J. Political
BC, 1997. Economy 106, 395– 422 (1998).
S. L. BRUMELLE AND J. I. MCGILL, A General Model for C. DEETMAN, “Booking Levels,” in AGIFORS Symposium
Airline Overbooking and Two-Class Revenue Manage- Proc., New York, NY, 1964.
ment with Dependent Demands, Working Paper, Uni- G. DOBSON AND P. J. LEDERER, “Airline Scheduling and
versity of British Columbia, Vancouver, BC, 1989. Routing in a Hub-and-Spoke System,” Transp. Sci. 27,
S. L. BRUMELLE AND J. I. MCGILL, “Airline Seat Allocation 281–297 (1993).
with Multiple Nested Fare Classes,” Opns. Res. 41, R. J. DOLAN AND A. P. JEULAND, “Experience Curves and
127–137 (1993). Dynamic Demand Models: Implications for Optimal
S. L. BRUMELLE, J. I. MCGILL, T. H. OUM, K. SAWAKI, AND Pricing Strategies,” J. Marketing 45, 52– 62 (1981).
M. W. TRETHEWAY, “Allocation of Airline Seat between M. DROR, P. TRUDEAU, AND S. P. LADANY, “Network Mod-
Stochastically Dependent Demands,” Transp. Sci. 24, els for Seat Allocation on Flights,” Transp. Res. 22B,
183–192 (1990). 239 –250 (1988).
J. BUHR, “Optimal Sales Limits for Two-Sector Flights,” in E. D’SYLVA, O and D Seat Assignment to Maximize Ex-
AGIFORS Symposium Proc. 22, Lagonissi, Greece, pected Revenue, Unpublished internal report, Boeing
1982. Commercial Airplane Company, Seattle, WA, 1982.
W. J. CAROLL AND R. C. GRIMES, “Evolutionary Change in H. N. DUNLEAVY, “Airline Passenger Overbooking,” in The
Product Management: Experiences in the Car Rental Handbook of Airline Economics, D. Jenkins (ed.), The
Industry,” Interfaces 25, 84 –104 (1995). Aviation Weekly Group of the McGraw-Hill Compa-
G. S. CARPENTER AND D. M. HANSSENS, Market Expan- nies, New York, NY, 469 – 482, 1995.
sion, Cannibalization, and International Airline Pric- M. J. DURHAM, “The Future of SABRE,” in The Handbook
ing: An Aggregrate Analysis of UTA’s Paris–Abidjan of Airline Economics, D. Jenkins (ed.), The Aviation
Route, Working Paper, Kellogg School, Northwestern Weekly Group of the McGraw-Hill Companies, New
University, Evanston, IL, 1991. York, NY, 485– 491, 1995.
G. S. CARPENTER AND D. M. HANSSENS, “Market Expan- M. M. ETSCHMAIER AND M. ROTHSTEIN, “Operations Re-
sion, Cannibalization, and International Airline Pric- search in the Management of the Airlines,” Omega 2,
ing Strategy,” Internat. J. Forecasting 10, 313–326 160 –175 (1974).
(1994). L. M. FALKSON, “Airline Overbooking: Some Comments,”
R. E. CHATWIN, Optimal Airline Overbooking, Ph.D. the- J. Transp. Econ. Policy 3, 352–354 (1969).
sis, Department of Operations Research, Stanford Uni- Y. FENG AND G. GALLEGO, “Optimal Starting Times for
versity, Palo Alto, CA, 1993. End-of-Season Sales and Optimal Stopping Times for
R. E. CHATWIN, “Continuous-Time Airline Overbooking Promotional Fares,” Management Sci. 41, 1371–1391
with Time Dependent Fares and Refunds,” Transp. (1995).
Sci. 33, 182–191 (1999a). C. GAIMON, “Simultaneous and Dynamic Price, Produc-
R. E. CHATWIN, “Multiperiod Airline Overbooking with a tion, Inventory and Capacity Decisions,” Eur. J. Oper.
Single Fare Class,” Opns. Res. 46, 805– 819 (1999b). Res. 35, 426 – 441 (1988).
Y. S. CHOW, H. ROBBINS, AND D. SIEGMUND, Great Expec- G. GALLEGO, A Demand Model for Yield Management,
tations: The Theory of Optional Stopping, Houghton & Technical Report, Columbia University, Dept. of In-
Mifflin, Boston, MA, 1971. dustrial Engineering and Operations Research, New
A. CIANCIMINO, G. INZERILLO, S. LUCIDI, AND L. PALAGI, “A York, NY, 1996.
Mathematical Programming Approach for the Solution G. GALLEGO AND G. VAN RYZIN, “Optimal Dynamic Pricing
of the Railway Yield Management Problem,” Transp. of Inventories with Stochastic Demand over Finite Ho-
Sci. 33, 168 –181 (1999). rizons,” Management Sci. 40, 999 –1020 (1994).
S. A. CONRAD, “Sales Data and the Estimation of De- G. GALLEGO AND G. VAN RYZIN, “A Multi-Product Dynamic
mand,” Oper. Res. Q. 27, 123–127 (1976). Pricing Problem and Its Applications to Network Yield
R. G. CROSS, “An Introduction to Revenue Management,” Management,” Opns. Res. 45, 24 – 41 (1997).
in The Handbook of Airline Economics, D. Jenkins A. GARCIA-DIAZ AND A. KUYUMCU, “A Cutting-Plane Pro-
(ed.), The Aviation Weekly Group of the McGraw-Hill cedure for Maximizing Revenues in Yield Manage-
Companies, New York, NY, 443– 468, 1995. ment,” Comp. Indust. Eng. 33, 51–54 (1997).
RESEARCH OVERVIEW AND PROSPECTS / 253
M. J. GARVEY, “United to Manage Fares with Informix- The Handbook of Airline Economics, D. Jenkins (ed.),
SP2 Setup,” Informationweek, March 10, 1997, 28. The Aviation Weekly Group of the McGraw-Hill Com-
M. K. GERAGHTY AND E. JOHNSON, “Revenue Management panies, New York, NY, 477– 482, 1995.
Saves National Car Rental,” Interfaces 27, 107–127 S. LADANY, “Dynamic Operating Rules for Motel Reserva-
(1997). tions,” Decision Sci. 7, 829 – 840 (1976).
Y. GERCHAK AND M. PARLAR, “A Single Period Inventory S. LADANY, “Bayesian Dynamic Operating Rules for Opti-
Problem with Partially Controlled Demand,” Comput. mal Hotel Reservations,” Z. Opns. Res. 21, B165–B176
Opns. Res. 14, 1–9 (1987).
Downloaded from informs.org by [129.127.145.240] on 08 October 2017, at 09:50 . For personal use only, all rights reserved.
(1977).
Y. GERCHAK, M. PARLAR, AND T. K. M. YEE, “Optimal Ra- S. LADANY AND A. ARBEL, “Optimal Cruise-Liner Passen-
tioning Policies and Production Quantities for Prod- ger Cabin Pricing Policy,” Eur. J. Oper. Res. 55, 136 –
ucts with Several Demand Classes,” Can. J. Admin. 147 (1991).
Sci. 2, 161–176 (1985). S. P. LADANY AND D. N. BEDI, “Dynamic Rules for Flights
F. GLOVER, R. GLOVER, J. LORENZO, AND C. MCMILLAN, with an Intermediate Stop,” Omega 5, 721–730 (1977).
“The Passenger Mix Problem in the Scheduled Air- A. H. L. LAU AND H. S. LAU, “The Newsboy Problem with
lines,” Interfaces 12, 73–79 (1982). Price Dependent Demand Distribution,” IIE Trans. 20,
R. G. HAMZAEE AND B. VASIGH, “An Applied Model of Air- 168 –175 (1988).
line Revenue Management,” J. Travel Res. 35, 64 – 68 H. S. LAU AND A. H. L. LAU, “The Multi-Product Multi-
(1997). Constraint Newsboy Problem: Applications, Formula-
P. HANLON, Global Airlines: Competition in a Transna- tion, and Solution,” J. Opns. Mgmt. 13, 153–162
tional Industry, Butterworth-Heinemann, Oxford, UK, (1995).
247, 1996. H. S. LAU AND A. H. L. LAU, “The Newstand Problem: A
P. HARRIS AND G. MARUCCI, “A Short Term Forecasting Capacitated Multiple-Product Single-Period Inventory
Model,” in AGIFORS Symposium Proc. 23, Memphis, Problem,” Eur. J. Oper. Res. 94, 29 – 42 (1996).
TN, 1983. C. J. LAUTENBACHER AND S. J. STIDHAM, “The Underlying
M. HERSH AND S. P. LADANY, “Optimal Seat Allocation for
Markov Decision Process in the Single-Leg Airline
Flights with Intermediate Stops,” Comput. Opns. Res.
Yield Management Problem,” Transp. Sci. 33, 136 –146
5, 31–37 (1978).
(1999).
C. HOPPERSTAD, “The Application of Path Preference and
A. O. LEE, “Airline Reservations Forecasting,” in AGI-
Stochastic Demand Modelling to Market Based Fore-
FORS Symposium Proc. 28, New Seabury, MA, 1988.
casting,” in AGIFORS Reservations and Yield Manage-
A. O. LEE, Airline Reservations Forecasting: Probabilistic
ment Study Group Proc., Hong Kong, 1994.
and Statistical Models of the Booking Process, Ph.D.
F. INZERILLI AND S. R. JARA-DIAZ, “Uncertain Demand,
thesis, Flight Transportation Laboratory, Massachu-
Modal Competition and Optimal Price-Capacity Ad-
setts Institute of Technology, Cambridge, MA, 1990.
justments in Air Transportation,” Transportation 21,
T. C. LEE AND M. HERSH, “A Model for Dynamic Airline
91–101 (1994).
D. JENKINS, (ed.) Handbook of Airline Economics, The Seat Inventory Control with Multiple Seat Bookings,”
Aviation Weekly Group of the McGraw-Hill Compa- Transp. Sci. 27, 252–265 (1993).
nies, New York, NY, 1995. C. O. LENNON, “Market Reactions to Fare Constraints and
A. KANAFANI AND E. SADOULET, “The Partitioning of Long Its Effects on Passenger Revenue,” in AGIFORS Sym-
Haul Air Traffic—A Study in Multinomial Choice,” posium Proc. 12, Nathanya, Israel, 1972.
Transp. Res. 11, 1– 8 (1977). E. L’HEUREUX, “A New Twist in Forecasting Short-Term
I. KARAESMAN AND G. J. VAN RYZIN, Overbooking with Passenger Pickup,” in AGIFORS Symposium Proc. 26,
Substitutable Inventory Classes, Working Paper, Bowness-On-Windermere, England, 1986.
Graduate School of Business, Columbia University, M. Z. F. LI, Pricing Perishable Inventories by Using Mar-
New York, NY, 1998. keting Restrictions with Applications to Airlines, Ph.D.
S. E. KIMES, “Yield Management: A Tool for Capacity Con- thesis, Faculty of Commerce Business Administration,
strained Service Firms,” J. Opns. Mgmt. 8, 348 –363 University of British Columbia, Vancouver, BC, 1994.
(1989). M. Z. F. LI, On the Multi-Fare Seat Allocation Problem,
A. J. KLEYWEGT AND J. D. PAPASTAVROU, “The Dynamic Working Paper, Division of Applied Economics, Nan-
and Stochastic Knapsack Problem,” Opns. Res. 46, yang Business School, Nanyang Technological Univer-
17–35 (1998). sity, Singapore, 1998.
L. KOSTEN, “Een Mathematisch Model voor een reservings M. Z. F. LI AND T. H. OUM, Seat Allocation Game on
probleem,” Statist. Neerland. 14, 85–94 (1960). Flights with Two Fares, Working Paper, Division of
G. M. KOSTYRSKY, “The Evaluation of an Airline’s Pricing Applied Economics, Nanyang Business School, Nan-
Policy,” in AGIFORS Symposium Proc. 11, Benalma- yang Technological University, Singapore, 1998.
dena, Spain, 1971. Y. LIANG, “Solution to the Continuous Time Dynamic
D. G. H. KRAFT, T. H. OUM, AND M. W. TRETHEWAY, “Air- Yield Management Model,” Transp. Sci. (In press).
line Seat Management,” Logistics & Transp. Rev. 22, V. LIBERMAN AND U. YECHIALI, “Hotel Overbooking Prob-
115–130 (1986). lem—Inventory System with Stochastic Cancella-
S. S. KRETSCH, “Airline Fare Management and Policy,” in tions,” Adv. Appl. Probab. 9, 220 –230 (1977).
254 / J. I. MCGILL AND G. J. VAN RYZIN
V. LIBERMAN AND U. YECHIALI, “On the Hotel Overbooking Algorithm for Calculating Consistent Itinerary Flows,”
Problem,” Management Sci. 24, 1117–1126 (1978). Transp. Sci. 25, 225–239 (1991).
K. LITTLEWOOD, “Forecasting and Control of Passenger N. PRATTE, “Automated Full Fare Protection System,” in
Bookings,” in AGIFORS Symposium Proc. 12, Nath- AGIFORS Symposium Proc. 26, Bowness-On-Wind-
anya, Israel, 1972. ermere, England, 1986.
C. LYLE, “A Statistical Analysis of the Variability in Air- K. PROUSSALOGLOU AND F. KOPPELMAN, “Air Carrier De-
craft Occupancy,” in AGIFORS Symposium Proc. 10, mand: An Analysis of Market Share Determinants,”
Transportation 2, 371–388 (1995).
Downloaded from informs.org by [129.127.145.240] on 08 October 2017, at 09:50 . For personal use only, all rights reserved.
with Multiple Fare Types,” in AGIFORS Symposium Single-Flight Leg, unpublished company report, Doug-
Proc. 23, Memphis, TN, 1983. las Aircraft Company, McDonnell Douglas Corpora-
L. R. WEATHERFORD, Perishable Asset Revenue Manage- tion, Long Beach, CA, 1986b.
ment in General Business Situations, Ph.D. thesis, R. D. WOLLMER, A Hub-Spoke Seat Management Model,
Darden Graduate School of Business Administration, unpublished company report, Douglas Aircraft Com-
University of Virginia, Charlottesville, VA, 1991. pany, McDonnell Douglas Corporation, Long Beach,
L. R. WEATHERFORD, Optimization of Perishable-Asset CA, 1986c.
Revenue Management Problems that Allow Prices as R. D. WOLLMER, “An Airline Seat Management Model for a
Downloaded from informs.org by [129.127.145.240] on 08 October 2017, at 09:50 . For personal use only, all rights reserved.
Decision Variables, Working Paper, University of Wy- Single Leg Route when Lower Fare Classes Book
oming, Laramie, WY, 1994. First,” Opns. Res. 40, 26 –37 (1992).
L. R. WEATHERFORD, “Optimization of Joint Pricing and J. T. WONG, Airline Network Seat Allocation, Ph.D. thesis,
Allocation Perishable-Asset Revenue Management Northwestern University, Evanston, IL., 1990.
Problems with Cross-Elasticity,” J. Combin. Optimiza- J. T. WONG, F. S. KOPPELMAN, AND M. S. DASKIN, “Flexible
tion 1, 277–304 (1997). Assignment Approach to Itinerary Seat Allocation,”
L. R. WEATHERFORD AND S. E. BODILY, “A Taxonomy and Transp. Res. 27B, 33– 48 (1993).
Research Overview of Perishable-Asset Revenue Man- R. WYSONG, “A Simplified Method for Including Network
agement: Yield Management, Overbooking, and Pric- Effects in Capacity Control,” in AGIFORS Symposium
ing,” Opns. Res. 40, 831– 844 (1992). Proc. 28, New Seabury, Massachusetts, 1988.
L. R. WEATHERFORD, S. E. BODILY, AND P. E. PFEIFER, P. S. YOU, “Dynamic Pricing in Airline Seat Management
“Modeling the Customer Arrival Process and Compar- for Flights with Multiple Legs,” Transp. Sci. 33, 192–
ing Decision Rules in Perishable Asset Revenue Man- 206 (1999).
agement Situations,” Transp. Sci. 27, 239 –251 (1993). Y. YOUNG AND R. VAN SLYKE, Stochastic Knapsack Models
L. R. WEATHERFORD AND P. E. PFEIFER, “The Economic of Yield Management, Technical Report 94-76, Poly-
Value of Using Advance Booking of Orders,” Omega 22, technic University, New York, NY, 1994.
105–111 (1994). W. ZHAO, Dynamic and Static Yield Management Models,
E. L. WILLIAMSON, Comparison of Optimization Tech- Ph.D. thesis, The Wharton School, Operations and In-
niques for Origin–Destination Seat Inventory Control, formation Management Department, University of
Master’s thesis, Flight Transportation Laboratory, Pennsylvania, Philadelphia, PA, 1999.
Massachusetts Institute of Technology, Cambridge, W. ZHAO AND Y. S. ZHENG, General Dynamic Models for
MA, 1988. Airline Seat Allocation, Working Paper, The Wharton
E. L. WILLIAMSON, Airline Network Seat Inventory Con- School of the University of Pennsylvania, Philadelphia,
trol: Methodologies and Revenue Impacts, Ph.D. the- PA, 1998a.
sis, Flight Transportation Laboratory, Massachusetts W. ZHAO AND Y. S. ZHENG, Optimal Policy for Multi-Class
Institute of Technology, Cambridge, MA, 1992. Seat Allocation with Nonstationary Demand, Working
R. D. WOLLMER, An Airline Reservation Model for Open- Paper, The Wharton School, University of Pennsylva-
ing and Closing Fare Classes, unpublished company nia, Philadelphia, PA, 1998b.
report, Douglas Aircraft Company, McDonnell Douglas M. ZIMMERMAN, “Computer Glitch Costs AA $50MM,” Aus-
Corporation, Long Beach, CA, 1986a. tin American-Statesman, Sunday, September 11th,
R. D. WOLLMER, An Airline Seat Management Model for a 1988.