PRODUCTION AND OPERATIONS MANAGEMENT POMS

Vol. 19, No. 1, January–February 2010, pp. 40–60 DOI10.3401/poms.1080.01046

ISSN 1059-1478|EISSN 1937-5956|10|1901|0040 r 2009 Production and Operations Management Society

Optimal Dynamic Pricing of Perishable Items by a

Monopolist Facing Strategic Consumers
Yuri Levin, Jeff McGill, Mikhail Nediak
School of Business, Queen’s University, Kingston, ON, Canada K7L 3N6, [email protected], [email protected],
[email protected]

e introduce a dynamic pricing model for a monopolistic company selling a perishable product to a finite population of
W strategic consumers (customers who are aware that pricing is dynamic and may time their purchases strategically). This
problem is modeled as a stochastic dynamic game in which the company’s objective is to maximize total expected revenues,
and each customer maximizes the expected present value of utility. We prove the existence of a unique subgame-perfect
equilibrium pricing policy, provide equilibrium optimality conditions for both customer and seller, and prove monotonicity
results for special cases. We demonstrate through numerical examples that a company that ignores strategic consumer
behavior may receive much lower total revenues than one that uses the strategic equilibrium pricing policy. We also show
that, when the initial capacity is a decision variable, it can be used together with the appropriate pricing policy to effectively
reduce the impact of strategic consumer behavior. The proposed model is computationally tractable for problems of realis-
tic size.
Key words: customer behavior; dynamic pricing; strategic consumers; stochastic games
History: Received: February 2008; Accepted: December 2008 by Costis Maglaras; after 3 revisions.

1. Introduction valuation for the product while ignoring future prod-

The recent growth in internet-based marketing has uct availability and purchasing opportunities. This
stimulated widespread experimentation with dynamic assumption leads to tractable models, significantly
pricing—the practice of varying prices for the same simplifies analysis, and can lead to attractive struc-
goods over time or across customer classes in an at- tural properties for pricing policies. But many real-life
tempt to increase total revenues for the seller. In this consumers are strategic; that is, they attempt to ‘‘get
paper, we examine a particular version of dynamic more for less’’ by delaying purchase in anticipation of
pricing that has been widely applied to sales of price reductions. When supplies are limited, such
perishable items over a finite time period. Familiar strategic consumers must also take into account prod-
examples include passenger transportation tickets, uct availability and the strategic behavior of other
hotel bookings, automobile rentals, end-of-season consumers. Thus, a comprehensive model of strategic
style goods, and expiring models of electronic equip- consumer behavior needs to capture interactions be-
ment or other high-value items. tween the seller and the consumers as well as among
Particularly notable successes of dynamic pricing the consumers.
have been reported in the airline industry, which has,
perhaps, the longest history of controlled dynamic 1.1. Survey of Existing Literature
pricing through their revenue management (RM) sys- Dynamic pricing models for an infinite population of
tems (traditional RM is more properly viewed as myopic customers include those of Feng and Xiao
inventory control across price classes; however, the (2000), Gallego and van Ryzin (1994), and Zhao and
distinction between RM and dynamic pricing is Zheng (2000). In their book, Talluri and van Ryzin
becoming increasingly blurred as the use of online (2004) discuss myopic vs. strategic consumer models
booking expands). There is an extensive literature on (Section 5.1.4.1). They point out that the myopic as-
RM and related practices. For a review of this area, see sumption provides an acceptable approximation
Phillips (2005) and Talluri and van Ryzin (2004). when customers are relatively spontaneous in mak-
In dynamic pricing, it is crucial to take into account ing decisions (typical with low-price products) or
key characteristics of the market such as consumer when they do not have enough time or information to
behavior and the availabilities of goods and services. behave strategically. However, for more expensive
Classical dynamic pricing models often assume that and durable products and when information is avail-
consumer behavior is myopic – a consumer makes a able, this assumption becomes less realistic. They
purchase as soon as the price drops below his/her argue that failure to account for strategic customer
40
Levin, McGill, and Nediak: Dynamic Pricing with Strategic Consumers
Production and Operations Management 19(1), pp. 40–60, r 2009 Production and Operations Management Society 41

behavior could significantly reduce expected revenues Papers which study two-period pricing models
from dynamic pricing. with strategic consumers include Liu and van Ryzin
The literature on consumer behavior issues in RM (2008) who use quantity decisions (rather than pric-
for perishable products is growing rapidly; for a sur- ing) as means to induce early purchases. Zhang and
vey see Shen and Su (2007). Here, we only review Cooper (2008) also consider the effect of strategic
models that are the most relevant in terms of their customers in a two-period model with regular and
assumptions and conclusions relative to the present clearance prices and an option of restricting product
work. In general terms, any analysis of individual availability in the second period. Cachon and
strategic consumer behavior requires simplifying as- Swinney (2007) study the value of an additional prod-
sumptions. For example, many studies in this area uct procurement, termed ‘‘quick response,’’ in a two-
assume deterministic demand, and, even in the more period model where the second period price is always
realistic stochastic case, most articles restrict the a markdown. The model considers three segments of
dynamics of price changes to markdowns or consider customers: myopic, bargain hunting (those who only
problems with only two decision periods. Each of purchase in the second period), and strategic. The
these make additional specific assumptions on con- paper finds that ‘‘quick response’’ offers more value in
sumer behavior. the presence of strategic consumers.
An early deterministic model of dynamic pricing A two-channel model is studied by Caldentey and
with strategic consumers is presented in Besanko and Vulcano (2007) who consider a setting in which stra-
Winston (1990). They show that the subgame-perfect tegic consumers, who arrive according to a Poisson
equilibrium pricing policy for the firm is to lower process, decide between buying immediately at a
prices over time in a manner similar to price-skimming. (fixed) posted price and buying later by entering their
The selling capacity in this model is unlimited. Su bids to a sealed-bid multi-unit uniform price auction.
(2007) considers a deterministic demand model with Two cases are studied: the single-auction channel case
customers partitioned into four segments according to where the company only manages the auction, and
their valuation level and waiting costs, and derives the dual-channel case where the company manages
conditions relating to when the seller should use both the auction and the fixed-price channel. The
markdowns or markups. The company controls both authors discuss the equilibrium and a simple, asymp-
the price and the fill rate in this work. totically optimal, heuristic rule for its approximation.
Aviv and Pazgal (2008) study the optimal pricing Several articles study the effects of customer re-
of fashion-like seasonal goods in the presence of stra- sponse to RM over multiple selling seasons; see, for
tegic customers who arrive according to a Poisson’s example, Ovchinnikov and Milner (2007). In this
process but have deterministically declining valua- work, the market is treated in an aggregated form,
tions over time. This work considers subgame-perfect and some fraction of customers learns to expect end-
Nash equilibria between a seller and strategic consum- of-season discounts based on previous interactions
ers for the cases of inventory-contingent discounting with the seller. Within each season, given the full and
strategies and announced fixed discount strategies. The sale prices, the seller selects the number of units
paper finds that the seller cannot effectively avoid the available at the end-of-season discount. The article
adverse impact of strategic consumer behavior even provides analytical results for the explicit multi-
under low levels of initial inventory, and that fixed season model. Gallego et al. (2008) investigate a sim-
discounting strategies may outperform contingent pric- ilar class of models with multiple selling seasons and
ing strategies. Elmaghraby et al. (2008) study optimal two-period pricing within each season. Customers
markdown mechanisms in the presence of strategic update their expectations about the probability of
consumers who have fixed valuation throughout the acquiring an item at a discount based on previous
selling season. interaction with the seller. Both of these papers show
A number of authors consider fully dynamic pric- the existence of complex dynamic patterns of con-
ing but under specific behavioral assumptions. Xu sumer–seller interactions.
and Hopp (2004) study a continuous-time model
with strategic customers who arrive according to a 1.2. Inherently Stochastic Markets
Poisson process and whose price sensitivity varies In a deterministic setting, the company does not
with time. In this model, customers commit to a pur- experience uncertainty in demand and, given its
chasing time which is determined from an open-loop pricing/sales policy, can always accurately predict
(state independent) decision process. One of their consumer behavior and evaluate the performance of
results is that the optimal prices follow a submartin- the policy. However, most markets in which dynamic
gale if the price sensitivity is constant or decreas- pricing is employed are inherently stochastic in the
ing, and supermartingale if the price sensitivity is sense that outcomes (sales paths) are not wholly
increasing. determined by company pricing decisions but also
 Levin, McGill, and Nediak: Dynamic Pricing with Strategic Consumers
42 Production and Operations Management 19(1), pp. 40–60, r 2009 Production and Operations Management Society

depend on other random events. In our view, cus- control of the fill rate. Dynamic explicit control of
tomer valuations or propensities to consume can be the fill rate may not be easy to implement since se-
uncertain up to the very time of consumption, partic- lective rejection of purchase attempts over time can
ularly for advance purchases. Appropriate examples alienate customers, and it may be hard to keep a cus-
include discretionary hotel reservations, vacation tomer from making another purchase attempt if the
packages, and sports and entertainment events whose original attempt is rejected. Our model also differs
value may depend on uncertain weather and time from that of Su (2007) in the description of strategic
availability for the consumer. Uncertainty in valua- consumer behavior, since the latter introduces explicit
tions is also present for a wide variety of services that waiting costs for the consumers. In contrast to many
need to be reserved in advance (housekeeping, gar- prior stochastic models, we do not limit the dynamics
dening, snow removal, etc.). of prices to two time periods, and we do not limit the
An intention to purchase at a given time or price model to decreasing price paths (markdowns). While
point may not result in a purchase. For example, a markdown policies are appropriate in some retail set-
customer who feels that a price is acceptable may not tings, they are not universally applicable. These two
act immediately for a number of reasons, including aspects allow us to study much richer dynamics of
procrastination, congestion delay, or terms and con- price and sales processes than more restricted models
ditions of purchase. Also, many customers exhibit permit.
‘‘wavering’’ around a decision, particularly for dis- In models with deterministic valuations (or random
cretionary purchases – deciding one moment that they valuations that are realized at the beginning of the
will purchase and the next that they really shouldn’t. sales season and remain fixed thereafter), homoge-
It can also take more than a single inquiry before a neous consumers will all complete purchases
consumer settles on a purchase. (Uncertainty in indi- simultaneously when ‘‘conditions are right.’’ This, of
vidual choices, even under seemingly identical course, does not accord with the behavior of real
conditions, has been reported in empirical studies markets. Firms are much more likely to see relative
[see, e.g., Tversky 1972]. Moreover, Xie and Shugan increases or decreases in the pace of sales as prices and
indicate that ‘‘buyers are nearly always uncertain inventories change. This motivates a model in which
about their future valuations for most services’’ [Xie individual consumers respond to prices and other
and Shugan 2001, p. 219].) All of these behaviors are market conditions by controlling the intensity of their
difficult to capture with fixed valuation assumptions. demand processes. In this way, actual completions
Demand uncertainty often brings product availabil- of sales will be distributed over time stochastically,
ity into consideration for both the company and its and consumers will have to take this uncertainty into
customers. As a result, the pricing policy and con- account in their decision making. Such a strategic re-
sumer response will, in general, depend on the sponse of consumers in the form of shopping intensity
remaining capacity level and remaining time. Finally, is a distinctive feature of this paper.
implementation of a deterministic policy assumes that This approach to consumer behavior removes the
the company can credibly commit to a price path, an need for explicit rationing controls for the firm be-
assumption that does not always hold in practice. The cause, much like in real life, random rationing of
issue of credible price commitment is not important the product to consumers occurs as a natural conse-
when consumers are myopic and, in that case, a quence of market operation. This model also obviates
deterministic policy can be asymptotically optimal the assumption that the choice behavior of consumers
Gallego and van Ryzin (1994), However, as shown by is predetermined at the beginning of the sales season.
Aviv and Pazgal (2008), this becomes an important Such generality comes with a cost – the model re-
issue in the presence of strategic consumers. quires an assumption that consumers re-sample their
In this paper, we present a dynamic pricing model valuations at each decision point, independently of
for a finite population of strategic consumers in a previous valuations. While this can be viewed as a
market which is inherently stochastic in its operation. limiting model for consumer ‘‘wavering,’’ a more sat-
The seller is a monopolist firm that contingently prices isfactory model would allow dependency across time;
a fixed stock of items over a finite time horizon, and for example, with a random-walk valuation process.
the consumers exercise choice based on the strategic This is an interesting topic for future research.
factors discussed above. Our consumer choice model
is similar in spirit to that of Besanko and Winston 1.3. Chief Contributions
(1990), but we extend that work by taking demand The model described here is a stochastic dynamic
uncertainty into consideration and allowing for ca- game in which the company controls prices contin-
pacity restrictions. In contrast to the deterministic gent on time and present market conditions to
model of Su (2007), which also treats capacity restric- maximize total expected revenues, and each customer
tions, we do not assume that the firm has explicit controls demand intensity to maximize the expected
Levin, McGill, and Nediak: Dynamic Pricing with Strategic Consumers
Production and Operations Management 19(1), pp. 40–60, r 2009 Production and Operations Management Society 43

present value of utility. In the presence of imperfect creases significantly with an increase in supply level
information, such stochastic games are notoriously when compared with the myopic case.
difficult to analyze. Therefore, we make some simpli- The model described here can be used as a bench-
fying assumptions which guarantee that this is a mark for the importance of strategic consumer
perfect information game, and its participants can ra- behavior. For example, it can be used to estimate the
tionally anticipate the behavior of others. In this aspect, losses a company can incur due to strategic consumer
our game structure is similar to that of Besanko and behavior, both when this behavior is ignored and
Winston (1990). when it is optimally anticipated in the company’s
We prove that a unique subgame-perfect equilib- pricing policy. Numerical examples demonstrate that
rium pricing policy exists, and that the ‘‘strategicity’’ a company that ignores strategic consumer behavior
of consumers affects the choice of an optimal pricing may receive much lower total revenues than one that
policy. We also provide equilibrium optimality con- uses the strategic equilibrium pricing policy, and that
ditions for both consumer and seller, and study losses are much more profound when supply levels
properties of the equilibrium policy. The equilibrium are high.
obtained here is a direct generalization of the optimal While the focus of our paper is on contingent pric-
pricing policy for myopic consumers which would be ing policies, the model can be appropriately modified
obtained in a discrete-time, finite population version to evaluate consumer response to pre-announced, de-
of the classical model of Gallego and van Ryzin (1994). terministic, pricing policies. We numerically evaluate
Practical insights derived from this work relate to the effects of two-period pre-announced pricing pol-
the structure of pricing policies and their relation to icies in comparison with fully dynamic, contingent,
capacity decisions. We show that, when consumer ra- ones and find that the latter perform better when the
tionality is limited, the pricing policy exhibits natural supply level is low and consumer behavior is closer to
monotonic properties: price decreases as remaining myopic. However, contingent policies may underper-
capacity increases or remaining time decreases. Such form when the capacity level is high. This agrees with
properties are familiar from the analysis of Gallego the findings of Aviv and Pazgal (2008) obtained under
and van Ryzin (1994) in the myopic case. When con- different modelling assumptions.
sumers are fully rational and strategic, prices decrease Finally, the expected revenue as a function of initial
monotonically in time in the special case of high capacity provided by our model can also be used in a
product supply. This result is also in agreement with newsvendor-type formulation in which the initial
Besanko and Winston (1990) in the deterministic case. procurement decision is followed by dynamic pricing
However, these authors also find that price approaches in a strategic consumer market. This is made possible
marginal cost as the discount factor approaches one, by the stochastic market feature of the model. We
whereas in our case price remains strictly bounded show that, when the initial capacity is a decision vari-
away from marginal cost. This difference arises be- able, its proper use together with dynamic pricing is
cause of the stochastic nature of the sales process. crucial for effective reduction of the effects of strategic
We also demonstrate that monotonic properties do consumer behavior. In short, dynamic pricing may not
not hold in general. This contrasts with well-known be able to compensate for inappropriate capacity de-
results in the theory of dynamic pricing for the my- cisions. Also, the appropriate capacity level is lower
opic case. In practical terms, if strategic customers are for the case of strategic consumers. On the computa-
only partially rational and are not aware of the pres- tional side, the proposed model is tractable for
ent state of the system or cannot take it into account, realistic-size problems. In summary, the chief contri-
then a familiar monotonic property holds: price in- butions of this paper are:
creases as inventory decreases. In contrast, if the cus-
tomers are rational and aware of the current state of  pricing policies (based on a game between a mo-
the system, then the company may have to counter- nopolist and strategic consumers) that generalize
act their strategic behavior with policies that deviate classical optimal pricing policies for the myopic
from monotonicity. For example, there must be a case in an unrestricted dynamic setting,
credible threat to increase prices even in the absence  monotonic properties of pricing policies in the
of sales. presence of strategic consumers, and
Numerical simulations of pricing policies in the  an analysis of relations between and effects
general case also indicate that price paths are mark- of appropriate/inappropriate procurement deci-
edly different for the cases of myopic and strategic sions and dynamic pricing policies in the pres-
consumers. In the myopic case, the average level of ence of strategic consumers.
prices starts high and decreases over time, whereas
the average level is more constant in the strategic case. These results are restricted to the case of monopo-
Also, for strategic consumers, pricing flexibility de- listic firms that possess knowledge of consumer
 Levin, McGill, and Nediak: Dynamic Pricing with Strategic Consumers
44 Production and Operations Management 19(1), pp. 40–60, r 2009 Production and Operations Management Society

valuation distributions. Such results are much harder finite number of individual consumers. The pop-
to obtain in more complex market settings. In a com- ulation demand intensity is the sum of consumer
panion paper, Levina et al. (2009), we relax the demand intensities.
assumption of knowledge of the valuation distribu-  Consumers control the intensity of their demand
tion and allow for learning of consumer behavior as processes (which we call shopping intensity) in
the sales season unfolds. That model cannot explicitly response to price and other market conditions.
capture the dynamic aspect of the game between the  There exists a common upper bound l for the
company and the consumers. A game-theoretic model shopping intensity of each consumer.
of interactions between the company and strategic
consumers in the presence of consumer and company In this interpretation of demand, consumers cannot
learning is an interesting direction for future research. control the precise timing of their purchases, but
In Levin et al. (2009), we examine dynamic pricing in respond to more favorable prices and other market
an oligopoly with strategic consumers in multiple conditions by increased eagerness to purchase, which
market segments. The complexity of the models in translates to a higher expected frequency of pur-
both of these papers does not permit detailed study chases. This assumption is motivated by two practical
of monotonic properties of pricing policies and the considerations. First, in a real marketplace, all inter-
importance of capacity decisions. ested consumers cannot purchase at exactly the same
The organization of this paper is as follows. We time because both the capacity of the purchasing
present the notation and game-theoretic model in channel and the timing of access by consumers to that
Section 2 and show existence of the unique subgame- channel are constrained. Second, as pointed out by
perfect equilibrium in Section 3. We explore monoto- Su (2006), real-world consumers have a tendency to
nicity results of the equilibrium pricing policies in procrastinate. The total demand intensity is bounded
Section 4, and show the effect of strategic consumer in practice: even if the product is available for free,
behavior on the performance of pricing policies and there will be a maximum number of customers per
capacity decisions in Section 5. We conclude and unit of time who can purchase the product. Since each
indicate several directions for further research in consumer contributes a fraction of the total intensity,
Section 6. Additional discussions of the modeling as- it is reasonable that the individual intensity is also
sumptions, a glossary of notation, and mathematical bounded. The value l captures transaction uncertain-
proofs are provided in the supporting information ties in the market as well as consumer tendency
Appendix S1. toward procrastination, and has a precise interpreta-
tion as the reciprocal of the expected time for an eager
2. Model Description customer to purchase the product. It can also be in-
We describe the model in three stages. First, in Section terpreted as the intensity of shopping opportunities,
2.1, we specify general model assumptions with par- and the quantity lT is the expected number of shop-
ticular attention to the demand process, which is ping opportunities for a customer who is eager
characterized by the shopping intensity of consumers. throughout the sales season.
In Section 2.2, we describe the consumer choice model Finally, to simplify the model we take the approach
and define the stochastic dynamic game that captures employed in Talluri and van Ryzin (2004) and dis-
company–consumer interactions. Finally, in Section cretize the time into short time steps which we call
2.3, we derive the objective functions for consumers decision periods. The number of decision periods is
and the company and the equilibrium response for sufficiently large that any continuous-time counting
consumers. process in the model can be well approximated by its
discrete-time analogue. Under such a choice of dis-
2.1. General Modeling Assumptions cretization, the probability of more than one event
Consider a product with limited availability sold by occurring in a decision period has to be relatively
a monopolistic company over a finite planning hori- small, and we can assume that at most one event in
zon [0, T]. A common model of demand from a any of the processes can occur per decision period.
population of consumers is a counting process with For simplicity of notation, we assume that each deci-
intensity generally dependent on time and price (see, sion period has a length of exactly one time unit. Then
e.g., Gallego and van Ryzin 1994). Since we want to the shopping intensity of a customer is equivalent to
describe behavior of individual consumers, this as- the probability that this particular customer purchases
sumption takes the following form: an item in the next decision period. Consequently, the
Demand process: demand intensity of the entire market is equivalent to
the probability that any consumer purchases an item
 The population demand process is a sum of in the given decision period. The purchase probabil-
counting demand processes originating from a ities retain the additive property of intensities: the
Levin, McGill, and Nediak: Dynamic Pricing with Strategic Consumers
Production and Operations Management 19(1), pp. 40–60, r 2009 Production and Operations Management Society 45

probability of a purchase occurring is the sum of pur- the initial number of customers N  Y (the case of
chase probabilities of individual customers. We NoY can be reduced to N 5 Y since any excess items
summarize this as: cannot be sold and should be disregarded. The case of
Representation of intensity as probability: Consumer N 5 Y is appropriate for a company offering a make-
shopping intensity decisions are expressed as pur- to-order product to strategic customers). All custom-
chase probabilities in the decision periods of the ers are present from the beginning and can monitor
discrete-time model. the market as well as make purchases during the
Consumer behavior, discussed in detail in the next entire planning horizon.
section, can be summarized as follows. In each pe- With regard to the information structure of the
riod t, each customer draws a valuation according to a game, we assume by default:
given distribution. Draws are independent across pe- Perfect market information: All game participants
riods and are exchangeable random variables across have perfect information about the market including
customers. Each individual customer then uses his its parametric, distributional, and dynamic character-
draw to select a shopping intensity from the interval istics (except for random valuation realizations).
½0; l: To ensure that our time discretization is suffi- A perfect information assumption implies that the
ciently fine, we must select the time step (time units) remaining inventory level y  Y and the number of
so that the upper bound  l on demand intensity is less customers who have not yet acquired the product,
than the reciprocal of the total maximum market size. n  N, are public information. Because the customer
The individual customer’s decision will generally also population is homogeneous, it is unimportant which
depend upon the posted price, the time remaining in particular customer has not yet acquired the product,
the horizon, the remaining inventory, and the cus- and the state of the system in decision period
tomer’s assessment of the expected value of delaying tAf0, 1, . . ., Tg can be described by the pair y, n. How-
purchase. Once all customers have simultaneously ever, since the current inventory level can be
selected their probabilities, then at most one customer recovered from n as y 5 Y  (N  n), we can drop it
will, conditionally independent of the past, make a from the state description. A transition from state n to
purchase in period t according to the selected prob- n  1 represents a sale of one unit of the product to
abilities. one of the customers. Given maximum shopping in-
Next, we introduce notation, describe how consum- tensity l, the value of n, together with the remaining
ers decide between shopping now and waiting – the time, allows us to quantify the risk that a consumer
consumer choice model – and how the company can will fail to complete the purchase (rationing risk). This
influence this behavior. is how a randomized sales process can replace explicit
rationing controls considered in some models. We
2.2. The Game view this as an advantage of the model since, as
The goal of our model is to capture the intertemporal pointed out in the introduction, explicit rationing
behavior of customers who strategically attempt to controls may be hard to implement when the sales
time their purchase to periods with lower price. They process is truly dynamic.
indirectly control timing of their purchases by mod- The key modeling issue is how strategic consumers
erating their shopping intensity relative to that of compare a current purchase with a possible purchase
myopic customers. Customer shopping intensity is in the future. One of the common approaches is based
selected dynamically in response to the company’s on random utility models (see Section 3.2 of Anderson
pricing policy, which is also dynamic. The resulting et al. 1992). The use of such models is pervasive in
model belongs to the general class of stochastic economics, marketing, and other fields. In a dynamic
dynamic games. setting under linear random utility, a consumer values
For convenience, the notation introduced here and a purchase at price p at time tAf0, 1, . . ., T  1g as
throughout the paper is summarized in a glossary in at1et  p, where at is a known constant describing
Appendix S1. Recall that there are T decision periods consumer perception of quality or value of the prod-
in our model. The initial inventory of the product at uct, and et is a random variable with mean zero
time 0 is Y, with no replenishment possible during the (usually assumed to be continuously distributed). The
planning period. At time T, the product expires and quantity p0 t ¼ at þ et formally corresponds to valua-
all unsold items are lost. tion of the product at time t. One modeling option
The product is offered to a finite number N of cus- would be to assume that for each consumer the val-
tomers, each of whom can purchase at most one item. uation is drawn from some known distribution at the
To ensure that probabilities in our discrete-time model beginning and then remains constant throughout
adequately represent intensities of the demand pro- the selling season. However, in reality, individual con-
cesses, we choose time discretization so that N l is sumer valuations cannot be measured precisely and
much o1. Without loss of generality, we assume that are unlikely to be certain. In general, an individual
 Levin, McGill, and Nediak: Dynamic Pricing with Strategic Consumers
46 Production and Operations Management 19(1), pp. 40–60, r 2009 Production and Operations Management Society

consumer valuation can be modelled over time as a  There is no customer ‘‘disutility’’ associated with
stochastic process. For example, one may assume it failing to purchase an item, and a customer with-
is a continuous-state random walk (Markov) process. out an item at time T derives a utility of zero.
By changing the degree of statistical dependence be-  The utility of buying an item in the future is dis-
tween consecutive states of this process, we get a counted by a factor bA[0, 1] per decision period.
range of possible models for valuations. At one end of The customer values future opportunities as the
this spectrum, we have a fixed (deterministic) model – expected value of bU(t11,  ) over all possible
an assumption made, for example, in deterministic states resulting from the current one.
models of Besanko and Winston (1990), Su (2007),
where it is completely appropriate. In a model like
ours, a finite-population model with a stochastic sales The degree of strategicity of a customer is determined
process, the fixed valuation model will lead to a dis- by the parameter bA[0, 1]. Indeed, the value b 5 0
continuous total demand intensity response to posted means that the customer completely disregards the
prices (intuitively, as the price increases, customers possibility of a future purchase; that is, the customer
will switch from shopping to non-shopping one by is myopic. The value b 5 1 means that the customer
one based on their fixed utility thresholds). Moreover, values the current purchase the same as a purchase at
as the company prices its product, it will also have any point in the future and will exhibit fully strategic
to learn the unknown and uncertain demand inten- behavior. Intermediate values of b determine how
sity profile. Such a problem falls into the class of long customers can postpone their purchase without
incomplete information control problems. In such a excessive loss of utility. Our use of identical charac-
formulation, the state description would have to in- teristics to describe different customers: b, u(  ), Ft(p)
clude the histories of price and sales processes, which and U(  ) implies that the pool of customers is homo-
would make the resulting control problem very geneous (in a stochastic sense). In practice, these
difficult. Consumer evaluation of a possible future parameters could be estimated from such sources as
purchase would have to depend on initial valuation as past sales history and marketing surveys. In e-com-
well as price and sales history, resulting in a major merce applications, where it is often possible to track
complication for a strategic consumer behavior model. behavior of individual customers, these parameters
An intermediate level of statistical dependence of can be estimated from consumer responses to prices
consecutive valuations is more realistic, but it does not over time. The maximum shopping intensity param-
resolve the issue of dealing with the demand-learning eter l could be estimated from the number of
problem. At the other end of the spectrum – complete individual consumer inquiries. Some approaches to
independence between valuations at different time – learning of customer behavior in this way are dis-
the demand-learning problem can be avoided if the cussed in a separate paper: Levina et al. (2009).
valuation distribution is known. This leads to the fol- The company’s objective is to maximize the ex-
lowing: pected total revenues by selecting the pricing policy in
Strategic choice model: Suppose that the current de- the class of state-feedback policies: p 5 p(t, n) for state
cision period is t, and the price announced by the n at time t. A customer’s objective is to maximize the
company is p; then, expected present value of utility of acquiring the
product. The objectives of the company and its cus-
 The consumer values the purchase that can occur tomers determine the payoff functions in a stochastic
in the current decision period as u(p 0  p), where dynamic game which unfolds over the decision peri-
p 0 is a possible valuation level and u(  ) is a ods tAf0, . . ., T  1g. If, at time toT, the system is in
strictly increasing utility function with an inverse state n4N  Y, the company announces the price
u  1(  ). An arbitrary utility reference value is p 5 p(t, n) (move of the company), the customers ob-
chosen so that u(0) 5 0. serve their valuation realization p 0 (which is different
 Consumer valuations are distributed according to for each consumer and unknown by other consumers),
Ft(p). Valuations of different customers at time and respond by choosing their shopping intensity
t are exchangeable random variables. from the interval ½0; l. The equilibrium shopping in-
 Valuations are independent random variables for tensity response function of a customer is denoted as
different t. lU(t, n, p, p 0 ), and its average over realizations of p 0 (the
 Each consumer’s value of future purchase oppor- average response that the company would expect in
tunities as perceived in the next decision period the present state) as lU(t, n, p). The game ends either at
is the same known function of time and state, time T regardless of the state, or if the company sells
U(t11, n). Consumers may assess this value out prior to time T (that is, in state n 5 N  Y). At the
exactly or heuristically depending on their ratio- end of the game all unsold items (if any) expire, and
nality level. all customers still in the market obtain a utility of 0.
Levin, McGill, and Nediak: Dynamic Pricing with Strategic Consumers
Production and Operations Management 19(1), pp. 40–60, r 2009 Production and Operations Management Society 47

We now turn to the nature of equilibrium in this the expected payoffs and equilibrium responses of all
game. customers are identical. We present the formal treat-
ment in the general case of exchangeable draws in
2.3. Equilibrium Appendix S1.
To arrive at a strategic decision concerning shopping Suppose that the current decision period is t and
intensity, the customer needs to know the expected consider a particular customer. A future purchase can
present value of utility from a possible future pur- only occur in one of the decision periods t11,
chase. This value may be computed exactly or . . ., T  1. The expected value of utility at time t11
approximately. The exact computation requires the from a possible future purchase, given that the state at
customers to know the price which will be used in time t11 is n, is assessed by each customer as
every possible state in the future. This is possible if we U(t11, n). Given strategicity parameter b, the expected
suppose that customers are fully rational and can present utility (at time t) from a possible future pur-
compute the optimal pricing policy used by the com- chase is then bU(t11, n). At the time of decision, the
pany or if the customers can resort to third-party customer has also observed the posted price p and
services. An alternative is to assume partial rational- his/her own valuation level p 0 . We denote the deci-
ity: customers use a given heuristic rule to estimate sion variable as l 2 ½0; l. Since valuation realizations
the expected present value of utility from a possible of all other customers are identically distributed and
future purchase. In either case, we assume independent of the value observed by the customer,
Rationality: The company is fully rational and their shopping intensities can be estimated by the
knows how customers evaluate future purchasing customer at the expected value lU(t, n, p). Given
possibilities. We further assume that all participants of t, n, p, p 0 , the value function VU(t, n, p, p 0 ) is the ex-
the game assume the optimality (in an exact or a heu- pected present value of utility over all possible
ristic sense) of the future strategies of all opponents purchase events at time t (recall that lf and lU(t, n, p)
(prices and customer responses), and each customer are probabilities):
assumes that the concurrent responses of other cus-
tomers are exactly optimal. VU ðt; n; p; p0 Þ ¼ max fluðp0  pÞ þ bðn  1ÞlU ðt; n; pÞ
0ll
The sequence of moves at each t described in the
previous subsection corresponds to a hierarchical  Uðt þ 1; n  1Þ þ bð1  l  ðn  1Þ
equilibrium structure with the company as the leader  lU ðt; n; pÞÞUðt þ 1; nÞg:
and the customers as the followers. The equilibrium at
each t can be described as a Stackelberg equilibrium The first term in the above expression refers to the
between the company and the customers and a Nash certain utility that will be realized if the customer
equilibrium with asymmetric information (since each completes the purchase now. The other terms repre-
customer knows their own current valuation but not sent the expected present value of utility when a
those of others) between the customers. According to purchase by another customer occurs (second term) or
our rationality assumptions, the behavior of the com- does not occur (third term). After collecting terms, this
pany is always time consistent. We can make the expression can be rewritten as
customer response time-consistent as well under full
V U ðt; n; p; p0 Þ ¼ max flðuðp0  pÞ  bUðt þ 1; nÞÞg
customer rationality. Thus, under full rationality, we 0ll
have a subgame perfect equilibrium with asymmetric
þ bðn  1ÞlU ðt; n; pÞðUðt þ 1; n  1Þ
information. The equilibrium between the customers
at each t can also be viewed as a variation of a quantal  Uðt þ 1; nÞÞ þ Uðt þ 1; nÞ:
response equilibrium, see McKelvey and Palfrey (1995,
From the linearity of the objective in l, we conclude
1998), because the payoff of each customer is modu-
that the customer optimal shopping intensity is given
lated by a random perturbation drawn from the
by the expression
valuation distribution, and decisions of each customer
can be simplified to a binary choice between shopping lU ðt; n; p; p0 Þ ¼ lI ½uðp0  pÞ  bUðt þ 1; nÞ: ð1Þ
now (with the maximum possible intensity) or wait-
ing (that is, shopping with intensity 0). The interpretation of this result is straightforward:
We now derive the objective function for a cus- the hypothetical customer will shop with the maxi-
tomer and the equilibrium decision in terms of the mum possible intensity of l whenever the utility of
shopping intensity. This derivation is for a simpler the valuation/price difference is greater than or equal
case in which valuation draws across different cus- to the present value of utility of a possible future
tomers are independent and identically distributed purchase. The condition inside the indicator function
instead of exchangeable (which admits some depen- can also be expressed as p 0  p1u  1(bU(t11, n)),
dency). Under this assumption, it is easy to see that where u  1(  ) is the inverse of u(  ).
 Levin, McGill, and Nediak: Dynamic Pricing with Strategic Consumers
48 Production and Operations Management 19(1), pp. 40–60, r 2009 Production and Operations Management Society

We thus have a simple rule for how customers Rðt; nÞ ¼Rðt þ 1; nÞ þ max
p
choose their shopping intensities after observing their  U
private valuations, but we must also specify how  nl ðt; n; pÞðp þ Rðt þ 1; n  1ÞRðt þ 1; nÞÞg;
shopping intensity can be assessed without knowl- n ¼ N  Y þ 1; . . . ; N; t 2 f0; . . . ; T  1g;
edge of those valuations. This is needed by the ð4Þ
company, by customers considering other customers
(in the fully rational case), and by customers assessing with the terminal conditions
their own behavior in the future. To accomplish this, RðT; nÞ ¼ 0; n ¼ N  Y þ 1; . . . ; N; and ð5Þ
we take expectations of the optimal shopping inten-
sity and the corresponding objective value over the Rðt; N  YÞ ¼ 0; t 2 f0; . . . ; T  1g: ð6Þ
valuation distribution Ft(p) and obtain the following
Given the rule U(  ) for consumer evaluation of
proposition (where we let VU ðt; n; pÞ ¼ Ep0 ½VU ðt; n;
expected utility from a possible future purchase, pU(  )
p; p0 Þ):
and lU(  ) describe the game equilibrium.
One particularly interesting case results when U(  )
PROPOSITION 1. Given the evaluation of the expected utility
corresponds to fully rational customers. It can be
from a possible future purchase U(t11,  ), which is iden-
obtained by
tical for all remaining customers, the expected equilibrium
shopping intensity of any customer observing price p in Uðt; nÞ ¼V U ðt; n; pU ðt; nÞÞ; n ¼ N  Y þ 1; . . . ; N;
state n 5 1, . . ., N at time t is given by t 2 f0; . . . ; T  1g
 
lU ðt; n; pÞ ¼  t p þ u1 ðbUðt þ 1; nÞÞ :
lF ð2Þ and the terminal conditions on the customer utility
UðT; nÞ ¼ 0; n ¼ N  Y þ 1; . . . ; N ð7Þ

The expected present value of utility for each remaining Uðt; N  YÞ ¼ 0; t 2 f0; . . . ; T  1g: ð8Þ
customer is given by We will denote the expected utility for this case as
Z þ1 U ðÞ and the corresponding equilibrium policies as
 þ
U
V ðt; n; pÞ ¼l fðuðp0  pÞ  bUðt þ 1; nÞÞg dFt ðp0 Þ p ðÞ and l ðÞ.
0 If, in addition to the pricing policy, the company can
þ bððn  1ÞlU ðt; n; pÞðUðt þ 1; n  1Þ control its initial capacity (procurement) decision, it
 Uðt þ 1; nÞÞ þ Uðt þ 1; nÞÞ; can use the optimal revenue at time 0 for fixed N but
different values of Y to make the optimal capacity
ð3Þ
choice. For the time being, however, we assume that
where, for any real x, its positive part fxg1 5 x if x  0 Y is fixed and fully concentrate on the pricing deci-
and fxg1 5 0 if xo0. sions of the company. We return to the question of the
An important observation from this analysis is that optimal capacity decision in Section 5 where we nu-
the customer reaction to price given in Equation (2) merically examine the effects of customer strategicity.
depends on the state n only through his/her evalu-
ation of the future given by U(t11, n). Thus, for an 3. Existence and Uniqueness of
exogenously specified U, the result contained in Equilibrium
Equation (2) can be generalized to the case in which A typical approach in the analysis of dynamic pricing
the customer is uncertain about state information: we models is to regard the aggregate demand rather than
would replace U(t11, n) by the expected value over n. the price as a decision variable for the company.
We do not pursue this generalization here but note In our case, it is also convenient to change the decision
that this would allow extension to a variety of limited- variable of the company to a quantity closely related
rationality cases. to the aggregate demand. We note that a particular
The company can use the equilibrium consumer price p in state n at time t results in the following
shopping intensity in its policy optimization. The eagerness to purchase averaged over the valuation dis-
probability of a sale occurring in the current decision tribution:
period and state n is equal to nlU(t, n, p). The expres-  
w¼F t p þ u1 ðbUðt þ 1; nÞÞ :
sion for the optimal expected revenue of the company
is obtained in the same fashion as the expression for The purchase intensity corresponding to w is lU ðt;
the optimal expected customer’s utility. The com- n; pÞ ¼ lw:
pany’s equilibrium pricing strategy pU(  ) is chosen so To use eagerness w as a policy variable, we require
that pU(t, n) delivers the maximum expected future t ðÞ to have an inverse function gt(  ). The value gt(w)
F
revenues: of the inverse function specifies the quantile of the
Levin, McGill, and Nediak: Dynamic Pricing with Strategic Consumers
Production and Operations Management 19(1), pp. 40–60, r 2009 Production and Operations Management Society 49

valuation distribution at time t corresponding to the Defining

right-tail probability w. To this end, we will make the Z þ1  þ
following assumption regarding gt(w): Ht ðw; vÞ ¼ uðp0  gt ðwÞ þ u1 ðvÞÞ  v dFt ðp0 Þ;
0
ASSUMPTION (A). Let the valuation probability density we obtain the following relation
function be strictly positive and continuous on a (possibly  
p; 
infinite) interval ð pÞ, and zero outside the interval. V U ðt; nÞ ¼
lHt wU ðt; nÞ; bUðt þ 1; nÞ
 blðn  1ÞwU ðt; nÞDUðt þ 1; nÞ þ bUðt þ 1; nÞ;
Then the tail distribution function F t ðÞ is decreas-
n ¼ N  Y þ 1; . . . ; N; t 2 f0; . . . ; T  1g:
ing and has an inverse function gt(  ), which is
continuously differentiable and decreasing. The do- ð10Þ
p; 
main of gt(  ) is (0, 1), and its range is ð pÞ.
This relation becomes recursive in the fully rational
If Assumption (A) holds, then we can express p as
consumer case where Uðt; nÞ ¼ U ðt; nÞ ¼ VU ðt; nÞ for
the following continuously differentiable function
all n ¼ N  Y þ 1; . . . ; N and t ¼ 0; . . . ; T  1, and the
of w:
terminal conditions are given in (7) and (8).
p ¼ gt ðwÞ  u1 ðbUðt þ 1; nÞÞ: Even under full rationality, the computational effort
(required to find the optimal policy in our strategic-
In the sequel, we use the following abbreviated customer model in this parametrization) is only mod-
notations for the quantities marginal in n: estly greater than the computational effort needed in a
time-discretization of the myopic-customer model of
DRðt; nÞ ¼ Rðt; nÞ  Rðt; n  1Þ; Gallego and van Ryzin (1994) with the same initial
DUðt; nÞ ¼ Uðt; nÞ  Uðt; n  1Þ; inventory level. The increase results entirely from the
Du1 ðbUðt; nÞÞ ¼ u1 ðbUðt; nÞÞ  u1 ðbUðt; n  1ÞÞ: need to solve recursion (10) and is modest as long as
the function Ht(w, v) can be evaluated efficiently.
The dynamic programming recursion (4) for the Thus, our strategic customer model is computation-
expected future revenues of the company can now be ally tractable for problems of realistic size. This result
rewritten in terms of eagerness w as follows: could be achieved because of our reliance on time-
consistency and a hierarchical structure of the equi-
Rðt; nÞ ¼Rðt þ 1; nÞ þ n
l max w librium.
0w1
 1 Using the dynamic programming recursion (9) for
 gt ðwÞ  u ðbUðt þ 1; nÞÞDRðt þ 1; nÞ;
the company, we arrive at the following existence and
n ¼ N  Y þ 1; . . . ; N; t 2 f0; . . . ; T  1g: uniqueness result (the proof of this and all subsequent
ð9Þ formal statements is provided in the Appendix S1).
The result applies to both the full U ¼ U and limited
The value of the expression under the maximum is consumer rationality cases (exogenous U). It also
defined at 0 and 1 in the limiting sense. An optimal gives the equilibrium pricing policy of the company.
policy value wU(t, n) delivers the maximum in this We make two additional assumptions:
expression. Note the role of the u  1(bU(t11, n)) term.
When the customers are myopic, this term is equal to ASSUMPTION (B). wgt(w) is strictly concave for all t, and
zero, and the price corresponding to w is given by
p 5 gt(w). The quantity u  1(bU(t11, n)) can then be ASSUMPTION (C). limw ! 0 wgt(w) 5 0 for all t,
viewed as a state and time-dependent transaction
cost. This is a portion of the total surplus the company which correspond to the requirements of concavity
has to give up to the consumer due to his/her stra- and regularity of the revenue rate in Gallego and van
tegicity. Ryzin (1994). These assumptions are satisfied for such
Next, we use Equation (3) and the unique customer common distributions as exponential F t ðpÞ ¼ expðpÞ,
response (2) to derive an expression for the equilib- and power F t ðpÞ ¼ ð1 þ pÞg ; g41 (in these cases,
rium expected present value of utility VU(t, n) in terms gt(w) 5  ln(w) and gt(w) 5 w  l/g  1, respectively).
of wU(t, n):
PROPOSITION 2. Suppose that the valuation distribution
V U ðt; nÞ ¼V U ðt; n; pU ðt; nÞÞ
Z þ1 satisfies assumptions (A), (B) and (C). Then there exists a
 0 unique equilibrium such that, for every (t, n), the customer
¼l uðp  gt ðwU ðt; nÞÞ þ u1 ðbUðt þ 1; nÞÞÞ
0 eagerness to purchase, wU(t, n), is
bUðt þ 1; nÞgþ dFt ðp0 Þ
 zero if limw!0 ðw g0 t ðwÞ þ gt ðwÞÞ  u1
 b
lðn  1ÞwU ðt; nÞDUðt þ 1; nÞ þ bUðt þ 1; nÞ: ðbUðt þ 1; nÞÞ þ DRðt þ 1; nÞ,
 Levin, McGill, and Nediak: Dynamic Pricing with Strategic Consumers
50 Production and Operations Management 19(1), pp. 40–60, r 2009 Production and Operations Management Society

 one if limw!1 ðw g0 t ðwÞ þ gt ðwÞÞ  u1 u1 ðbUðt þ 1; nÞÞ þ DRðt þ 1; nÞ  ðÞ u1 ðbUðt þ 2; nÞÞ
ðbUðt þ 1; nÞÞ þ DRðt þ 1; nÞ, þ DRðt þ 2; nÞ
 and, otherwise, satisfies the equation
holds if and only if wU ðt; nÞ  ðÞ wU ðt þ 1; nÞ.
wU ðt; nÞg0t ðwU ðt; nÞÞ þ gt ðwU ðt; nÞÞ
ð11Þ
¼ u1 ðbUðt þ 1; nÞÞ þ DRðt þ 1; nÞ: This result sheds some light on the consequences of
consumer rationality for monotonic properties of the
The corresponding optimal price is
company’s policy. One of the common monotonicity
pU ðt; nÞ ¼ gt ðwU ðt; nÞÞ  u1 ðbUðt þ 1; nÞÞ: ð12Þ results in dynamic pricing models is that the marginal
revenues decrease in capacity; that is, DR(t11, n)
 DR(t11, n  1). On the other hand, intuition sug-
Given the customer perception of the future spec-
gests that, for rational consumers, inequality
ified by U, the equilibrium eagerness value wU(t, n)
U(t11, n)  U(t11, n  1) should typically hold since
specifies the optimal customer reaction to the equi-
the customer competition for remaining items be-
librium price pU(t, n). When assessing their reaction,
comes more acute when there is one less item
customers generally take into account how many
available (even though there is also one less cus-
items are still available as well as competition from
tomer). Because this inequality for utility is in the
other customers as summarized by the state descrip-
opposite direction to the inequality for marginal rev-
tion n. Thus, the game developed here is a dynamic
enues, we conclude that consumer rationality works
Stackelberg game between the company and a homo-
against typical monotonicity properties. Some excep-
geneous population of individual consumers. If
tions to this are possible if U(t11, n) is independent of
customers are unaware of the state, the equilibrium
n. The reader will see that this intuitive conclusion
takes a somewhat simpler form in which the company
is confirmed in the structural results and numerical
plays Stackelberg games with customers who make
experiments to follow. As a final remark about Prop-
decisions independently from each other. However,
osition 2, we show that certain monotonic properties
for the company, these games are linked through the
of prices (in contrast to the result of Corollary 1 which
limited inventory. To interpret (11), observe that, on
is in terms of eagerness) can be established under the
the left-hand side, wg0 t ðwÞ þ gt ðwÞ is the marginal rate
following assumption:
of revenue per customer while, on the right-hand side,
marginal expected revenue per item DR(t11, n) can
ASSUMPTION (D). wg0 t ðwÞ is increasing for all t.
be viewed as an opportunity cost of capacity. A stan-
dard interpretation of optimality conditions in
This assumption is also satisfied for exponential
dynamic pricing is that the marginal revenue rate
and power distributions that we mentioned earlier
equals the opportunity cost of capacity (see, for ex-
(after Assumptions (B) and (C)).
ample, the interpretation of equation (5.13) on page
203 of Talluri and van Ryzin (2004)). In our model, this
COROLLARY 2. Let Assumption (D) hold in addition to (A),
is modified by addition of the certainty equivalent
(B), and (C). Then the pair of inequalities
from a possible future purchase u  1(bU(t11, n)),

which we interpreted as a cost of transaction with DRðt þ 1; nÞ ðÞDRðt þ 1; n  1Þ;
strategic customers. wU ðt; nÞ ðÞwU ðt; n  1Þ;
The following corollary immediately follows from
the optimality conditions described in the above implies pU ðt; nÞ  ðÞ pU ðt; n  1Þ. If, moreover, gt(  ) is
proposition and the (strictly) decreasing property of stationary (independent of t), then the pair of inequalities
wg0 t ðwÞ þ gt ðwÞ in w. It relates monotonic properties of 
DRðt þ 1; nÞ ðÞDRðt þ 2; nÞ;
the optimal policy to those of u  1(bU(t11, n))1
DR(t11, n) (the sum of the certainty equivalent of wU ðt; nÞ ðÞwU ðt þ 1; nÞ;
purchasing in the future and the marginal future ex- implies pU ðt; nÞ  ðÞ pU ðt þ 1; nÞ.
pected revenues).
To explain this result, we rewrite Equation (12) us-
COROLLARY 1. The relation ing Equation (11) as pU ðt; nÞ ¼ DRðt þ 1; nÞ  wU ðt; nÞ
u1 ðbUðt þ 1; nÞÞ þ DRðt þ 1; nÞ  ðÞ g0t ðwU ðt; nÞÞ. Since, under strategic consumer behavior,
 u1 ðbUðt þ 1; n  1ÞÞ þ DRðt þ 1; n  1Þ wU(t, n) is generally time and state dependent, appro-
priate monotonic relations on both DR(t11, n) and
holds if and only if wU ðt; nÞ  ðÞ wU ðt; n  1Þ. In addi- wU(t, n) are needed to guarantee corresponding mono-
tion, if gt(  ) is stationary (independent of t) then the tonic relations between prices. On the other hand, in
relation the case of myopic consumers it suffices to have
Levin, McGill, and Nediak: Dynamic Pricing with Strategic Consumers
Production and Operations Management 19(1), pp. 40–60, r 2009 Production and Operations Management Society 51

monotonic relations on DR(t11, n). In managerial lish that the equilibrium customer utility is
terms, an interpretation of the first claim of the cor- nonnegative: VU(t, n)  0, since we can regroup the
ollary is that, if under equilibrium policies a marginal terms in Equation (10). For U , we can establish the
value of an extra sale opportunity (one extra item plus nonnegativity by reverse induction from terminal
one extra customer) decreases and the shopping in- conditions (7) and (8).
tensity of an individual customer increases with n,
then the price decreases with n. Similar interpreta- 4.1. Special Case: State Unaware, Limited
tions apply to the second claim and also if inequalities Rationality Consumers
are reversed. We analyze the case of a finite population of limited
As a reality check of our model, we have also ex- rationality customers whose evaluation of the future
amined what happens if the original problem is scaled is such that bU(t, n) is independent of the state n for
to an arbitrarily large population size, with capacity every t. Note that myopic consumer behavior, which
proportionally large but still smaller than the popu- is a special case of our model when b 5 0, satisfies this
lation. Under such scaling, we also need to take the restriction. The following proposition establishes that
number of decision periods to infinity. Through the the marginal revenues (with respect to the number of
law of large numbers, the demand process becomes remaining customers) are decreasing with respect to n
increasingly deterministic from the point of view of and t. These properties generally coincide with those
the company, but each consumer still faces a poten- found in many myopic infinite population models
tial rationing risk because of insufficient capacity. (see, for example, Proposition 5.2 on page 203 of Tall-
In Appendix S1, we show that in the limit the price uri and van Ryzin (2004)). We also remark that this
optimality conditions in our discrete-time model cor- monotonicity result provides a bound on the decre-
respond to a Hamilton–Jacobi–Bellman equation for a ment of DR(t, n) with respect to t:
deterministic continuous time control problem. Thus,
we find that the stochastic discrete-time model has a PROPOSITION 3. Let U be such that bU(t, n) does not depend
natural deterministic continuous-time analog. on n for each t. Then we have
DRðt; nÞ  DRðt; n  1Þ; t ¼ 0; . . . ; T; n
4. Monotone Properties of the Optimal ¼ N  Y þ 2; . . . ; N; ð13Þ
Policy 
In this section, we study monotone properties of the DRðt; nÞ  DRðt þ 1; nÞ þ lwU ðt; nÞ gt ðwU ðt; nÞÞ
optimal policy. We show that the optimal price and 
u1 ðbUðt þ 1; nÞÞ  DRðt þ 1; nÞ ; ð14Þ
marginal revenues possess monotone properties in
two special cases: first, when customers have limited t ¼ 0; . . . ; T  1; n ¼ N  Y þ 1; . . . ; N:
rationality and evaluate the expected utility from The first inequality can also be restated as D2R(t, n)
future purchases independently of the state (this  0, and implies concavity of the expected revenues
includes myopic case), and, second, when customers in n.
are strategic but do not need to compete for the items The following corollaries establish monotonic prop-
since the inventory is sufficient to satisfy all custom- erties of the policy variables in the state unaware,
ers. Our results show that under these conditions the limited rationality case. The first corollary shows that
monotonic properties well-known in the infinite pop- the customer eagerness to purchase increases with
ulation myopic case hold in our model as well. We respect to n, and that prices pursue the opposite
also present a numerical illustration which shows that behavior:
the price may not be monotone in general. This is in
contrast with the results in dynamic pricing literature
for the myopic consumer case. In practical terms, this COROLLARY 3. If U is such that bU(t, n) does not depend on
suggests that the company may have to deviate from n for each t, then the following inequalities hold for all
monotonic pricing policies to counteract strategic con- t 5 0, . . ., T  1, and all n 5 N  Y12, . . ., N:
sumer behavior. All results of this section are obtained wU ðt; nÞ  wU ðt; n  1Þ; ð15Þ
under Assumptions (A)–(C).
We start our discussion with basic observations pU ðt; nÞ  pU ðt; n  1Þ: ð16Þ
which apply to the most general form of our model.
First, note that the optimal expected future revenues
are decreasing in time: R(t, n)  R(t11, n), since the Recall that smaller n also means lower inventory
maximum of expression (28) under maximization of level y; thus, this property is natural because, under
the recursive relation (9) for revenues is nonnegative. otherwise identical market conditions, our intuition
Given that U  0, it is also straightforward to estab- suggests increasing price if there are fewer items left
 Levin, McGill, and Nediak: Dynamic Pricing with Strategic Consumers
52 Production and Operations Management 19(1), pp. 40–60, r 2009 Production and Operations Management Society

to sell. Our model provides an additional insight that DRðt; nÞ ¼DRðt þ 1; nÞ þ 

lw ðt; nÞ½gt ðw ðt; nÞÞ
the price increases in spite of the fact that the market 1

u ðbU ðt þ 1; nÞÞDRðt þ 1; nÞ ; ð21Þ
is also smaller after a sale.
t ¼ 0; . . . ; T  1; n ¼ N  Y þ 1; . . . ; N;
The next corollary establishes monotonic properties
of the pricing policy with respect to time in the case of
myopic customers and a stationary valuation distri- DRðt; nÞ ¼ DRðt; n  1Þ; t ¼ 0; . . . ; T; n
bution. An additional assumption made is that the ¼ N  Y þ 2; . . . ; N; ð22Þ
consumer evaluation of expected utility from a pos-
sible future purchase decreases in time (a natural U ðt; nÞ ¼ U ðt; n  1Þ; t ¼ 0; . . . ; T; n
property for a heuristic utility evaluation rule). ¼ N  Y þ 2; . . . ; N: ð23Þ

The result of this proposition is that the equilibrium

COROLLARY 4. If U is such that bU(t, n) does not depend on price, the consumer reaction, and the expected utility
n, bUðt; Þ  bUðt þ 1; Þ for all t, and gt(  ) 5 g(  ), then are independent of the state, while the expected rev-
the following inequality holds for all t ¼ 0; . . . ; T  2 and enues increase linearly with the state (thus, the
n ¼ N  Y þ 1; . . . ; N revenue is concave in the state here as well). Such
equilibrium structure is natural since there is no com-
wU ðt; nÞ  wU ðt þ 1; nÞ ð17Þ petition between consumers, and they only need to
take into account time remaining to the end of the
with inequality being strict if wU ðt þ 1; nÞ40: In addition, planning horizon. Equation (21) also implies that the
under Assumption (D) or in the myopic case of b 5 0 marginal expected revenues decrease in time.
Based on the above proposition, we also obtain a
pU ðt; nÞ  pU ðt þ 1; nÞ: ð18Þ time monotonicity result for the equilibrium eager-
ness to purchase and price in the special case of fully
strategic customers.
This corollary asserts that, under otherwise iden-
tical market conditions, consumer eagerness to pur-
COROLLARY 5. In the case of b 5 1 N 5 Y and gt(  ) 5 g(  ),
chase increases when there is less time remaining. It
for all t ¼ 0; . . . ; T  2 and n ¼ 1; . . . ; N
is much less obvious that the price is monotonic since
there are generally two opposite effects: gðwU ðt; nÞÞ  w ðt; nÞ  w ðt þ 1; nÞ: ð24Þ
gðwU ðt þ 1; nÞÞ and u1 ðbUðt; nÞÞ  u1 ðbUðt þ 1;
nÞÞ, and it is not clear which one dominates. When In addition, under Assumption (D)
b 5 0, the last inequality holds as an equality which p ðt; nÞ  p ðt þ 1; nÞ: ð25Þ
implies monotonicity of prices.
4.3. General Case: Numerical Illustrations
4.2. Special Case: Fully Rational Strategic The monotonicity properties discussed above for spe-
Consumers Under High-Supply Market cial cases do not hold for more general strategicity
Next, we analyze the case of fully rational strategic or inventory level conditions. We illustrate this with
consumers in a ‘‘high product supply’’ situation, numerical counterexamples below. In these examples,
when there are enough items in inventory to satisfy we use a risk-neutral utility uðp0  pÞ ¼ p0  p and a
the purchasing requirements of all customers. The stationary exponential tail distribution of the valua-
results imply that the game at each time t and state tion Ft ðp0 Þ ¼ expðp0 Þ: We solve the model for
n decomposes into n identical games played simul- T 5 10,000 decision periods, N 5 100 customers and
taneously between the company and each of the maximum individual shopping intensity l ¼ 0:005.
remaining customers individually. This is possible be- The purpose of choosing a large value for T is pri-
cause customers are statistically identical. marily to provide a sufficiently close approximation to
continuous time.
PROPOSITION 4. In the case of N 5 Y, the following equalities
hold 4.3.1 Monotonicity with Respect to Time. Our first
numerical experiment concerns the non-monotonicity
w ðt; nÞ ¼ w ðt; n  1Þ; t ¼ 0; . . . ; T; n of price in time when the number of customers is
¼ N  Y þ 2; . . . ; N; ð19Þ fixed. The initial inventory level for this experiment is
fixed at Y 5 75 items. Five values of the strategicity
parameter b are examined: zero (myopic), one (fully
strategic), and three intermediate values such that the
p ðt; nÞ ¼ p ðt; n  1Þ; t ¼ 0; . . . ; T; n value of a purchase after 10,000 time steps is 0.1, 0.01,
¼ N  Y þ 2; . . . ; N; ð20Þ and 0.0001 of the value of a purchase if it is made
Levin, McGill, and Nediak: Dynamic Pricing with Strategic Consumers
Production and Operations Management 19(1), pp. 40–60, r 2009 Production and Operations Management Society 53

Figure 1 
The Optimal Price p ðt ; 100 Þ, Future Expected Revenues R(t, 100) and the Booking Curve k100w ðt ; 100 Þ as Functions of Time for Y = 75, N = 100
and Different Values of b
5 300
4.5

Expected future
250
4

revenues
3.5 200
Price

3 150
2.5 100
2
1.5 50
1 0
0 2000 4000 6000 8000 10 000 0 2000 4000 6000 8000 10 000
Time Time

0.1
 = 1.0
Booking curve

0.08 T =0.1
0.06 T =0.01
T =0.0001
0.04  = 0.0
0.02
0
0 2000 4000 6000 8000 10 000
Time

immediately. These values were chosen because of the as well although not as prominently as for p ðt; 100Þ
logarithmic effects of strategicity on the optimal (see Figure 1). Another interesting feature observed
policies, and produce graphs which fill the visual from this plot is that the shopping intensity is not
gaps between myopic and fully strategic customers monotonic in b. In the beginning of the interval, the
quite well. Figure 1 shows the optimal price p ðt; 100Þ, shopping intensity is almost the same for b 5 0 and
the optimal expected revenues R(t, 100) and the 1, while being slightly larger for the intermediate
‘‘booking curve’’ for the state with 100 customers values of b. A possible explanation is that both the
and all five values of b. The graph of the optimal price myopic and fully strategic (b 5 1) customers in
for the myopic case is, of course, monotone decreas- the beginning of the planning interval do not care
ing, in full agreement with Corollary 4. However, as when exactly their purchases occur. On the other
the value of b increases, we see a tendency towards hand, customers with intermediate values of b effec-
non-monotonic behavior. For b 5 1, the change in tively have less time than T to make sure the item is
direction can be seen very clearly and takes place acquired than the customer with b 5 1 who can take
around time steps t 5 6200. . .7200. A possible expla- advantage of the whole planning interval.
nation of this non-monotonic behavior is that the
supply level Y/N 5 75% is relatively high. Therefore, 4.3.2 Monotonicity with Respect to n. The second
in the beginning of the interval, the state n 5 100 will experiment examines the monotonic behavior of
provide a plentiful opportunity for the customer to prices with respect to n at fixed time t. The initial
buy. As the expiration date gets closer, the customers inventory Y is fixed, which implies that for smaller
will have fewer possibilities to purchase. This allows values of n we look at respectively smaller values of
the company to increase prices even if all other the remaining inventory y 5 Y  (N  n). Four values of
parameters of the model stay the same. Note that this Y were examined corresponding to the supply levels
is possible only because the customers are strategic Y/N of 25%, 50%, 75%, and 100%. The strategicity
and aware of the reduction in future purchase parameter b is fixed at 1. The graphs of the optimal
opportunities. Fully rational behavior of the com- price p ðt; nÞ as functions of n are shown in Figure 2
pany ensures that a price increase is a credible threat, for fixed values of t 5 0 (the decision period near
thus stimulating consumers to shop earlier. which p ðt; nÞ is decreasing in t, a natural monotonic
Another observation from this plot is that the op- behavior) and t 5 7000 (where we observed some
timal price tends to decrease in b. Naturally, the deviations from monotonic behavior). The value of p
same can be expected from the optimal expected ðt; nÞ for the supply level of 100% remains constant in
revenues. We can see this type of behavior of agreement with Proposition 4. For lower values of the
R(t, 100) in Figure 1. Recall that the total shopping supply level, the value of p ðt; nÞ is decreasing in n at
intensity of n customers at time t is given by  lnw the beginning of the planning period (t 5 0). This is
ðt; nÞ: This can be viewed as a ‘‘booking curve’’ over not the case for t 5 7000 and the supply level of 75%.
time. Non-monotonic behavior in time is seen here We recall that the price is also non-monotone in t for
 Levin, McGill, and Nediak: Dynamic Pricing with Strategic Consumers
54 Production and Operations Management 19(1), pp. 40–60, r 2009 Production and Operations Management Society

Figure 2 The Optimal Price p ðt ; nÞ as a Function of the Number of Remaining Customers n for b = 1, fixed t = 0, t = 7000 and Different Values of Initial
Supply Level
t=0 t = 7000
8 8
7 SupplyLevel 7 SupplyLevel
100% 100%
6 75% 6 75%
Price

Price
5 50% 5 50%
4 25% 4 25%

3 3
2 2
1 1
10 20 30 40 50 60 70 80 90 100 10 20 30 40 50 60 70 80 90 100
Number of Customers Number of Customers

this supply level around t 5 7000. For supply levels of price level of about 2.3 with progressively larger price
25% and 50%, the prices are monotonic for both fluctuations as the end of selling season gets closer. At
values of t. s 5 75%, we see a steady decline in prices. Even the
starting price is just slightly above 2 and there is
4.3.3 Typical Price Path Realizations. In addition practically no price fluctuation around the average
to monotonic properties of policies themselves, it is price path. This suggests that appropriate choice of
interesting to examine typical realizations of these capacity level is especially important in the presence
policies – the price paths, since this is how policies are of strategic consumers.
perceived in the market place. Therefore, we have The non-monotonicity of prices for the high supply
simulated 10,000 realizations of pricing policies for level also suggests that we may discover interesting
myopic and fully strategic (b 5 1) consumers at behavior in expected revenues when the values of Y
supply levels of 50% and 75%. The averages of and N are close. A numerical experiment addressing
simulated prices plus/minus one standard deviation this and other issues is described in the next section.
for all four combinations are shown in Figure 3. In the
myopic case, increased supply level leads to reduced
price flexibility (as evidenced by a smaller standard
5. Effects of Strategic Consumer
deviation) and significantly lower prices in the second Behavior on Company Performance
half of the planning horizon. When consumers are and Decisions
strategic, the effect of increased supply level is even In this section, we examine the interplay between
more dramatic. At s 5 50%, there is relatively steady strategicity and company performance and policies.

Figure 3 Averages of Simulated Price Paths for Myopic/Fully Strategic Consumers, Supply Levels 50%/75%, N = 100 and T = 10,000

Supply level - 50% / Customers - myopic Supply level - 75% / Customers - myopic
5 5
4 4
3 3
Price

Price

2 2
Average Average
1 Average±StDev 1 Average±StDev
0 0
0 2000 4000 6000 8000 10000 0 2000 4000 6000 8000 10000
Time Time

Supply level - 50% / Customers - strategic Supply level - 75% / Customers - strategic
5 5
4 4
3 3
Price

Price

2 2
Average Average
1 1 Average±StDev
Average±StDev
0 0
0 2000 4000 6000 8000 10000 0 2000 4000 6000 8000 10000
Time Time
Levin, McGill, and Nediak: Dynamic Pricing with Strategic Consumers
Production and Operations Management 19(1), pp. 40–60, r 2009 Production and Operations Management Society 55

We start by assuming that the initial capacity Y is The utility values are computed recursively accord-
specified exogenously (for example, a flight capacity is ing to
determined by the aircraft following a pre-determined
schedule) and, in Section 5.1, study strategicity effects
0
Ub;s ðt; nÞ ¼
lHt w0b;s ðt; nÞ; bUb;s
0
ðt þ 1; nÞ
on the pricing policies, expected revenues, and utilities.  b
lðn  1Þw0b;s ðt; nÞDUb;s
0 0
ðt þ 1; nÞ þ bUb;s ðt þ 1; nÞ;
In particular, we examine performance of pricing pol- n ¼ N  Y þ 1; . . . ; N; t 2 f0; . . . ; T  1g:
icies obtained under the assumption that customers are
ð26Þ
myopic in a situation when customers are in fact fully
rational and strategic. This examination is done by The expected future revenues in the situation when
testing the effect of company deviation from the equi- the myopic policy p0;s ðÞ is used for customers with
librium decisions. We assume that the customers are arbitrary b is denoted by R0b;s ðÞ, and can be computed
aware of the fact that the company is behaving non- via
optimally, can compute their actual expected utility in
this situation exactly, and adjust their response to the R0b;s ðt; nÞ ¼R0b;s ðt þ 1; nÞ þ nlw0b;s ðt; nÞ
h
observed prices accordingly. Major observations made
gt ðw0b;s ðt; nÞÞ  u1 bUb;s 0
ðt þ 1; nÞ
in this subsection include much lower or even negative i
marginal revenue in terms of extra product supply for DR0b;s ðt þ 1; nÞ ; n ¼ N  Y þ 1; . . . ; N;
supply levels close to 100%, possibly high costs of
assuming that customers are myopic when they are t 2 f0; . . . ; T  1g:
strategic, and a description of conditions when these
The derivation of these recursions is straight-
costs are low.
forward and closely follows that of Equations (9)
In Section 5.2, we relax the assumption of exoge- 0
nous Y and study the effects of strategicity on the and (10). The boundary conditions for Ub;s ðÞ and R0b;s
initial capacity decisions. In particular, we observe ðÞ are identical to those for Ub;s ðÞ and Rb;s ðÞ (equa-
importance of appropriate initial capacity reduction to tions of the same form as (7), (8), and (5), (6),
counteract strategic consumer behavior, possibly sig- respectively).
nificant profit reduction even if strategic behavior is For the numerical experiments described below, we
correctly taken into account, and truly disastrous use a risk-neutral utility uðp0  pÞ ¼ p0  p and a sta-
effects on profits if strategicity is ignored in procure- tionary exponential tail distribution of the valuation
t ðp0 Þ ¼ expðp0 Þ: We solve the model for N 5 100
F
ment decisions. We also study contribution of pricing
vs. capacity decisions to the proper company response customers, the maximum individual shopping inten-
to strategicity. sity of l ¼ 0:001 per time step and a time horizon of
T 5 1,000,000. In addition to b 5 0 and b 5 1, we ex-
5.1. Effects on Revenues and Pricing Policies amine two values of b which correspond to a
In the study of pricing decisions, we keep all param- purchase after T and T/10 time steps being valued
eters of the model fixed except for the values of b (the at 10% of utility for an immediate purchase.
strategicity parameter) and Y. It is convenient to present
the results in terms of the supply level s 5 Y/N. We 5.1.1 Optimal Expected Total Revenues as a
denote by pb;s ðÞ the pricing policy that is optimal for Function Supply Level. Figure 4 shows the optimal
fixed values of b and s, and by Rb;s ðÞ and Ub;s ðÞ the expected total revenues Rb;s ð0; 100Þ as a function of the
corresponding (optimal) future expected revenues and supply level s 5 Y/N (as well as the ratio of the
expected present values of customer utility in all states, expected total revenues R0b;s ð0; 100Þ=Rb;s ð0; 100Þ, the
respectively. The pricing policy which is optimal for ratio of the expected present value of consumer
0
myopic customers (under the assumption that b 5 0) is utilities Ub;s ð0; 100Þ=Ub;s ð0; 100Þ; and the optimal
p0;s ðÞ. If the company uses this policy in a situation price pb;s ð0; 100Þ). The plot of the expected revenues
when b40, the behavior of the customers will be (top left) contains four graphs which correspond to
different from the one expected by the company, and different values of b. The most prominent feature is
this policy will no longer be optimal. Given that the that, although in the case of myopic customers the
company applies the policy p0;s ðÞ, let the actual (exact) revenues are monotonically increasing in s, this is not
utility values for customers with strategicity parameter generally true in the presence of strategic customers.
0 As b increases, the graph of Rb;s ð0; 100Þ acquires a
b be Ub;s ðÞ. The eagerness to purchase in response to
p0;s ðÞ will be denoted by w0b;s, and is computed via maximum in s. This drop in revenue after a certain
0 point can be explained by a decrease in competition
Equation (2) where we use U ¼ Ub;s resulting in
between customers for the remaining inventory. We
t p ðt; nÞ þ u1 ðbU 0 ðt þ 1; nÞÞ : also see that the optimal revenues for all values of s
w0b;s ðt; nÞ ¼ F 0;s b;s
are decreasing in b. Moreover, the slope of the graphs
 Levin, McGill, and Nediak: Dynamic Pricing with Strategic Consumers
56 Production and Operations Management 19(1), pp. 40–60, r 2009 Production and Operations Management Society

0
Figure 4 The Optimal Expected Total Revenues Rb;s ð0 ; 100 Þ, the Ratio of the Expected Total Revenues Rb;s ð0 ; 100 Þ=Rb;s ð0 ; 100 Þ, the Ratio of the Expected
0
Present Value of Consumer Utilities Ub;s ð0 ; 100 Þ=Ub;s ð0 ; 100 Þ and the Optimal Price pb;s ð0 ; 100 Þ as Functions of the Supply Level s = Y/N for
Different Values of b and T = 1,000,000
600 1.05
Expected Revenues

500  = 1.0 1

Revenue Ratio
T = 0.1
400 T/10 = 0.1 0.95
300  = 0.0 0.9  =1.0
200 0.85 T = 0.1
100 0.8 T/10 = 0.1
0 0.75
0% 20% 40% 60% 80% 100% 0% 20% 40% 60% 80% 100%
Supply Level Supply Level

1 12
0.9 10
Utility Ratio

0.8
8

Price
0.7  = 1.0
 = 1.0 6 T = 0.1
0.6 T = 0.1 T/10 = 0.1
0.5 T/10 = 0.1 4  = 0.0
0.4 2
0% 20% 40% 60% 80% 100% 0% 20% 40% 60% 80% 100%
Supply Level Supply Level

(essentially, marginal revenues in Y) is decreasing. i.e., bT/10 5 0.1. This suggests that taking into account
Therefore, marginal analysis would suggest a lower the strategic nature of consumers is unnecessary if
initial stocking level when customers are strategic customers’ strategicity is such that they prefer to make
compared to the myopic case. The final observation their purchases within much shorter time spans than the
regarding these plots is a plateau-type behavior for planning period of the company. Thus, we see that even if
the highest supply levels. This behavior occurs the exact value of b is unknown, this model may allow
because the company will be less likely to sell its a company to assess whether or not strategic behavior
entire inventory when the customers are strategic if is important. If company losses from disregarding
the initial inventory level is too high. strategic behavior for a wide range of strategicity
parameter values are minor, then the company may
5.1.2 Relative Cost of Assuming Myopic Behavior. stick with a pricing policy which assumes myopic
Next, we consider the ratio of the expected total behavior. The validity of this conclusion may be
revenues R0b;s ð0; 100Þ=Rb;s ð0; 100Þ, which represents a affected if the capacity is not costless and the per-
relative cost to the company of its assumption that the formance is measured in terms of profit. We examine
customers are myopic in a situation where they are this issue further in Section 5.2.
not. The ratio is shown in the top right plot of Figure 4
as a function of s. A few observations are in order. 5.1.3 Relative Boost in Revenues Resulting from
First, the loss in revenues due to ignoring the strategic Myopic Behavior Assumption for Supply Levels near
nature of the customers can be very significant. For b 5 1 100%. An additional observation about Figure 4
the drop is as high as 20% for the supply level near concerns a somewhat surprising boost in revenues
87%. Other numerical trials showed that for a time resulting from the assumption that the customers are
horizon of T 5 100,000 the maximum loss is smaller myopic when they are, in fact, strategic for supply
and is approximately 13%, for even shorter time hori- levels close to 100% (also observed for shorter time
zons the maximum loss decreases further. Therefore, horizons). This type of phenomenon is not unusual in
taking into account the strategic nature of consumers game theory in general. It occurs in this case, since the
may not be as important if customers only have a few pair of company-customer strategies under consider-
shopping opportunities. Another useful observation ation is no longer an equilibrium (indeed, the com-
from the plots is that, for the supply levels under pany assumes in its pricing policy that the customers
about 30%, the loss in revenues does not exceed 1%. are myopic, while the actual strategic customers are
We also remark that the loss in revenues becomes very aware of that and adjust their behavior accordingly).
small (well under 1%) for all supply levels if the In the well-known prisoner’s dilemma, for example,
customer utility of a purchase after 1/10 of the both players gain if they simultaneously deviate from
planning period is 10% of an immediate purchase, the equilibrium. We can confirm that the prisoner’s
Levin, McGill, and Nediak: Dynamic Pricing with Strategic Consumers
Production and Operations Management 19(1), pp. 40–60, r 2009 Production and Operations Management Society 57

dilemma-type situation occurs here by looking at the announced, fully rational consumers would be able to
0
plots of utility ratios Ub;s ð0; 100Þ=Ub;s ð0; 100Þ pre- reconstruct it from the available information. There-
sented in the bottom left plot of Figure 4. We see fore, we can assume that the deterministic policy is
that the customers gain (at any supply level) if the known to the consumers, and substitute it into the
company uses the correct value of b instead of as- utility recursion (a similar computation was done in
suming myopic behavior (b 5 0) since the ratio never Equation (26) for the pricing policy which assumes
exceeds 1. To conclude, we remark that the company that consumers are myopic). Since the expected con-
can avoid this non-Pareto-optimal equilibrium, if the sumer utility for each time, state, and each pre-
supply level is a decision variable, and it is known announced policy is available, the company can use
that the customers are strategic. Indeed, there is no this information to optimize pre-announced prices.
gain in revenue from bringing the supply level close We have implemented such calculation for the case of
to 100%. The company will always work with smaller two prices, the time horizon of T 5 10,000, and a price
values of s as suggested by the plot of the expected switch at time t 5 5000. The best prices were selected
revenues. using a search over the price grid with step 0.02. The
resulting two-price policies for the cases of myopic
5.1.4 Optimal Price as a Function of Supply Level and fully strategic consumers are shown in the top
for Different b’s. We also examine the optimal price two plots of Figure 5. Similarly to the fully dynamic
pb;s ð0; 100Þ as a function of s in the bottom right plot of case, we observe that the first period prices in a
Figure 4. The optimal price in case of strategic con- market with strategic consumers are lower. However,
sumers is decreasing for smaller-to-medium values of the company prefers to keep prices higher in the
s but may exhibit an upward jump at a higher value of second period when supply levels are high. This can
s. The optimal price for higher values of b is smaller. be explained by the desire of the company to
Incidentally, this explains why the consumer utility discourage waiting on the part of strategic consum-
increases if the company knows that the customers are ers. We also observe very little price flexibility for
strategic – because the price goes down. supply levels above 40%. The ratio of expected reve-
nues of the pre-announced policies over fully dy-
5.1.5 Comparison of Fully Dynamic Contingent and namic ones is shown in the bottom left plot of Figure
Pre-announced Pricing Policies. Some businesses 5. When consumers are myopic, we see that pre-
operate under deterministic pricing policies rather announced policies always perform worse than fully
than contingent ones, which are the focus of this paper. dynamic ones (up to 4% for low supply levels and
In practice, such policies are often pre-announced, as in more than 1% for high supply levels). For the case of
the case of retail chain Filene’s Basement with its strategic consumers, we also see that pre-announced
automatic 25–50–75% price reduction scheme. How- pricing policies underperform when supply levels
ever, even if a deterministic policy were not pre- are low. However, if supply levels are higher than

Figure 5 Prices in Preannounced Two-Price Policy and the Ratios of Expected Revenues of the Preannounced Two-Price Policies Over the Fully Dynamic
Ones for the Myopic and Fully Strategic Cases as Functions of the Supply Level s = Y/N for T = 10,000
Consumers – myopic Consumers – strategic
7 7
6 Period 6 Period
5 First 5 First
Price

Price

4 Second 4 Second
3 3
2 2
1 1
0% 20% 40% 60% 80% 100% 0% 20% 40% 60% 80% 100%
Supply Level Supply Level

1.1
1.08
Revenue Ratio

1.06  = 1.0
1.04  = 0.0
1.02
1
0.98
0.96
0% 20% 40% 60% 80% 100%
Supply Level
 Levin, McGill, and Nediak: Dynamic Pricing with Strategic Consumers
58 Production and Operations Management 19(1), pp. 40–60, r 2009 Production and Operations Management Society

approximately 50%, the pre-announced policy outper- impact of consumer strategicity can be quite substantial,
forms the fully dynamic one (up to 9% for supply exceeding 25% of profits if customer behavior changes
levels above 70%). This finding is in agreement with from myopic to fully strategic, even if the company can
those of Aviv and Pazgal (2008), who also observed correctly take it into account. On the other hand, when
that announced pricing policies can be advantageous customers are only moderately strategic (bT/10 5 0.1),
to the seller compared with contingent pricing schemes then the impact is o5% of profits in the entire range
(although this conclusion was obtained under differ- of c. The strategicity impact is stronger for the lower
ent assumptions). A study of pre-announced pricing values of unit cost c, but becomes smaller as c
strategies in a model setting similar to ours is an approaches 10. As suggested in the previous sub-
interesting topic for future research. section, the correct way for the company to respond is to
decrease the initial capacity, thus limiting strategic
5.2. Implications of Strategic Consumer Behavior on consumer behavior. We can see this from the top
Optimal Initial Capacity Decisions right plot of Figure 6, which shows the optimal initial
In this subsection, we assume that the initial market capacity for different values of strategicity parameter
size N is fixed but the company can select the amount b as functions of cA[0, 10]. For lower values of c the
of initial capacity Y at the variable cost c per unit. The optimal initial capacity in the strategic case is up to
optimal Y is then selected to maximize the expected approximately 30% lower than in the myopic case.
present value of profit at time 0: However, when c is high and the optimal Y is small to
  start with, the difference between myopic and
max Rb;Y=N ð0; NÞ  cY : ð27Þ
Y¼0;...;N strategic case also becomes small since customers
have only limited opportunities for strategic behavior.
Different values of c will result in different optimal
solutions, which we denote as Yb(c). The correspond- 5.2.2 Possible Loss of Profitability if Strategicity is
ing optimal profits are denoted as pb(c). All results Ignored. If the company ignores strategic consumer
discussed below use the same parameter settings as in behavior both in pricing and capacity decisions, the
the previous subsection. impact on profits can be disastrous. The bottom left
plot of Figure 6 shows the actual expected profits
5.2.1 Effects of Strategicity on Profits and Optimal R0b;Y0 ðcÞ=N ð0; NÞ  cY0 ðcÞ resulting from using the
Procurement Decision When it is Correctly Taken pricing policy p0;s ðÞ and the initial capacity decision
into Account. The top left plot of Figure 6 shows the Y0(  ) (both optimal for the myopic case) when the
ratios of the optimal expected profit pb(c) for different customers are in fact strategic as functions of c for
values of strategicity parameter b to the optimal different values of b. Some of the worst situations – near
expected profit p0(c) in the myopic case as functions of zero or negative profit – are when the unit cost is in the
cA[0, 10]. We examine the same values of b as in the range [3,5] and near 10. One may be interested in how
previous subsection. The main conclusion is that the much of this dramatic drop in profit is due to pricing.

Figure 6 The Ratio of Optimal Expected Profits pb(c)/p0(c), the Optimal Initial Capacity yb(c), the Actual Expected Profit Resulting from Using the Myopic
Case Policies when Customers are Strategic, and the Ratio of the Actual and Optimal Expected Profits p0b ðcÞ=pb ðcÞ as Functions of Unit Cost c for
Different Values of b and T = 1,000,000
1 100
Optimal/myopic

0.95
Optimal initial

80
profit ratio

capacity

0.9
60  = 1.0
0.85
 = 1.0 40 T = 0.1
0.8
T = 0.1 T/10 = 0.1
0.75 20  = 0.0
T/10 = 0.1
0.7 0
0 2 4 6 8 10 0 2 4 6 8 10
Unit cost Unit cost
Actual expected profit

600 1
500  = 1.0 0.98
Actual/optimal

400 T = 0.1
profit ratio

0.96
300 T/10 = 0.1
200  = 0.0 0.94  = 1.0
100 0.92 T = 0.1
0 0.9 T /10 = 0.1
–100 0.88
0 2 4 6 8 10 0 2 4 6 8 10
Unit cost Unit cost
Levin, McGill, and Nediak: Dynamic Pricing with Strategic Consumers
Production and Operations Management 19(1), pp. 40–60, r 2009 Production and Operations Management Society 59

5.2.3 Contribution of Capacity vs. Pricing which it would be beneficial for the company to
Decisions in Proper Response to Strategicity. We reveal the information about n and/or y to the
also examine the actual expected profit for the case customers; and
when the capacity is selected optimally under the (2) a dynamic game model of interactions between
realization that the pricing policy will result in the company and strategic consumers in the
revenues R0b;Y=N ð0; NÞ rather than Rb;Y=N ð0; NÞ: presence of consumer and company learning.
n o
p0b ðcÞ ¼ max R0b;Y=N ð0; NÞ  cY : Acknowledgment
Y¼0;...;N
This research was supported by Natural Sciences and
Engineering Research Council of Canada (grant numbers
In fact, the company would be unable to realize
261512-04 and 341412-07). The authors thank the Depart-
that the expected revenue is R0b;Y=N ð0; NÞ under its ment and Senior Editors, and the Referees for their
incorrect myopic model of demand, but the ratio p0b constructive suggestions, which helped to better explain
ðcÞ=pb ðcÞ can be used to gauge how much of the op- the model and improve the results. The third author sends
timal profit must be attributed to using the correct prayers of utmost gratitude to God in Whom he sought
pricing policy. This ratio is shown in the bottom right inspiration during this work.
plot of Figure 6 as a function of c for different values
of b. When the unit cost is low, the pricing policy
correctly accounting for the strategic behavior con- References
tributes up to around 10% of the optimal profit in the Anderson, S., A. de Palma, J. Thisse. 1992. Discrete Choice Theory of
Product Differentiation. MIT Press, Cambridge, MA.
fully strategic case. When the unit cost is high, a
Aviv, Y., A. Pazgal. 2008. Optimal pricing of seasonal products in the
correct capacity decision appears to be more impor- presence of forward-looking customers. Manuf. Service Oper.
tant since the profit ratio attained by it is close to 1. Manage. 10(3): 339–359.
This can be explained by generally low levels of the Besanko, D., W. Winston. 1990. Optimal price skimming by a mo-
optimal capacity when the unit cost is high. nopolist facing rational consumers. Manage. Sci. 36(5): 555–567.
Overall, the results of this subsection suggest that Cachon, G., R. Swinney. 2007. Purchasing, pricing, and quick re-
taking into account strategic behavior of consumers sponse in the presence of strategic consumers. Working Paper,
University of Pennsylvania.
is very important in capacity decisions. Moreover, a
Caldentey, R., G. Vulcano. 2007. Online auction and list price rev-
proper capacity choice can partially compensate for enue management. Manage. Sci. 53(5): 795–813.
consumer strategicity. Elmaghraby, W., A. Gülcü, P. Keskinocak. 2008. Designing optimal
pre-announced markdowns in the presence of rational custom-
6. Conclusions and Directions for ers with multi-unit demands. Manuf. Service Oper. Manage.
10(1): 126–148.
Future Research Feng, Y., B. Xiao. 2000. A continuous time yield management model
The paper presents a new, inherently stochastic, with multiple prices and reversible price changes. Manage. Sci.
game-theoretical dynamic pricing model for a finite 46(5): 644–657.
population of strategic customers. This model explic- Gallego, G., R. Phillips, Ö. Rahin. 2008. Strategic management of
itly incorporates the strategic nature of consumers in distressed inventory. Prod. Oper. Manage. 17(4): 402–415.
the decision-making process. The model falls within Gallego, G., G. van Ryzin. 1994. Optimal dynamic pricing of inven-
tories with stochastic demand over finite horizons. Manage. Sci.
the general class of stochastic dynamic games but 40(8): 999–1020.
generalizes stochastic consumer behavior in a man- Levin, Y., J. McGill, M. Nediak. 2009. Dynamic pricing in the pres-
ner that, to our knowledge, has not previously been ence of strategic consumers and oligopolistic competition.
achieved in the literature. We demonstrate the Manage. Sci. 55(1): 32–46.
existence of a unique subgame-perfect equilibrium Levina, T., Y. Levin, J. McGill, M. Nediak. 2009. Dynamic pricing
with online learning and strategic consumers: An application of
pricing policy under reasonable assumptions, provide the aggregating algorithm. Oper. Res. (in press).
optimality conditions, and explore some of the struc- Liu, Q., G. van Ryzin. 2008. Strategic capacity rationing to induce
tural properties of solutions. We also discuss impli- early perchases. Manage. Sci. 54(6): 1115–1131.
cations of strategic consumer behavior for overall McKelvey, R., T. Palfrey. 1995. Quantal response equilibria for nor-
performance as well as pricing and procurement mal form games. Games Econom. Behav. 10(1): 6–38.
decisions. The proposed computational procedure McKelvey, R., T. Palfrey. 1998. Quantal response equilibria for
allows evaluation of this model on problems of real- extensive form games. Exp. Econom. 1(1): 9–41.
istic size. Ovchinnikov, A., J. Milner. 2007. Revenue management with end-of-
period discounts in the presence of customer learning. Working
The future research related to this model may in- Paper, University of Toronto.
clude the following topics: Phillips, R. 2005. Pricing and Revenue Optimization. Stanford Univer-
sity Press, Stanford, CA.
(1) considering situations when the parameters n Shen, Z., X. Su. 2007. Customer behavior modeling in revenue
and/or y are not known to the company or cus- management and auctions: A review and new research oppor-
tomers exactly, and describing the situations in tunities. Prod. Oper. Manage. 16(6): 713–728.
 Levin, McGill, and Nediak: Dynamic Pricing with Strategic Consumers
60 Production and Operations Management 19(1), pp. 40–60, r 2009 Production and Operations Management Society

Su, X. 2006. A model of consumer inertia with applications to Zhao, W., Y. Zheng. 2000. Optimal dynamic pricing for perishable
dynamic pricing. Working Paper, University of California, assets with nonhomogeneous demand. Manage. Sci. 46(3):
Berkeley. 375–388.
Su, X. 2007. Inter-temporal pricing with strategic customer behavior.
Manage. Sci. 53(5): 726–741.
Talluri, K., G. van Ryzin. 2004. The Theory and Practice of Revenue
Management. Kluwer Academic Publishers, Norwell, MA. Supporting Information
Tversky, A. 1972. Elimination by aspects: A theory of choice. Psychol. Additional supporting information may be found in the
Rev. 79(4): 281–299.
online version of this article:
Xie, J., S. Shugan. 2001. Electronic tickets, smart cards, and online
prepayments: When and how to advance sell. Market. Sci. 20(3):
219–243.
Appendix S1. Assumptions, Notation, and Proofs
Xu, X., W. Hopp. 2004. Customer heterogeneity and strategic
behavior in revenue management: A martingale approach. Please note: Wiley-Blackwell is not responsible for the
Working Paper, Nothwestern Univeristy. content or functionality of any supporting materials sup-
Zhang, D., W. Cooper. 2008. Managing clearance sales in the plied by the authors. Any queries (other than missing
presence of strategic customers. Prod. Oper. Manage. 17(4): material) should be directed to the corresponding author for
416–431. the article.