Optimal Iterative Pricing over Social Networks

Hessameddin Akhlaghpour ∗ Mohammad Ghodsi ∗ Nima Haghpanah ∗

Hamid Mahini ∗ Vahab S. Mirrokni † Afshin Nikzad ∗

ABSTRACT ing through word-of-mouth. This can be done by understanding

In this paper, we study the optimal pricing for revenue maximiza- the externalities among buyers in a social network. The increasing
tion in the presence of positive network externalities. In our model, popularity of these networks has allowed companies to collect and
the value of a digital good for a buyer is a function of the set of buy- use information about inter-relationships among users of social net-
ers who have already bought the item. In this setting, a decision to works. In particular, by designing certain experiments, these com-
buy an item depends on its price and also on the set of other buy- panies can determine how users influence each others’ activities.
ers that have already owned that item. The revenue maximization Consider an item or a service for which one buyer’s valuation is
problem in the context of social networks has been studied by Hart- influenced by other buyers. In many settings, such influence among
line, Mirrokni, and Sundararajan [11], following the previous line users are positive. That is, the purchase value of a buyer for a ser-
of research on optimal viral marketing over social networks [13, vice increases as more people use this service. In this case, we say
14, 16, 19]. that buyers have positive externalities on each other. Such phenom-
In contrast to the previous work by Hartline et. al. [11], we con- ena arise in various settings. For example, the value of a cell-phone
sider this problem without price discrimination. We consider the service that offers extra discounts for calls among people using
Bayesian setting in which there are some prior knowledge of the the same service, increases as more friends buy the same service.
probability distribution on the valuations of buyers. In particular, Such positive externality also appears for any high-quality service
we study two iterative pricing models in which a seller iteratively through positive reviews or the word-of-mouth advertising. In this
posts a new price for a digital good (visible to all buyers), and any paper, we explore optimal pricing strategies for revenue maximiza-
interested buyer can buy the item at the posted price. In one model, tion in the presence of positive network externalities.
re-pricing of the items are only allowed at a limited rate. For this By taking into account the positive externalities, sellers can em-
case, we give a FPTAS for the optimal pricing strategy in the gen- ploy forward-looking pricing strategies that maximize their long-
eral case. In the second model, we allow very frequent re-pricing term expected revenue. For this purpose, there is a clear trade-
of the items. We show that the revenue maximization problem in off between the revenue extracted from a buyer at the beginning,
this case is inapproximable even for simple deterministic valuation and the revenue from future sales. That is, the lower the price of-
functions. In the light of this hardness result, we present constant fered to a buyer, the lower the extracted revenue, but the higher
and logarithmic approximation algorithms for a special case of this the probability of the sale as well as the expected influence on fu-
problem when the individual distributions are identical. ture buyers. For example, the seller can give large discounts at the
beginning to convince buyers to adopt the service. These buyers
will, in turn, influence other buyers and the seller can extract more
1. INTRODUCTION revenue from the rest of the population, later on. Other than being
Despite the rapid growth, online social networks have not yet explored in research papers [11], this idea has been employed in
generated significant revenue. Most efforts to design a comprehen- various marketing strategies in practice, e.g., in selling TiVo digital
sive business model for monetizing such social networks [21, 22], video recorders [30].
are based on contextual display advertising [31]. An alternative In an earlier work, Hartline, Mirrokni, and Sundararajan [11]
way to monetize social networks is viral marketing, or advertis- study the optimal marketing strategies in the presence of such posi-
∗ tive externalities. They study optimal adaptive ordering and pricing
Department of Computer Engineering, Sharif University of Tech- by which the seller can maximize its expected revenue. However,
nology, {akhlaghpour,haghpanah,mahini,nikzad}@ce.sharif.edu,
ghodsi@sharif.edu in their study, they consider the marketing settings in which the
†
Google Research NYC, 76 9th Ave, New York, NY 10011, mir- seller can go to buyers one by one (or in groups) and offer a price
rokni@google.com to those specific buyers. Allowing such price discrimination makes
the implementation of such strategies hard. Moreover, price dis-
crimination, although useful for revenue maximization in some set-
tings, may result in a negative reaction from buyers. For example,
Oliver and Shor [20] suggest that price discrimination has negative
Permission to make digital or hard copies of all or part of this work for effect on the likelihood of purchase. They do this by studying web-
personal or classroom use is granted without fee provided that copies are sites that allow users to enter promotion-codes and get discounts
not made or distributed for profit or commercial advantage and that copies on the price of the product. They claim that the existence of such
bear this notice and the full citation on the first page. To copy otherwise, to
republish, to post on servers or to redistribute to lists, requires prior specific
codes suggests price promotions that may not be available to the
permission and/or a fee. customer, which results in the reduction of the user’s trust, and thus
Copyright 200X ACM X-XXXXX-XX-X/XX/XX ...$5.00.
in the likelihood of purchase. They assume that the seller is capa- allow the influence amongst buyers to propagate in the same time
ble of providing this discounts directly to the targeted users. Shor step. In this setting, as we change the price per time step, we as-
and Oliver [23] question this assumption by showing that with the sume the influence among buyers will be effective on the next time
existence of coupon repositories on the web, the users will be seg- step (and not on the same time step). In the following, we define
mented according to their technical competence, instead of their these two problems formally.
price sensitivity. This casts further doubts on the applicability of
price discrimination as a method to increase the revenue of the D EFINITION 1.1. The Basic(k) Problem: In the Basic(k) prob-
seller and supports our study of pricing strategies without price dis- lem, our goal is to find a sequence p1 , . . . , pk of k prices in k con-
crimination. Finally, the problem of disallowing price discrimina- secutive time steps. A buyer decides to buy the item during a time
tion in designing marketing strategies was also raised as an open step as soon as her valuation is more than or equal to the price
research direction by Hartline et. al. [11]. offered in that time step. In contrast to the Rapid(k) problem, the
In this paper, we explore algorithmic problems for optimal pric- buyer’s decision in a time step immediately affects the valuations
ing without price discrimination in the presence of network exter- of other buyers in the same time step. More precisely, a time step
nalities. In particular, we assume that a seller can iteratively post is assumed to end when no more buyers are willing to buy the item
a price for an item at several time steps. Although the price can at the price at this time step. At this time, we move to the next step
change in different time steps, it is visible to all the buyers at all and offer a new price. Our goal is to find a sequence p1 , . . . , pk of
times, and a buyer may buy the item at any time step. We assume a k prices that maximizes our expected revenue given the probability
Bayesian setting in which we have a prior (or a probability distribu- distributions fi,S for each buyer i and each subset S ⊂ V .
tion) over the valuation of buyers. One can define various models
for the iterative pricing based on the number of time steps at which Note that in the Basic(k) problem, the price sequence will be
the seller can change the price, and on restrictions on the price se- decreasing. If the price posted at any time step is greater than the
quence we can offer. We study two iterative pricing models that previous price, no buyer would purchase the product at that time
allow different rates of re-pricing the item, and will study the com- step.
plexity of the revenue maximization problems in these two settings.
D EFINITION 1.2. The Rapid(k) Problem: Consider a seller
We also discuss incentive issues at the end of the paper. We elabo-
who wants to sell a digital good, and a set V of n buyers each with a
rate on our results after defining our models formally.
valuation vi (S) for each subset S ⊂ V \{i}. We assume that vi (S)
Preliminaries. Consider a case of selling multiple copies of a dig- is a random variable that comes from a probability distribution with
ital good (with no cost for producing a copy) to a set V of n buy- an accumulative distribution function Fi,S . Given a number k, the
ers. In the presence of network externality, the valuation of buyer Rapid(k) problem is to design a pricing policy for k consecutive
i for the good is a function of buyers who already own that item, days or time steps. In this problem, a pricing policy is to set a
vi : 2V → R, i.e., vi (S) is the value of the digital good for buyer public price pi at the start of time step (or day) i for each 1 ≤
i, if set S of buyers already own that item. We say that users have i ≤ k. At the start of each time step, after the public price pi
positive externality on each other, if and only if vi (S) ≤ vi (T ) for is announced, each buyer decides whether to buy the item or not,
each two subsets S ⊆ T ⊆ V . In general, we assume that the seller based on the price offered on that time step 1 and her valuation. In
is not aware of the exact value of the valuation functions, but she the Rapid(k) problem, the decision of a buyer during a time step is
knows the distribution fi,S with an accumulative distribution Fi,S not affected by the action of other buyers in the same time step. Our
for each random variable vi (S), for all S ∈ V and any buyer i. goal to find a pricing policy consisting of k prices that maximizes
Also, we assume that each buyer is interested only in a single copy the expected revenue to the seller 2 .
of the item. The seller is allowed to post different prices at different
time steps and buyer i buys the item in a step t if vi (St ) − pt ≥ 0, For more insight on the Rapid(k) problem, we will study an ex-
where St is the set of buyers who own the item in step t, and pt is ample (Figure 1) later. One drawback of the Rapid(k) problem is
the price of the item in that step. Note that vi (∅) does not need to that the re-pricing may be done very frequently, and after only one
be zero; in fact vi (∅) is the value of the item for a user before any level of influence propagation. Similarly, we can say that buyers
other buyer owns the item and influence him. react slowly to the new price and the seller can change the price
We study optimal iterative pricing strategies without price dis- before the news spreads through the network. On the other ex-
crimination during k time steps. In particular, we assume an itera- treme, we can consider a model in which buyers immediately be-
tive posted price setting in which we post a public price pi at each come aware of the new state of the network and react before the
step i for 1 ≤ i ≤ k. The price pi at each step i is visible to all seller is capable of changing the price. This idea lead to the other
buyers, and each buyer might decide to buy the item based on her iterative pricing model studied in this paper. In this model, after
valuation for the item and the price of the item in that time step. setting the price at one time step, the influence among buyers prop-
An important modeling decision to be made in a pricing problem is agate until no other buyer has incentive to buy the item. The formal
to whether model buyers as forward-looking (strategic) or myopic definition follows:
(impatient). For the most of this paper, we consider myopic or im- Example. For more insight about the above two models and their
patient buyers who buy an item at the first time in which the offered differences, we study the following example: Consider n buyers
price is less than their valuations. We discuss this issue along with numbered 1 to n. For buyer i (1 ≤ i ≤ n), the initial bid is
forward-looking buyers in more details in Section 4 after stating ǫ · (i − 1). Any purchase by any buyer i 6= 1 increases buyer 1’s
our results for myopic buyers. valuation by L. The valuations of the rest of the buyers does not
In order to formally define the problem, we should also define change (See Figure 1).
each time step. A time step can be long enough in which the influ- 1
In discussions for the Rapid(k) problem, we use the terms time
ence among users can propagate completely, and we can not modify step and day interchangeably.
the price when there is a buyer who is interested to buy the item at 2
Note that in the Rapid(k) problem, a pricing policy is adaptive in
the current price. On the other extreme, we can consider settings that the price pi at time step i may depend on the actions of buyers
in which the price of the item changes fast enough that we do not in the previous time steps.
 ǫ the function h(p) = f (p)/(1−F (p)) is monotone non-decreasing.
2ǫ
0 Finally, we consider the non-decreasing version of the Rapid(k)
3ǫ problem in which the sequence of prices has to be non-decreasing,
i.e, p1 ≤ p2 ≤ . . . pk . By doing so, we make sure that the best

...
strategy of self-interested buyers is to buy the item in the first day
(n − 1)ǫ in which the price is less than their valuations.
Our Contributions. We explore the complexity of the Rapid(k)
and Basic(k) problems, and present approximation algorithms and
Figure 1: Each node represents a buyer. The numbers writ-
hardness results for these problems.
ten next to each buyer is her initial bids. The arrows represent
We first show that the deterministic Basic(k) problem is polynomial-
influences of L. L (and ǫ) are arbitrarily large (and small) num-
time solvable. Moreover, for the Bayesian Basic(k) problem, we
bers respectively.
present a fully polynomial-time approximation scheme. We discuss
incentive issues for strategic agents at the end of the paper.
Next we show that in contrast to the Basic(k) problem, the Rapid(k)
problem is intractable. For the Rapid(k) problem, we show a strong
Consider the Rapid(n) problem on this example. One may think hardness result: we show that the Rapid(k) problem is not approx-
the Rapid(n) problem can be easily solved by posting the highest imable within any reasonable approximation factor even in the de-
price that anyone is willing to buy the product for it. Using this terministic case (in which the valuation functions of buyers are ex-
naive approach for this example, in the second time step, the seller actly known) unless P=NP. This hardness result holds even if the in-
would sell the product for price L to buyer 1. But if she decides to fluence functions are submodular and the probability distributions
post a public price ǫ on the first time step, she can sell the product satisfy the monotone hazard rate condition. In light of this hardness
for price (n−1)L to player 1 afterwards. For the Basic(n) problem, result, we give an approximation algorithm using a minor and nat-
no matter what the seller does, she can not sell the product to buyer ural assumption. We show that the Rapid(k) problem for buyers
1 for more than 1 + (n − 1)ǫ. This example shows that the seller with submodular influence functions and probability distributions
can get much more revenue in the Rapid(n) problem compared to with the monotone hazard rate condition, and identical initial dis-
the Basic(n) problem. tributions admits logarithmic approximation if k is a constant and
Since Rapid(k) is hard to approximate, we study a special vari- 1
a constant-factor approximation if k ≥ n c for any constant c.
ant of it, the submodular valuation with identical initial distribu- Related work. Optimal pricing mechanisms in the presence of
tions variant. Here, we give the required definitions for this prob- network externalities have been considered in the economics liter-
lem. ature [3, 5, 8, 12, 18, 25]. Carbal, Salant, and Woroch [5] con-
A common assumption studied in the context of network exter- sider an optimal pricing problem with network externalities when
nalities is the assumption of submodular influence functions. This the buyers are strategic, and study the properties of equilibrium
assumption has been explored and justified by several previous work prices. In their model, buyers tend to buy the product as soon as
in this framework [7, 11, 13, 14, 19]. In the context of revenue possible because of a discount factor, which reduces the desirabil-
maximization over social networks, Hartline et. al. [11] state this ity of late purchases. Previous work in economic literature such as
assumption as follows: suppose that at some time step, S is the [10, 6, 4, 9, 28] had shown that without network externalities, the
set of buyers who have bought the item. We use the notion of equilibrium prices decrease over time. On the other hand, Carbal
optimal (myopic) revenue of a buyer for S, which is Ri (S) = et.al. show that in a social network the seller might decide to start
maxp p · (1 − Fi,S (p)). Following Hartline et.al [11], we consider with low introductory prices to attract a critical mass of players,
the optimal revenue function as the influence function, and assume when the players are large (i.e, the network effect is significant).
that the optimal revenue functions (or influence functions) are sub- They observe that this pattern (of increasing prices) also happens
modular, which means that Ri (S)+Ri (S ′ ) ≥ Ri (S∪S ′ )+Ri (S∩ when there is uncertainty about customer valuations, no matter how
S ′ ). An equivalent definition of submodularity says that function strong the network effect is.
Ri is submodular if and only if for any two subsets S ⊂ T , and any Bensaid and Lesne [3] studied the problem when a monopolist
element j 6∈ S, Ri (S ∪ {j}) − Ri (S) ≥ Ri (T ∪ {j}) − Ri (T ). sells a durable good. They discuss two types of network external-
In other words, submodularity corresponds to a diminishing return ities: word of mouth externality and learning by doing externality.
property of the optimal revenue function which has been observed Similar to our model, the authors assume the value of the good de-
in the social network context [7, 13, 14, 19]. Next, we define the pends on number of earlier users. This means that players are ex-
identical initial distribution assumption. cluded from additional externalities after the purchase of the good.
They use the example of software products to justify their model.
D EFINITION 1.3. We say that all buyers have identical initial
The quality of software increases after more buyers purchase the
distributions if there exists a distribution F0 so that the valuation of
product and discover and report the bugs in it. This means that
a player given that the influence set is equal to S is the sum of two
after the purchase, a player will no longer benefit from the exter-
independent random variables, one from F0 , and another one from
nalities produced by other players, unless she pays for the updated
Fi,S , with Fi,∅ = 0. Note that we allow any kind of dependency
software. This assumption also appears in our model. Also, exter-
between the distributions of the form Fi,S .
nalities has been modeled as a linear function of earlier buyers in
Next, we define probability distributions satisfying the monotone [3] which is special case of our valuation function. On the other
hazard rate condition. Several natural distributions like uniform hand, it has been shown that under negative network externalities,
distributions and exponential distributions satisfy this condition. the seller sets higher prices in order to increase the value of the good
(Kessing and Nuscheler [15]). In addition, in a recent work, Saask-
D EFINITION 1.4. A probability distribution f with accumula- ilahti [24] studies the effect of network topology on the monopoly
tive distribution F satisfies the monotone hazard rate condition if pricing of network goods, when price discrimination is not allowed.
 buyers
They assume that the network is not necessarily symmetric; e.g,
n
some players have a few deep relations, while others have a large
p0
number of shallow friendships. In such networks, they identify a
set of players, they call critical players, who have a more important
role in the network. They show that the topological effect, which p1
is caused by the existence of critical players, is a dominating ef-
fect on the optimal price. They observe that when critical players
are present, the seller tends to set lower prices in order to increase
the probability of such players buying the item. Finally, Sundarara-
jan [26, 27] study the pricing in presence of network externalities pk
2
in a game theoretic setting with incomplete information (in both
monopolistic and competitive environments). 1
Optimal viral marketing over social networks have been studied v0 v1 vn−1 vn price
v2
extensively in the computer science literature [16]. For example,
Kempe, Kleinberg and Tardos [13] study the following algorithmic
question (posed by Domingos and Richardson [7]): How can we Figure 2: The valuation of players are assumed to be constant,
identify a set of k influential nodes in a social network to influence exactly known, and sorted (v0 is the least valuation). The graph
such that after convincing this set to use this service, the subse- shows the number of players who will buy the item given any
quent adoption of the service is maximized? Most of these models price. Both Basic(k) and Rapid(k) problems then reduce to
are inspired by the dynamics of adoption of ideas or technologies maximizing the area under the graph with k rectangles.
in social networks and only explore influence maximization in the
spread of a free good or service over a social network [7, 13, 14,
19]. As a result, they do not consider the effect of pricing in adopt-
ing such services. On the other hand, the pricing (as studied in this one of vi ’s. As a result, both of the problems reduce to finding a
paper) could be an important factor on the probability of adopting set of prices to maximize the area under k rectangles, fitted under
a service, and as a result in the optimal strategies for revenue max- the curve of Figure 2. It can be done easily by a dynamic program-
imization. ming algorithm. We will later observe that this is closely connected
Aviv and Pazgal [1] consider Inventory contingent strategies (adap- to the Basic(k) problem with externalities.
tive) and announced fixed-discount strategies (non-adaptive) (not in
social networks), when the players are strategic. It is obvious that 2.1 Deterministic Basic(k)
the seller can do better using adaptive strategies, since she can re- As defined earlier, in the Basic(k) problem, any buyer’s decision
act to the events she observes. But [1] observes that non-adaptive to purchase an item affects the valuations of other buyers during
prices are as good as adaptive strategies when the players are my- the same time step. The time step ends when no more buyers are
opic. They argue that the reason is that a credible precommitment willing to buy the item. We first show that the order that buyers
to a fixed price removes the (rational) expectation of players to face decide to buy the item, has no effect on the state after the time step
a large discount in the future. They also observe that in both vari- has ended.
ants, increasing prices are better when players are relatively more We define B 1 (S, p) := {i|vi (S) ≥ p} ∪ S. Assume a time
strategic, and decreasing prices are better if players are relatively step where at the beginning, we set the global price p, and the set
myopic. Following this line of thought, we note that the Rapid(k) S of players already own the item. B 1 (S, p) specifies the set of
problem as defined in this paper is adaptive, but our hardness result buyers who immediately want to buy (or already own) the item.
also holds for the non-adaptive variant of the problem. On the other As B 1 (S, p) will own the item before the time step ends, we can
hand, the Basic(k) problem is defined in the non-adaptive model. recursively define B k (S, p) = B 1 (B k−1 (S, p), p) and use induc-
We elaborate on some related work about pricing for impatient tion to reason that B k (S, p) will own the item in this time step. Let
buyers versus pricing for forward-looking buyers in Section 4.
B(S, p) = B k̂ (S, p), where k̂ = max{k|B k (S, p)−B k−1 (S, p) 6=
∅}, knowing that all buyers in B(S, p) will own the item before the
2. THE Basic(k) PROBLEM time step ends.
In this section, we first study the case where the valuation func- Since the valuation functions vi for each buyer i are monotone
tions of buyers are deterministic, and for this problem, we show non-decreasing, one can easily argue that the set B(S, p) identifies
that in contrast to the Rapid(k) problem which is hard to approx- the set of buyers who own the item at the end of this time step, and
imate even in the deterministic case, the Basic(k) problem can be this set does not depend on the order of users who choose to buy
solved exactly using a dynamic programming approach. Later, we the item.
extend the dynamic programming approach and present an FPTAS First Step: Solving Deterministic Basic(1). First, we state the
for the Basic(k) problem. following lemma, which can be proved by induction.
Before we start presenting technical results, let us first study the
following warmup example. Both of the problems defined above L EMMA 2.1. For any a and b such that a < b, B(∅, b) ⊆
reduce to simple optimization problems, if we ignore the external- B(∅, a).
ities in the network. Assume that the valuation of each player is
independent of the set of players currently owning the item, and In the Basic(1) problem, the goal is to find a price p1 such that
also this value is exactly known., i.e., for each player i, her val- p1 · |B(∅, p1 )| is maximized. Let βi := sup{p|i ∈ B(∅, p)} and
uation is vi . Since the valuations of players do not change, the β := {βi |1 ≤ i ≤ n}. WLOG we assume that β1 > β2 . . . > βn .
Basic(k) problem will be similar to Rapid(k), and we can assume Using Lemma 2.1, player i will buy the item if and only if the price
that the optimal price sequence is non-increasing . Also it can be is set to be less than or equal to αi . The definition of β tells us that
proved that all prices in the optimum solution should be equal to βi+1 is the maximum price p for which B(∅, βi ) ( B(∅, p).
 L EMMA 2.2. The optimal price p1 is in the set β. using a price p for a polynomial number of trials and and taking the
P ROOF. By increasing p to the smallest price in β which is average of the number of buyers who bought the item in each trial.
larger than p, we would achieve a better revenue without losing any Since 0 ≤ Xp ≤ |V |, using Chernoff-Hoeffding concentration in-
customers. For the extreme case where p > β1 , we can decrease p equality, we can easily show that E[Xp ] can be computed within
to β1 to achieve a better revenue. an error factor of ǫ with high probability.
Using the sampling method, we estimate E[Cp ] with high proba-
In light of the above discussion, we provide an algorithm to find bility for any given p. Assuming values pmin = 1 and a maximum
p1 by finding all elements of the set β and considering the profit price pmax such that pmin ≤ pOP T ≤ pmax , the idea is to check
βi · B(∅, βi ) of each of them, to find the best result. Throughout some specific price values between pmin and pmax and choose one
the algorithm, we will store a set S of buyers (who have bought of them with respect to their estimated revenue. The algorithm is
the item) and a global price g. In the beginning of the algorithm as follows:
S = ∅ and g = ∞. The algorithm consists of |β| steps. At the i-th
p /pmin
step, we set the price equal to the maximum valuation of remaining 1. Let imax = log1+ǫ
max
+1.
players, considering the influence set to be S. We then update the
state of the network until it stabilizes, and moves to the next step. 2. Define values pi = (1 + ǫ)i pmin for any integer i, 0 ≤ i <
Our main claim is as follows. At the end of the i-th step, the set imax , and compute E[Cpi ] for each pi .
who own the item is B(∅, βi ), and the maximum valuation of any 3. Return pi with the maximum calculated E[Cpi ].
remaining player is equal to βi+1 .
The Algorithm of the i-th Step: By induction, we know that T HEOREM 2.3. For every m ≥ 3 and t ≥ 1, there exists an
S = B(∅, βi−1 ), and the maximum valuation of any remaining polynomial time algorithm which finds a price p such that E[Cp ] ≥
player is βi < βi−1 . As a result of lemma 2.1, we know that 1−ǫ
E[CpOP T ] (1+ǫ) 2 with high probability.
S ⊆ B(∅, βi ). By the argument presented at the beginning of the
section, if we set g ← βi and wait until the network stabilizes, the Now, we are ready to design an algorithm for solving Basic(k)
final set of owners will be equal to S ′ = B(∅, βi ). Then, using problem. In this problem, we have k time steps and we need to set
the definition of β, the maximum valuation of any player would be a price pi in time step i. Let Xp1 ,p2 ,...,pt be the number of buyers
equal to βi+1 . who buy the item at time step t when the price is pi at time step
Generalization to Deterministic Basic(k). i ≤ t. Thus, the total revenue is
We attempt to solve the Basic(k) problem by executing the Basic(1) revenue = p1 Xp1 + p2 Xp1 ,p2 + ... + pk Xp1 ,p2 ,...,pk .
algorithm consisting of m steps and by using a dynamic algorithm.
We are looking for an optimal sequence (p1 , p2 , . . . , pk ), which In Basic(k), our goal is to maximize the expected revenue, which
is a decreasing sequence. Considering that, and is:
Pklemma 2.1, the
value that we are attempting to maximize is i=1 |B(∅, pi ) − E[revenue] = p1 E[Xp1 ] + p2 E[Xp1 ,p2 ] + ... + pk E[Xp1 ,p2 ,...,pk ]
B(∅, pi−1 )| · pi . (1)
We claim that an optimal sequence exists such that for every i, Using standard sampling technique and Chernoff-Hoeffding bound,
pi = βj for some 1 ≤ j ≤ |β|. This can be shown by a proof sim- we can estimate E[Xp1 ,p2 ,...,pt ] within a small error with high
ilar to that of lemma 2.2. Thus the problem Basic(k) can be solved probability (since 0 ≤ Xp1 ,p2 ,...,pt ≤ n).
by considering the subproblem A[k′ , m] where we must choose an Given any set of valuations, the set of all buyers who have bought
increasing sequence π of k′ prices from the set {β1 , β2 , . . . βm }, to the item after time step t, is the same as the set of buyers who have
maximize the profit, and setting the price at the last day to βm . This bought the item when we have only one time step in which we set
subproblem can be solved using the following dynamic program: the price to be pt . Thus, Xp1 + Xp1 ,p2 + ... + Xp1 ,p2 ,...,pt = Xpt
A[k′ , m] = max A[k′ − 1, t] + |B(∅, βm ) − B(∅, βt )| · βm which implies Xp1 ,p2 ,...,pt = Xpt −Xpt−1 . Thus, E[Xp1 ,p2 ,...,pt ] =
1≤t<m
E[Xpt ] − E[Xpt−1 ].
One can observe a fine connection between Basic(k) problem First, we present a simple dynamic program with a polynomial
and the warmup problem discussed at the beginning of this section. running time in pmax . Then we modify it and design an FPTAS
In both cases, we are trying to maximize the area under a non- for the problem. As a warm-up example, let’s assume that E[Xp ]
increasing step function using a number of rectangles. In the first can be computed precisely for every p, and we are allowed to offer
case, the drop in the function happens at vi s, and in the second case only integer prices. We design a dynamic programming algorithm
in βi s. In both cases, however, this drop happens when we lose a to solve the problem. Consider the state of the network when we
player because of incrementing the price. set the price to p and wait until the network becomes stable. In this
2.2 An FPTAS for Basic(k) state, some subset of buyers has bought the item. We call this state
of the network, Net-Stable(p). We define A[t, p] as the maximum
Throughout this section, we assume a minimum price pmin = 1
expected revenue when we have t time steps and the state of the net-
for the item, and design pricing algorithms that optimize the rev-
work is Net-Stable(p). In order to calculate A[t, p] we can search
enue using a minimum price of 1$. Let ROU N D be the running
for the price to be offered in the first time step. It could be any
time of simulating one trial of the buying process during one time
price below p. The recurrence relation for the dynamic program is
step (given probability distributions f (i, S) for each user i and each
as follows:
subset S of users). We first observe a simple FPTAS for Basic(1),
and then use it to solve Basic(k). In the Basic(1) problem, our A[t, p] = max
′
{A[t − 1, p′ ] + p′ (E[Xp′ ] − E[Xp ])} (2)
0<p <p
goal is to find a price p that maximizes the expected revenue of the
seller when she sets the price to p. Let Cp and Xp be random vari- Note that in state Net-Stable(pmax + 1), no buyer has bought the
ables for the revenue and number of buyers who buy the item when item and the network is in the initial state. Therefore our solution
we set the price to p. Our goal is to find a price p that maximizes is stored in A[k, pmax + 1]. The above algorithm is based on some
E[Cp ] = pE[Xp ]. Note that we can estimate E[Xp ] using a stan- unrealistic assumption and its running time is polynomial in terms
dard sampling method, i.e., by sampling from the random process of to pmax .
 1−ǫ
A 2(1+ǫ) -approximation Algorithm. Let imax = ⌈log(pmax )⌉. 1
First we observe that if k ≥ imax + 1, one can easily design a
1 − F (x)
1
2
-approximation algorithm: Offer price pi = 2imax −i at time step 1 − F (p)
S1
i. Assume that in the optimum solution, we offer price poi at time
step i. Consider the set of buyers who has bought the item in Net-
Stable(p) and call this set Sp . It is clear that Sp ⊆ Sp′ for p ≥ p′ .
S2
Let buyer x be a buyer who has bought the item at price 2r ≤
px < 2r+1 in the optimum solution. Thus buyer x will buy the a p b
item when the price is 2r in our solution. Since 2r ≥ px /2, the
above simple algorithm is a 21 -approximation algorithm for the case Figure 3: the graph of 1 − F (x). The expectation of f is the
k ≥ imax + 1. But how can we solve the problem for the case of area under the graph, partitioned to S1 and S2 , and p is the
k ≤ imax ? value maximizing p(1 − F (p)).
Assume that we are only allowed to set prices among values
1, 2, ..., 2imax at each time step. As discussed above, we can show
that the optimum expected revenue in this case will be at least 12 of
the optimum expected revenue in the general case. Now, we pro- probability distributions satisfying the monotone hazard rate con-
pose an algorithm which solves the problem in the case that we are dition, and buyers have identical initial distributions. For this prob-
allowed to set prices to one of values 1, 2, ..., 2imax in each time lem, we present an approximation algorithm whose approximation
step. Let B[t, q] be the maximum expected revenue when we have factor is logarithmic for a constant k and its approximation factor
t time steps and the state of the network is Net-Stable(2q ). In this 1
′ is constant for k ≥ n c for any constant c > 0. We start by stating
situation, we can set the price to 2q in the first time step for q ′ < q. two lemmas from [11]:
Thus, we can calculate B[t, q] as follows:
′ L EMMA 3.1. Let S be the set formed by sampling each element
B[t, q] = max {B[t − 1, q ′ ] + 2q (E[X2q′ ] − E[X2q ])}. (3)
′
0≤q <q from a set V independently with probability at least p. Also let f
An issue with the above equation is that we do not have E[X2q′ ]− be a submodular set function defined over V , i.e., f : 2V → R.
E[X2q ] precisely, but we can estimate it within a small error with Then we have E[f (S)] ≥ pf (V ).
high probability using standard sampling and Chernoff-Hoeffding
L EMMA 3.2. If the valuation of a buyer is derived from a dis-
bound. Let matrix Bs be the matrix corresponding to equation 3
tribution satisfying the monotone hazard rate condition, she will
when we use an estimate value for E[X2q′ ] − E[X2q ] instead
accept the optimal myopic price with probability at least 1/e.
of its exact value in this equation. Since the estimation value for
E[X2q′ ] − E[X2q ] is within a small error of the exact value with Now, we prove a key lemma about probability distributions satis-
high probability, using union bound, we can show that the follow- fying monotone hazard rate condition, which states that the optimal
ing inequalities also hold with high probability: myopic revenue of such a distribution is close to its expected value.
(1 − ǫ)B[t, q] ≤ Bs [t, q] ≤ (1 + ǫ)B[t, q].
L EMMA 3.3. Suppose that f , defined over [a, b], is a probabil-
At the end, the expected revenue of the solution will be stored in ity distribution satisfying the monotone hazard rate condition, with
Bs [k, imax +1]. In order to compute the sequence of prices, we can expected value µ and myopic revenue R = maxp p(1 − F (p)).
′
store the index q ′ which maximizes Bs [t − 1, q ′ ] + 2q (E[X2q′ ] − Then we have R(1 + e) ≥ µ.
Rb
E[X2q ]) while computing Bs [t, q]. By storing these values, we can P ROOF. We know that µ = a 1 − F (t)dt, which is the area
compute the price at time step 1 ≤ i ≤ t by following an appropri- under the graph of figure 3. Also, R is the area of the largest rect-
ate sequence of [t, q] pairs in the dynamic program. Since B[q, t] is angle under that graph. Let p be the price for which p(1 − F (p)) is
between of (1 − ǫ)Bs [q, t] and (1 + ǫ)B[q, t] with high probability, maximized. The area under the graph is the sum of two parts, first
the above algorithm produces a sequence of prices (2p1 , 2p2 , . . . , 2pk ) the integral from a to p, named S1 , and then from p to b, named S2 .
(1−ǫ)
whose expected revenue is within a 2(1+ǫ) of the optimum. According to lemma 3.2, we know that 1 − F (p) ≥ 1/e. As a re-
Changing the algorithm to an FPTAS. We can easily change the sult, we can conclude that S1 ≤ 1 · p ≤ e(1 − F (p))p = eR. Also,
constant-factor dynamic-programming-based algorithm discussed p is the value maximizing p(1−F (p)), so we have h(p) = 1/p (by
above and give an algorithm with approximation factor (1+ǫ) 1−ǫ
2 . To setting p(1 − F (p))′ = 1 − F (p) − pf (p) = 0). Also, since h(p)
do so, we should consider the prices of form (1 + ǫ)i instead of 2i . is a non-decreasing function, we know that for any p′ ≥ p, we have
f (p′ )
Our algorithm is similar to the algorithm of the previous section. h(p′ ) ≥ h(p). Therefore, we conclude that h(p′ ) = 1−F (p′ )
≥
The term log pmax will be replaced with logp1+ǫmax
in the running h(p) = 1/p, and thus f (p′ ) ≥ (1 − F (p′ ))/p. Integrating both
1−ǫ Rb Rb
time and the approximation factor is (1+ǫ)2 . sides from p to b we have p f (t)dt ≥ (1/p) p (1 − F (t))dt. But
Rb Rb
p
(1 − F (t)) is equal to S2 , and p f (t)dt is equal to 1 − F (p).
3. THE Rapid(k) PROBLEM Therefore we have S2 ≤ p(1 − F (p)). So the area under the graph,
S1 + S2 , is at most (1 + e)R. Note that we do not have any condi-
3.1 Identical Initial distributions tions on a and b.
As we will see in subsection 3.2, the Rapid(k) problem is hard to Now, we present an algorithm to approximate Rapid(k). The
approximate even with submodular influence functions and proba- algorithm A is as follows:
bility distributions satisfying the monotone hazard rate condition.
In light of this hardness result, we study a special variant of the 1. Compute a price p0 which maximizes p(1 − F0 (p)) (the my-
problem and give an approximation algorithm for it. We consider opic price of F0 ), and let R0 be this maximum value. Also
the Rapid(k) problem with submodular influence functions and compute a price p1/2 such that F0 (p1/2 ) = 0.5.
 use Sp to denote the area of that part, and ep to denote the length
of the lower edge of that part. We use 3 rectangles for each part
in each step. First, using lemma 3.4 we know that we can use a
Sp
single rectangle to cover at least 1/ log ep of the total area of part p.
ep Then, we cover the two resulting uncovered parts by two rectangles,
which each equally divide the lower edge of the corresponding part.
As a result, four new uncovered parts are created, each with a lower
edge with length less than ep /2, therefore satisfying the necessary
conditions for the next step.
The area that we cover in each stepPof the algorithm is at least
Figure 4: The darker rectangles are selected at the first step, P sp P sp p sp n
p log ep ≥ p log n−(m−1)) = log n−(m−1) (since ep < 2m−1 ).
and the lighter ones at the second step. The fraction of the total covered area by the algorithm after step m,
1
assuming that at each step we cover exactly log n−(m−1) of the re-
2. With probability 21 , let c = 1, otherwise c = 2. maining area, is

3. If c = 1, set the price to the optimal myopic price of F0 (i.e, X

m
1 log n − (i − 1) m−1
· =
p0 ) on the first time step and terminate the algorithm after the log n − (i − 1) log n log n
i=1
first time step.
1
And if at any step i we cover more than log n−(i−1) of the re-
4. if c = 2, do the following: m−1
maining area, the algorithm still covers at least log n of the entire
(a) Post the price p1/2 on the first time step. area after m steps.
(b) Let S be the set of buyers that do not buy in the first As a result, after step m, we have used 4m − 1 = k rectangles,
(4m −1)
day, and let their optimal revenues be R1 (V − S) ≥ and have covered m−1log n
∈ Θ( log log n
) = Θ( log1 n ) of the entire
k
R2 (V − S) ≥ . . . ≥ R|S| (V − S). area.
(c) Let pj be the price which achieves Rj (V − S), and
P rj be the probability with which j accepts pj for any T HEOREM 3.6. The expected revenue of the above pricing strat-
1
1 ≤ j ≤ |S|. Thus we have Rj (V − S) = pj P rj . egy A as described above is at least 8e2 (e+1) logk n
of the optimal
revenue.
(d) Let d1 < d2 < . . . < dk−1 be the indices returned by
lemma 3.4 as an approximation of the area under the P ROOF. For simplicity assume that we are allowed to set k + 1
curve R(V − S). prices. In case c = 1, we set the optimal myopic price of all players
pd and therefore achieve the expected revenue of nR0 . If c = 2,
(e) Sort the prices ej for 1 ≤ j ≤ k − 1, and offer them
consider the second day of the algorithm, assuming that S is the set
in non-increasing order in days 2 to k.
of buyers who have not bought at the first day. By lemma 3.2, we
To analyze the expected revenue of the algorithm, we need the know that each remaining buyer accepts her optimal myopic price
following lemmas: with probability at least 1/e, so for every j we have P rj ≥ 1/e ≥
P ri /e. In addition, we know that for each j ≤ i, Rj (V − S) ≥
L EMMA 3.4. Let P (V − S) ≥ pi /e. We also know that Rj (V − S) ≤ pj . As
i be the index maximizing iai in the set {a1 , a2 , . . . , aRmi}.
Then we have iai ≥ m j=1 aj /(⌈log(m + 1)⌉).
a result, pj ≥ pi /e, for each j ≤ i. Therefore, if we offer the
Pm player j ≤ i the price pi /e, she will accept it with probability at
j=1 aj
P ROOF. Assume for each 1 ≤ j ≤ m, aj < j ⌈log(m+1)⌉ . By least P ri /e (she would have accepted pj with probability at least
summing P rj ≥ P ri /e; offering a lower price of pi /e will only increase the
Pm these inequalities
Pm Pwe have:
a j < ( a
j=1 j )( m 1
) ⇒ ⌈log(m+1)⌉ < probability of acceptance).
Pm j=1 1
j=1 j ⌈log(m+1)⌉
For now suppose that we are able to partition players to k differ-
j=1 j , a contradiction.
ent groups, and offer each group a distinct price. Ignore the addi-
tional influence that players can have on each other.
P In that case, we
L EMMA 3.5. For a set {a1 ≥ a2 ≥ . . . ≥ an }, let D = can find a set d1 < d2 < . . . < dk maximizing kj=1 (dj − dj−1 ) ·
{d1 ≤ d2 ≤ . . . ≤ dk } be the set of indices maximizing S(D) = Rdj (V − S). Assume that Di is the set of players k with di−1 <
Pk k ≤ di . As we argued above, if we offer each of these players the
j=1 (dj − dj−1 )adj (assuming P d0 = 0), over all sequences of
a i
price pdi /e, she will accept it with probability at least P rdi /e. So
size k. Then we have S(D) ∈ Θ( logi n ). the expected value of each of the players in Di when offered pdi /e
k

P ROOF. To give some intuition on the problem, assume the func- is at least P rdi /e · pdi /e = Rdi (V − S)/e2 . The total expected
P
tion f : [0, n] → R such that for x between i − 1 and i, f (x) = ai . revenue in this case will be kj=1 (di − di−1 ) · Rdj (V − S)/e2 ,
P
Our problem is to fit k rectangles under the graph of f , such that which, using lemma 3.4 is at least i Ri (V − S)/(e2 logk n).
the total covered area by the rectangles is maximized. An important observation is that, if the expected revenue of a
We present an algorithm that iteratively selects rectangles, such player when she is offered a price p is R, her expected revenue will
that after the m-th step the total area covered by the rectangles is not decrease when she is offered a non-increasing price sequence P
at least m/ log n using 4m − 1 rectangles. At the start of the m-th which contains p (note that we are still ignoring externalities). As
step, the uncovered area is partitioned into 4m−1 independent parts a result, we can sort the prices that are offered to different groups,
(see figure 4) . In addition, the length of the lower edge of each and offer them to all players in non-increasing order. As argued,
of these parts is at most n/(2m−1 ). The algorithm solves each of this will only increase the expected revenue. Now, considering the
these parts independently as follows. For each part p ≤ 4m−1 , we positive externalities, it is obvious that the expected revenue of the
 ai (K − 2 + 2iǫ) aj
players will not decrease when we take the externalities into ac- activators
count. A

Finally, using Lemma 3.1, and since every player buys at the first di (K − 2 + (2i − 1)ǫ)
day independently with probability 1/2, we conclude that any buyer
i that remains at the second day observe an expected influence of
Ri (V )/2 from all other buyers. b1 (0) b2 (0)
As a result, the expected revenue of our algorithm P is nR0 /2 C
e
c(K − 2)
(from setting p0 with probability 1/2 in the first day) plus i Ri (V )·
(1/8) · (1/(e2 logk n)). Since we set p1/2 with probability 1/2,
a player does not buy at first day with probability 1/2, and we
achieve 1/(e2 logk n) of the value of remaining players in the sec-
ond day. We also know that the expected revenue that can be ex-
tracted from any player is at most E(F0 ) + E(Fi,V ). Thus, using F (K − 2)

lemma 3.3, we conclude that the approximation factor of the algo- X(K − 1) Y (K − 1)
rithm is 8e2 (e + 1) logk n.
Theorem 3.6 tells us that if k is a constant, we can approximate Figure 5: The reduction of subsection 3.2. The number in the
√
Rapid(k) with a logarithmic factor. And in case k = Θ( c n) for parentheses next to the name of a vertex (set of vertices) is the
any c, the problem can be approximated within a factor Θ(1/c). initial valuation of that vertex (vertices). In this instance, the
edge e is adjacent to vertices i and j (in graph G of independent
3.2 Hardness set problem).
In this section, we prove the hardness of the Rapid(k) problem
even in the deterministic case with additive (modular) valuation
functions. Specifically, we consider the following special case of When visiting the i-th couple, if we want vertex i of G to be in the
the problem: (i) k = n; (ii) The valuations of the buyers are deter- independent set, we set the price to K − 1 + (2i − 1)ǫ (causing
ministic, i.e., fi,S is an impulse function, and its value is nonzero both di and ai to buy). Otherwise, set the price to K − 1 + 2iǫ. In
only at vi (S); and finally (iii) The influence functions are additive; both cases the activator will buy, and makes the next couple have
∀i, j, S such that i 6= j and i, j ∈ / S we have vi (S ∪ {j}) = the highest values. From now on, by selecting {i1 , i2 , . . . , il } from
vi (S) + vi ({j}), also each two buyers i 6= j, vi ({j}) ∈ {0, 1}, G or selecting {di1 , di2 , . . . , dil } from G∗ , we mean selling to
and each buyer has a non-negative initial value, i.e, vi (∅) ≥ 0. {di1 , di2 , . . . , dil }.
We use a reduction from the independent set problem; We show The set F is used to represent the edges in E. There is a vertex
1
that using any n1−ǫ -approximation algorithm for the specified sub- fe in F for each edge e in E. The initial values of all these vertices
problem of Rapid(k) , any instance of the independent set problem are K − 2. Let e = (i, j) be an edge in E. There is one edge
can be solved in polynomial time. In an instance of an independent from di , and one from dj to fe . In this way, if two endpoints of
set problem, given a simple graph G = (V, E) and an integer K, an edge e are selected to be in the independent set, the value of
we must specify whether a subset S ⊂ V exists such that |S| ≥ K the corresponding buyer fe increases to K. Otherwise, it would be
and ∀v, u such that (v, u) ∈ E, we have v ∈ / S or u ∈ / S. either K − 1, or K − 2. We are going to build the rest of the graph
For the special case of additive influence functions, it is conve- in a way that if the value of any vertex in F increases to K, we lose
nient to use a graph to represent the influence among buyers, i.e., a a big value. This will prevent the optimal algorithm from selecting
directed graph G∗ has node set V (G∗ ) equal to the set of all buy- two adjacent vertices.
ers, and (j, i) ∈ E(G∗ ) if and only if vi ({j}) = 1. Now we show The set C consists of only three vertices, the two counters b1 and
how to construct an instance of Rapid(k), the graph G∗ , from an b2 , and another vertex c. The initial values of the counters are zero,
instance G = (V, E) of the the independent set problem. Our goal and the initial value of c is K − 2. There are two edges from each
is to determine whether there exists an independent set of vertices, di going to b1 and b2 . Also there is an edge from b1 and b2 to c
larger than a given K. Let N = |V |. The set of vertices of G∗ is (note that b1 and b2 are completely identical). The selection of any
formed from the union of five sets, denoted by A, F , C, X, and vertex di will increase the value of the counters by one. So we can
Y . In the first set A, there are two vertices di and ai for each ver- use the counters to keep track of the number of vertices selected
tex i in G (see figure 5). As we will see later, selling the item to from the set A.
di corresponds to selecting i as a member of the independent set. Finally there are two important sets of X = {x1 , x2 , . . . , xL }
The activator of vertex i, ai , is used to activate (will be made clear and Y = {y1 , y2 , . . . , yL }, where L is a large number to be de-
shortly) the next vertex, di+1 . The initial values of d1 and a1 are termined later. The initial value of all the vertices in X and Y is
K −1+ǫ and K −1+2ǫ, respectively. For i > 1, the initial values K − 1. There is one edge from c to each vertex in X. There is
of di and ai are K − 2 + (2i − 1)ǫ and K − 2 + 2iǫ, respectively. also an edge from each vertex in F , to each vertex in Y . Finally
There are two edges from ai to di+1 , and ai+1 , for each i < n. We we add an edge from each vertex in X to each vertex in Y . These
observe that initially, the first couple have the highest valuations. L2 edges are so important that if we do not take advantage of them
We can consider selling the item to ai , di couples in order. On day when possible, we will be unable to approximate the revenue by
i, we sell the item to ai and we can choose to sell or not to sell to 1
the n1−ǫ factor; we must sell the item for price at least L to all
di . But we do not sell the item to any other buyer. Then next day buyers in Y , if possible. To do so, we have no choice but to select
(day i + 1), with the influence of the ai buyer on the next couple, at least K independent vertices from the G. First observe that all
which we referred to as activation, the (i + 1)-th couple will have the vertices in X are identical, and so are all the vertices in Y . So
the highest values. This allows us to sell to both of them, or only the prices with which any buyer in X buys the item, is equal to the
the activator, without having any other buyer buy the item. Thus we price with which any other buyer in X buys the item. The same ar-
can use the set A to show our selection of vertices for the indepen- gument is true for buyers in Y . In addition, the initial values of all
dent set as follows. Start from d1 and a1 and visit vertices in order. the vertices in X and Y are equal. In order to take advantage of the
L2 edges, we have to find a way to increase the value of the buyers day d − 2, at least K of the di buyers have bought the item. If any
in X to some value more than the value of the buyers in Y . To of these di buyers represent two endpoints of any edge in G. In this
do this, we should activate the only incoming edges of X without case, the valuation of at least one of the players in F will increase
activating any of the incoming edges of Y . This is only possible to K, and she will buy no later than c, which will cause the players
by selling to c, and not selling to any of the vertices in F . We are in Y to be influenced.
going to see that the only possible scenario is to increase the value Therefore if the item can be sold to the buyers of X before the
of b1 and b2 to K by selecting K vertices from V , and making sure buyers of Y buy it, there exists a set of at least K vertices in G
that the selected vertices form an independent set. which are independent.
Our goal is to set L to be so large, that the L2 edges between
X and Y becomes unavoidable. More specifically, if the optimal Now we find a suitable value for L. Assuming the maximum
revenue is greater than or equal to L2 , we have no choice but to revenue is at least L2 , our revenue must be at least n1−ǫ1
L2 . If
sell the item to all the Y buyers for price at least L; the revenue we do not sell the item for price L to all y buyers, the maximum
that is achievable from the rest of the buyers is negligible (less the revenue would be (n − L − 2)K + L(N 2 + N ) + 2N . The upper
1
· L2 ). Although we should make sure that L is polynomial bound of the value of any buyer is K, except b1 and b2 and buyers
n1−ǫ
relative to N = |V |; otherwise our reduction algorithm would not in Y . The counters have an upper bound of N , and the Y buyers
be polynomial. have an upper bound of |E|+K −1 ≤ N 2 +N . We must specify L
1
to be such that n1−ǫ L2 > (n−L−2)K +L(N 2 +N )+2N holds
L EMMA 3.7. If an independent set of size K exists in G, the true. We know that (n − L − 2)K + L(N 2 + N ) + 2N < n(N 2 +
maximum revenue is at least L2 . N ) + 2N < 2nN 2 . Therefore it is enough to prove n1−ǫ 1
L2 >
2 2 2−ǫ 2
P ROOF. We shall describe an algorithm that uses the indepen- 2nN which is equivalent to L > 2n N . Now considering
dent set S to gain a revenue of at least L2 . The algorithm works as that n = 2L + 2N + |E| + 3 ≤ 2L + 2N + N 2 + 3, if we assume
follows. For the first N days, at day i, if i ∈/ S it sets the price to L ≥ N 2 + 2N + 3, we conclude n ≤ 3L. Therefore it is enough to
be an amount so that only ai (and not di ) would buy. Otherwise the show that L2 > (2 · 32−ǫ )L2−ǫ N 2 . Taking a logarithm from both
2−ǫ
price is set so that only di and ai would buy. Stop this at the day D sides and replacing L by N α we get α > 1ǫ (2 + log(2·3
log N
)
).
in which the the K-st buys. We can no longer sell to the players in So the lemma is correct if we set L to be the maximum of (N 2 +
A because the valuations of b1 and b2 are now K. log(2·32−ǫ )
1+ 1 (2+ )
Knowing that all vertices in S are independent, there is no vertex 2N + 3) and N ǫ log N .
in F with a valuation more that K − 1; no two endpoints of any
edge in G has been chosen. On the other hand, |S| = K, hence the T HEOREM 3.9. The special variant of the Rapid(k) problem
1
value of b1 and b2 is equal to K. So on day D + 1 we can sell the defined in this section can be approximated within a factor n1−ǫ
item to both of the b buyers, without selling it to any other buyer, using a polynomial-time algorithm only if the independent set prob-
by setting the price to be K. Then these two buyers would increase lem is solvable in polynomial time.
the value of buyer c up to K, again allowing us to sell only to c at P ROOF. If such an algorithm exists, we can solve the indepen-
the next day. dent set problem by the described reduction. By running the al-
When we manage to sell to c without selling to any of the vertices gorithm on the constructed instance, and based on lemmas 3.7 and
in F , the valuations of vertices in X increases to K, while keeping 3.8, the independent set problem has a positive answer iff the max-
the valuations of vertices in Y at K − 1. We set the price to be K imum revenue from the instance is at least L2 .
on day D+3, causing all the vertices in X to buy. This results in an
large influence on each of the vertices in Y ; the value of the those Having proved the hardness of approximation for the unweighted
buyers would rise to L + K − 1. If we set the price to that value on graphs in which vi ({j}) ∈ {0, 1}, we now show that the prob-
day D + 4 (the last day), all of the L buyers y1 . . . yL would buy lem cannot be approximated within any multiplicative factor when
the item, and we can gain a total revenue of more than L2 . the edges of the corresponding graph are allowed to have arbitrary
weights.
L EMMA 3.8. If there is no independent set of size K in G, it is
impossible to sell the item to the buyers in X before the buyers in T HEOREM 3.10. The Rapid(k) problem with additive influence
Y buy it. functions can not be approximated within any multiplicative factor
unless P=NP.
P ROOF. Let d be the day the buyers in X buy the item. Note
that they all buy at the same day, if they do buy at all. At that day,
none of the buyers in Y should have been influenced by any of the 4. MYOPIC VS. STRATEGIC BUYERS
buyers in F , otherwise the value of the buyers in Y would have An important modeling decision to be made in a pricing problem
risen to K, which is equal to the upper bound for the value of any is to whether model buyers as forward-looking (strategic) or myopic
buyer in X. (impatient). Forward-looking players will choose their strategies
To have the buyers in X buy the item before the buyers in Y do, based on the prices that are going to be set in the future. As a result,
we must sell the item to c before we sell to any yi . This is because they might decide not to buy an item although the offered price
initially the value of any buyer in X is equal to the value of any is less than their valuations. On the other hand, myopic players
buyer in Y and the only edge going into any buyer of X comes are supposed to buy the item in the first day in which the offered
from c. It is obvious that before day d, we can not set a price equal price is less than their valuations. Buyers might be impatient for
to or less than K − 1, or else the y buyers would have bought it. several reasons. This happens when the good can be consumed
Therefore c must have paid more than K − 1 for the item. The only (food, beverage, ...), when the customers make impulse purchases
edges entering c are from b1 and b2 . So before day d − 1 buyers b1 ([29]), or when the item is offered in limited amount (so that it
and b2 must have bought the item. Similarly, these two buyers have might not be available in the future). For example, Zara, one of
paid more than K − 1 for the item. Since their value is always an the biggest apparel retailers, is known to set the stock level so low
integer, they should have paid at least K. This means that before that the customers are encouraged to buy the price instantly, rather
than waiting for the good to go on sale [17]. Also, players might 2003. ACM.
be myopic when the effect of the discount factor is stronger than [14] D. Kempe, J. Kleinberg, and Éva Tardos. Influential nodes in a diusion model
for social networks. In in ICALP, pages 1127–1138. Springer Verlag, 2005.
the possible decrement in prices in the future (that is why we buy
[15] S. Kessing and R. Nuscheler. Monopoly pricing with negative network effects:
electronic devices although we are aware of the discounted prices The case of vaccines. European Economic Review, 50:1061–1069, 2006.
in the future). [16] J. Kleinberg. Cascading behavior in networks: algorithmic and economic
Optimal pricing for myopic or impatient buyers have been also issues. Cambridge University Press, 2007.
studied from algorithmic point of view in the computer science lit- [17] K. F. M. A. Lewis and J. A. D. Machuca. ZaraŠs secret for fast fashion.
[18] R. Mason. Network externalities and the coase conjecture. European Economic
erature. For example, Bansal et. al. [2] study dynamic pricing prob- Review, 44(10):1981–1992, December 2000.
lems for impatient buyers, and design approximation algorithms for [19] E. Mossel and S. Roch. On the submodularity of influence in social networks.
this revenue maximization for this problem. In their setting, buy- In STOC ’07: Proceedings of the thirty-ninth annual ACM symposium on
ers may come and leave the market at different time steps and their Theory of computing, pages 128–134, New York, NY, USA, 2007. ACM.
[20] R. L. Oliver and M. Shor. Digital redemption of coupons: Satisfying and
goal is to find a pricing policy to maximize revenue. dissatisfying eőects of promotion codes. Journal of Product and Brand
In this paper, we focused on the myopic or impatient buyers, and Management, 12:121–134, 2003.
presented all our algorithms and models in this model. However, [21] E. Oswald. http://www.betanews.com/article/
some of our results can be extended to take into account forward- Google_Buy_MySpace_Ads_for_900m/1155050350.
[22] K. Q. Seeyle. www.nytimes.com/2006/08/23/technology/23soft.html.
looking behaviors. One trick in designing a pricing strategy to deal
[23] M. Shor and R. L. Oliver. Price discrimination through online couponing:
with forward-looking buyers is to make sure that the sequence of Impact on likelihood of purchase and profitability. Journal of Economic
prices offered is non-decreasing. If we put this as a constraint in our Psychology, 27:423–440, 2006.
pricing policy, at least, we know that even forward-looking buy- [24] P. SÃÃskilahti. Monopoly pricing of social goods. MPRA Paper 3526,
University Library of Munich, Germany, 2007.
ers should act similar to myopic players, since if they do not buy
[25] A. Sundararajan. Local network effects and network structure. Working paper,
the item at any time, the price of the item may only increase. We 2004.
can define the Rapid(k) and Basic(k) problem with this additional [26] A. Sundararajan. Nonlinear pricing of information goods. Management Science,
property that the price sequence should be non-decreasing. 50:1660–1673, 2004.
Here, we observe that most of our results can be extended for [27] A. Sundararajan. Network effects, nonlinear pricing and entry deterrence.
Working paper, 2005.
the non-decreasing variant of the Rapid(k) problem. First of all, [28] J. Thépot. A direct proof of the coase conjecture. Journal of Mathematical
our hardness result for the deterministic case can be extended for Economics, 29(1):57 – 66, 1998.
the non-decreasing variant of the problem. In particular, in the re- [29] G. van Ryzin and Q. Liu. Strategic capacity rationing to induce early purchases.
Management Science, 54:1115–1131, 2008.
duction from the independent set problem, the price sequence we
[30] R. Walker. http://www.slate.com/id/1006264/.
offered for the case when the independent set of size k exists is a [31] T. Weber. http://news.bbc.co.uk/1/hi/business/6305957.stm?lsf.
non-decreasing sequence. As a result, the hardness result holds for
the non-decreasing variant of the Rapid(k) problem.
We note that even with non-decreasing pricing policies, other 6. APPENDICES
strategic aspects of buyers makes the pricing policy challenging.
For example, strategic buyers may even buy an item earlier as they
expect that the price of the item and the number of people who APPENDIX
buy the item increases. It would be interesting to extend the rest
of our results for the case of forward-looking buyers, and study
A. EXTENDING TO NON-DECREASING VER-
the equilibria of the iterative pricing problem in the presence of SION
strategic buyers. Remember from subsection 3.1 that the pricing scheme A would
perform as follows when the number of days equals 2. It tosses a
5.[1] Y.REFERENCES fair coin to decide whether c = 1 or 2. If c = 1, it sets the price p0
Aviv and A. Pazgal. Optimal pricing of seasonal products in the presence of and terminates. If c = 2, it sets the price p1/2 in the first time step,
forward-looking consumers. Manufacturing and Service Operations
Management, 10:339–359, 2008. and p/e in the second time step, where p is the optimal myopic
[2] N. Bansal, N. Chen, N. Cherniavsky, A. Rudra, B. Schieber, and M. Sviridenko. price of a player i which maximizesP iRi . This scheme achieves the
Dynamic pricing for impatient bidders. In SODA, pages 726–735, 2007. expected revenue of nR0 /2 + i Ri (V ) · (1/8) · (1/e2 log n).
[3] B. Bensaid and J.-P. Lesne. Dynamic monopoly pricing with network Now modify the algorithm to terminate after the first day in case
externalities. International Journal of Industrial Organization, 14(6):837–855,
October 1996. c = 2 and p/e < p1/2 . Let A′ be this new pricing scheme. Clearly,
[4] J. I. Bulow. Durable-goods monopolists. The Journal of Political Economy, this gives a non-decreasing pricing scheme. We now prove that the
90(2):314–332, 1982. expected revenue of this scheme is asymptotically the same as the
[5] L. Cabral, D. Salant, and G. Woroch. Monopoly pricing with network previous one. Consider the case of p/e < p1/2 . In this case, we
externalities. Industrial Organization 9411003, EconWPA, Nov. 1994.
[6] R. H. Coase. Durability and monopoly. Journal of Law & Economics, have Ri ≤ p < ep1/2 = 2e(p1/2 · (1/2)) ≤ 2eR0 . From the
15(1):143–49, April 1972. analysis of scheme A, we know that our expected revenue when
[7] P. Domingos and M. Richardson. Mining the network value of customers. In we set p/e in the second day is iRi /e.
KDD ’01, pages 57–66, New York, NY, USA, 2001. ACM.
[8] J. Farrell and G. Saloner. Standardization, compatibility, and innovation. RAND
Journal of Economics, 16(1):70–83, Spring 1985. B. MISSING PROOFS
[9] F. Gul, H. Sonnenschein, and R. Wilson. Foundations of dynamic monopoly and
the coase conjecture. Journal of Economic Theory, 39(1):155–190, June 1986. We use the following theorem in our proofs.
[10] O. Hart and J. Tirole. Contract renegotiation and coasian dynamics. Review of
Economics Studies, 55:509–540, 1988. T HEOREM B.1. (Chernoff-Hoeffding Bound) Let X1 , ..., Xn be
[11] J. Hartline, V. S. Mirrokni, and M. Sundararajan. Optimal marketing strategies i.i.d. (independent and identically distributed) random variables
over social networks. In WWW, pages 189–198, 2008.
[12] M. L. Katz and C. Shapiro. Network externalities, competition, and
over a bounded domain [0, 1] with expectation E[XP i ] = µ (for all
compatibility. American Economic Review, 75(3):424–40, June 1985. i). Denote by X the average of X1 , ..., Xn , X = n1 n i=1 Xi . For
[13] D. Kempe, J. Kleinberg, and Éva Tardos. Maximizing the spread of influence ǫ2 nµ
through a social network. In KDD ’03, pages 137–146, New York, NY, USA, all 0 < ǫ < 1, P r[|X − µ| > ǫµ] ≤ 2e− 3 .
 L EMMA B.2. Given a price p, for every ǫ > 0, and δ < 1,
|V |
there is an O( ǫ2 E[X p]
log( 1δ )·ROU N D) algorithm which returns
value Xp such that P r[|Xp − E[Xp ]| > ǫE[Xp ]] ≤ δ.

T HEOREM B.3. For every ǫ > 0 and δ < 1, there exists a

polynomial time algorithm which finds a price p such that E[Cp ] ≤
1−ǫ
E[COP T ] (1+ǫ) 2 with probability at least 1 − δ.

P ROOF. The algorithm is as follows:

p /pmin
1. Let imax = log1+ǫ
max
+1.
2. Define values pi = (1 + ǫ)i pmin for any integer i, 0 ≤ i <
imax , and compute E[Cpi ] for each pi .
3. Return pi with maximum calculated E[Cpi ].
We prove that the above algorithm has our desired property. We
know that pmin ≤ pOP T ≤ pmax . Therefore there is an index j
such that pj ≤ pOP T < pj+1 . A specific people who has bought
the item when the price is pOP T will buy the item when the price
is pj certainly. Because the price decrease from pOP T to pj . So
for every random sampling we have Xpj ≥ XpOP T which means
E[Xpj ] ≥ E[XpOP T ]. We can conclude that:

(1 + ǫ)E[Cpj ] = (1 + ǫ)pj E[Xpj ] ≥ pOP T E[Xpj ]

≥ pOP T E[XpOP T ] = E[CpOP T ]
In order to estimated to E[Cpi ] = pi E[Xpi ] for every i, Let δ ′ =
δ
imax
in Lemma B.2. Let fi be the estimated value for E[Cpi ]. For
every i the value of fi will be out of interval [(1 − ǫ)E[Cpi ], (1 +
ǫ)E[Cpi ]] with probability at most δ ′ . So with probability at least
P = (1−δ ′ )imax ≤ 1−δ ′ imax = 1−δ we have (1−ǫ)E[Cpi ] ≤
fi ≤ (1 + ǫ)E[Cpi ] for all i. Now assume the algorithm returns po
as the best price. We know that with probability at least 1 − δ all
fi will be near to E[Cpi ]. So with probablity at least 1 − δ we can
bound E[Cpo ] With E[CpOP T ] as follows:
fo fj 1−ǫ 1−ǫ
E[Cpo ] ≥ ≥ ≥ E[Cpj ] ≥ E[CpOP T ]
1+ǫ 1+ǫ 1+ǫ (1 + ǫ)2
|V |
Our algorithm run in O( ǫ2 E[X p]
log( 1δ )·imax ·ROU N D) time.