Sie sind auf Seite 1von 32

Waterfall versus Sprinkler Product Launch

Strategy: Influencing the Herd


Manaswini Bhalla
Department of Economics
The Pennsylvania State University

Abstract
The herd behavior literature has concentrated on the social learning phenomenon where players do not influence the rate and degree of learning. However, under many economic scenarios agents can affect the speed of learning
through their actions, e.g. prices. We analyze the problem of a monopolist
that can influence consumer learning about its product by manipulating its
launching sequence in various regions and prices.
Every period, the monopolist decides the number of launches and the price of
its product. The firm and consumers do not know the true underlying quality of the product. However, consumers in each market inspect the good and
receive a private signal about its quality. They also learn about the quality from observing previous sales and prices. We find that initially, the firm
prefers prices and product launch strategies that allow greater transmission of
information from current to future buyers. The price sequence is found to be
supermartingale, i.e. the price decreases with time. Initially, the monopolist
prefers to launch the product at higher prices to influence learning. The number of trials introduced by the firm is a monotonic function of the belief about
the quality of the product. As the belief of the product increases the likelihood of launching the product sequentially, decreases. The number of trials is
found to be a submartingale, i.e. the number of launches increases with time.
However, the amount of learning encouraged by the monopolist is lower than
that under a social planner.
JEL Classification: C73, D62, D81, D82, D83
Keywords: Product Launch, Social Learning, Endogenous

I would like to thank my advisor Professor Kalyan Chatterjee for his valuable suggestions.

Introduction

Under uncertainty, agents usually rely on the decisions of others. For example, a
consumer deciding whether to purchase a particular good obtains valuable information by analyzing the consumption behavior of others. In cases when the quality
of the product is not verifiable, private information of other agents is more important to the decision maker. In this active learning literature, seminal papers like
Banerjee([3]) and Bikhchnadani etal ([4]) find that socially wasteful informational
cascade 1 eventually arises when agents rationally learn from each others actions.
This literature focuses on the social learning environment where agents do not influence the rate and degree of learning. However, under many economic scenarios
players can affect the speed of learning through their actions like the price charged by
a firm. This paper analyzes how a monopolist can influence consumer learning and
the eventual rise of a herd by controlling the prices and the number and sequence of
markets to launch the product in each period. Unlike previous literature, the role of
the monopolist is not passive and it is allowed to control learning through prices and
number of trials. One of the major issues facing firms in their decision of a global
launch of a new product is the pricing and sequencing of entry into the international
markets. This paper studies how these strategies can be used to the advantage of a
monopolist.
Waterfall or the hierarchical product launch strategy is a popular model for the
global rollover. Pioneered by Ayal and Zif [1], in a waterfall launch, products are
launched in a few regions, initially. The experience and success of the product in
these regions trickles down to the rest of the regions in a slow moving cascade. Thus,
after the successful domestic launch of a new product in a few regions, firms are
seen to launch it in other regions. Evidence of such a launching phenomenon has
been documented by Davidson and Harrigan [5]. They reported that during 194576, in planning for the global roll-over, U.S.-based multinationals initially focused on
English speaking markets (such as Canada or the United Kingdom), then on other
industrial markets, and finally to the less developed countries. It is also documented
2
that movies are released in a sequential order across the world. Consumers may
decide to observe the performance of the movie in other box offices before deciding
whether to see the movie when it is available. Production houses may wish to exploit
this consumer behavior in determining their launch and pricing strategies. Another
Information Cascade is the event where consumers disregard their own private information and
follow the average actions of their predecessors
2
www.imdb.com
1

example of a sequential launch is that of Colgate Sensation. Colgate Sensation was


launched sequentially first in Nordic countries then Mid European and then to the
Mediterranean countries. Another recent example of sequential product launch strategy is Apples i-phone. Apple launched its first iPhone only in U.S.A. on 29th June
2007. Subsequently, it was released in Germany on 8th November,2007 for 628 euros
and on 28th November in France for 399 euros. It was later released on 14th March,
2008 in Ireland, Austria for 399 euros(8GB). Recently, it was launched in 13 other
countries across the world. Though regulatory and preference differences across regions could be the stumbling blocks to a simultaneous product launch. Apple could
be exploiting the social learning aspect of consumer behavior by introducing the
product sequentially.
On the other hand, Omhae [6] and Riesenbeck and Freeling [7] advocate a sprinkler diffusion strategy or a simultaneous world attack. They suggest that entry
in all markets is the only viable choice in todays global marketplace as compared
to the waterfall diffusion strategy. It has been documented that Microsoft launched
Windows 95 in a similar fashion where 4-6 million customers worldwide bought the
operating system in the first three weeks after the launch. Thus, the marketing literature is unclear about the optimality of these two launch strategies. This paper
bridges the gap by shedding light on the optimal launch and pricing strategy in an
environment where both the monopolist and the consumers learn about the underlying quality of the product.
A few papers look at the role of prices in influencing the information flow across
the consumer base. For example, Welch [13] analyzes the optimal price to maximize the success of a new security to a sequence of partially informed investors. He
however considers, fixed-price sales, since the Securities and Exchange Commission
has banned variable-price sales in initial public offerings. Unlike Welch, this paper
characterizes the optimal flexible-price and launch strategy for a monopolist. Ceminal and Vives([8]) studies the price dynamics of a strategic firm in the presence of
duopoly and consumer learning about the uncertain quality of the products. It is
found that the firms manipulate consumers beliefs through prices. Consumers infer
the quality of the good only through the past market shares as the prices are assumed
to be unobservable. On the other hand, in our model, consumers take decisions based
on both the prices and the number of sales in the period. Bergemann and Valimaki
[10, 11] in a sequence of papers analyse the role of resolving the uncertainty about
the products quality through prices. In their model, the number of consumers in
each period remains constant. In contrast, in our model, the number of consumers is
3

endogenous. Also, they analyze the problem in a word of mouth envionment, where
consumers share their ex-post experiences directly to each other. In our model, consumer purchase decisions reveals the ex-ante private information.
Sgroi[12] analyzes the problem of a social planner and a monopolist that may influence the consumer learning by determining the number of offers to make in the
first period. Our paper differs from this work in many ways. First, the monopolist
in our problem can use both the number of offers and the price of the product as
an instrument to influence the herd. Second, in Sgroi the monopolist knows the
quality of the product whereas consumers are unaware of it. We on the other hand
analyze both two sided and one sided problem of learning. Also, Sgroi analyzes a
static problem where the monopolist chooses the number of trials at the beginning
of the game. We on the other hand solve a dynamic problem where the monopolist
can determine the number of offers in each period.
Bar-Issac[9] studies a dynamic learning model where the firm is privately informed
about the quality of the product and the consumers learn about the quality over
time. In contrast, in our paper the consumers are privately informed and both the
firm and the buyers learn the quality over time. Also, in Bar-Issac the seller only decides whether to trade or not. Once the trading decision is made, the product is sold
to all the consumers in that period. In contrast, we focus on the sellers launching
and pricing strategy in each period where the firm not only decides whether to trade
or not but also how many offers to trade and at what price. This paper is therefore
the first to analyze the role of both the prices and number of launches in a model of
bilateral learning.
We consider the situation of bilateral learning where neither the monopolist nor
the consumers are aware of the true value of the product. Though the monopolist
would know the characteristics of the product it could be unaware of the preferences
of the consumers. At the same time the consumers could be unsure of the true
quality and characteristics of the product. A monopolist can influence the decisions
of buyers and the social information accumulation by the price and the number of
trials that it allows in every time period. Prices and the number of trials play two
roles- rent extraction and information revelation. On one hand, greater number of
launches and higher prices increases current period rent. However, launching the
product at less than the maximum number of segments and at a higher price, may
favorably influence the belief about the product in the next period.

All buyers in the same region have similar preferences, and decide whether or not to
purchase a unit of the good offered by the monopolist. When the product is launched
in the market the consumers inspect the good and receive a private signal about the
quality of the product. This signal may be generated by the local advertising campaign or by the media. Each consumer decides whether to buy after observing the
price, a signal about the quality of the good, the decisions of previous consumers and
the previous prices. Observation of the behavior of other players is crucial to learn
about the quality of the product. Future consumers learn from previous purchases
and update their belief about the uncertainty. Thus, launching performances in previous markets influences sales in the remaining markets.
When the monopolist determines the number of regions to launch its product in
the current period, it may face a tradeoff between more informed purchase decisions
made by future decisions and less informed purchase decisions made by current consumers. We find that initially, the firm prefers prices and product launch strategies
that allow greater transmission of information from current to future buyers. That is
the monopolist sets the prices high and launches it at fewer than the maximum number of segments. An the act of consumption at the high price positively influences
the belief about the quality of the product in future. This favorable information to
others allows subsequent consumers to infer the private information of the buyer and
value it highly. Price is a function of the last periods belief. The price sequence
is found to be supermartingale, i.e. the price decreases with time. The number of
launches introduced by the firm is a monotonic function of the belief about the quality of the product. As the belief of the product increases the likelihood of launching
the product sequentially, decreases. The number of number of launches introduced
by the firm is found to be a submartingale, i.e. the number of launches increases
with time. We also observe that greater the horizon of the game and the number
of untapped consumers in the economy higher the probability that the seller would
allow experimentation. However, the amount of learning encouraged by the monopolist is lower than that under a social planner.
The paper is organized as follows: Section 2 explains the basic model of social learning. We characterize the equilibrium for the finite and infinite period problem in
Section 3. Section 4 compares the monopolists optimal policy with that of a social
planner. In section 5 some extensions of the model are discussed. Section 6 concludes

Model

This section describes the basic model of product launch and pricing decision of a
monopolist.

2.1

The Environment

There is a countable set of dates, T = {1, 2, 3, . . . } and N number of


regions. The regions represent a set of consumers which could belong to different
geographic or demographic groups. At each time period, t T , the monopolist
decides the number of regions, t and the price, pt < at which to launch its
product. The monopolists product is assumed not to be introduced in any region
at the beginning of the game. Each region i N has mi number of consumers.
To begin with we assume mi = m i. Denote M t1 to be the set of all regions
where the product can be launched at the beginning of t period. The product can
be launched in a region only once. Therefore, M t M t+1 . For every t, i such that
i M t1 denote lit = 1 if the product is available in region i at the end or during
the time period t and 0 otherwise. We do not explore the possibility of the product
being withdrawn from a region. Thus, if lit = 1 then lih = 1h t.
All consumers in region i such that lit = 1, decide whether to consume the product
or not 3 . Each consumer consumes the product only once. The consumers derive
utility from an unknown value of the good. The quality of the product of the
monopolist, could either be superior to the outside option, i.e. = G or inferior,
= B. At the beginning of the game all agents in all regions have the same common
prior about the state of the world, q0 = P r(G) 4 .

2.2

Signal Structure

The monopolists product is assumed to be an inspection good. Denote by ji the


consumers of region i. All consumers, ji i M t1 s.t. lit = 1 inspect the product
before their purchase and receive independent signals, sji {g, b} about the quality
of the product. For example, the consumers are allowed to inspect cars before a
purchase or are able to preview a movie before viewing the entire film. Also, most
products have showrooms where the products are on display for inspection. There is
The consumers are not strategic and assumed to be highly impatient. Thus, all consumers
in the a region decide immediately. Currently, the game restricts information spill overs within
countries. By making the consumers more strategic we hope to introduce both intra and inter
region diffusion.
4
The common prior in all regions could be a result of uniform advertising across all regions.
3

no cost of receiving such a signal. The signal sji is private information of each agent
ji . However, it is common knowledge that the signals are distributed independently
given the state of the world according to Table 1. 0.5. can be interpreted as
Table 1: Signal Structure
P (sji |)
G
B

b
1

the precision of the signal received. This precision is assumed to be the same across
all regions.
Given the signal received sji and the common prior at the end of period t 1 or
beginning of period t, q t1 , all consumers, ji i M t1 such that lit = 1, update their
belief about the quality of the product. Let fs(k) (q t1 ) = P r(G|s(k), q t1 )s(k)
{g, b} be the updated probability of the state 1 for the consumer that receives signal
s(k) {g, b} given q t1 , the common prior at the end of period t 1.
fg (q t1 ) =

q t1
q t1 + (1 )(1 q t1 )

fb (q t1 ) =

(1 )q t1
(1 q t1 ) + (1 )q t1

Notice that fg (q t1 ) and fb (q t1 ) are increasing functions in q and concave and convex
respectively. Conditional on the updated prior, fk (q t1 ) each consumer, ji i M t
such that lit = 1 decides whether to consume the product or not.

2.3

Strategy

The history at any time period t, ht H is a collection of all the actions taken by
the monopolist, k , pk and that of the consumers, k k < t
Monopolist
At the beginning of every period, t the monopolist has two decisions to take. First,
it has to decide the number of regions that it wishes to launch its product, t (ht ).
Second, it must also decide the price, pt (ht ) at which to launch it. Since the consumers in every region are homogenous in preferences, the price would be same in all
regions at any point in time t. The payoff relevant history for the monopolist is q t1 ,
7

the common prior at the end of period t 1 and the number of regions where the
product has not been introduced at the end of period t 1, M t1 Thus, the action
space of the monopolist at every point in time t is
(q t1 , M 1 t) = t (q t1 , M t1 ) pt (q t1 , M t1 ) M t1 R
Consumers
The strategy of all consumers ji is given by jti (ht , sji ) = {1, 0}ht H, ji , i. Where,
jti = 1 implies that the consumer bought the monopolists product. For all ji such
that i M t1 , lit = 0, ji = 0. However, for all agents ji such that i M t1 , lit = 1,
the payoff relevant history is q t1 . Thus the strategy can be written as
jti (q t1 , sji ) = {1, 0}ji , i M t1 , q t1 [0, 1]tsji {g, b}
Once an act of consumption t is made and observed by all, the common prior is
updated. The updated common prior is denoted by, q t = P r(G| t , t , q t1 ). Given
the updated prior q t , the firm and the rest of the consumers take their decisions in
the next period.

2.4

Payoffs

Consumers
Each consumer derives utility, rji from the outside option. For now we assume that
all agents in all regions are homogenous and have the same reservation utility, rji = 0
5
. The consumers utility is given by the following table.
When the true state of the world is Good(G) (or Bad(B)) the utility from conTable 2: Realized utility of agent ji

aj i = 1
aj i = 0

=G
z
z

=B
z
z

suming(or not consuming) the product is greater than that from not consuming (or
consuming) the product. If instead of z we assume that the payoff from taking
an incorrect action is z 0 z, the explicit value of the cutoffs would change but the
Section 6 discusses the extension of the model where the reservation utilities are different across
regions.
5

nature of the solution would remain the same. Hence, for mathematical convenience
we assume z 0 = z.
Monopolist
The monopolist maximizes the discounted expected payoff. The realized utility of
the monopolist at time period t,
t (q t1 , M t1 ) = pt (q t1 , M t1 )ct (q t1 , M t1 )t T
where, ct (q t1 , M t ) is the number of adoptions of the product at time period t. Since
we assume that the consumers are impatient and hence have no incentive to wait, the
number of consumers that would try the product at time period t would be equal to
the sum of the population of
Pthe regions where the product was launched in period
t. That is ct1 (q t1 , M t ) = i mi i s.t. lit = 1.
The dynamic problem solved by the monopolist is the following.
V t (q t1 , M t1 ) = M ax,p EV [p(q t1 , M t1 )c((q t1 , M t1 )] + EV (q 0 , M 0 )

3
3.1

Equilibrium Analysis
Example, T=2

In this section the discounted dynamic optimization problem of the monopolist which
intends to launch the product in N regions over two time periods is solved. In any
period the monopolist can decide not to sell, which gives a payoff of zero. However,
if the firm wishes to launch the product then it must do so at either of these two
prices: Separating price, PS (q) = 2fg (q) 1, the maximum price that the agent
which receives a good signal is willing to pay to for the good; and the Pooling price,
PP (q) = 2fb (q) 1, the maximum price at which the consumer that receives a bad
signal wishes to purchase the product, where q is the common public belief about
the state of the world at the beginning of the period. PS (q) sells with probability
P r(s1 |q), and yields an immediate payoff of q (1 ), while PP (q) yields the
immediate payoff of 2fb (q) 1 with certainty. The pooling price PP (q) stops social
learning. This is because both type of current consumers purchase the product at this
price. The future consumer can not infer the signal of the immediate predecessor who
buys the good regardless of the signal received. This implies that after a pooling price
the public belief in the next period will remain the same. However, the separating
price, PS (q) sells only to part of the market or only to the high type consumer. A
purchase at this price would reveal the signal received by the current consumer to
9

the future consumers. It allows for revelation of signals and hence for social learning.
The three prices Separating Price, Pooling Price and Exit are the strictly dominant
strategies of the firm. Therefore, from now on only these three dominant prices of
the monopolist would only be considered
The pricing decision of the firm that wishes to launch the product simultaneously is
given by the following lemma. This result also applies for the pricing decision of the
firm in the last period of a finite horizon sequential launch.
Lemma 1
The pricing decision of the firm in a simultaneous launch is the following:

0
x

Exit

xy

0.5
Screening

1
xy

Pooling

Proof. In any time period the monopolist can decide not to sell. If instead the
monopolist wishes to sell she has two prices to choose from. PS (q) = 2fg (q) 1, the
maximum amount that the high type consumer or the consumer that receives a high
signal is willing to spend; and the Pooling Price, PP (q) = 2fb (q) 1, the maximum
price that the low type is willing to spend on the good. All other prices being
strictly dominated by one of these. Notice that PS (q) > PP (Q). The separating
price,PS (q), 2fg (q) 1 sells with probability P r(s1 |q) and leads a profit of ES =
P r(s1 |q)(2fg (q) 1) = q (1 ). The pooling price, 2fb (q) 1 sells with certainty
and leads a profit of EP = 2fb (q) 1.
q 1 , PP (q) < PS (q) < 0. Hence the optimal strategy of the firm in the
last period is to not launch the product q 1 .
1 q , PP (q) < 0, PS (q) > 0. Hence the optimal strategy of the firm would
be to sell only to the high type at the separating price.
q , the separating price will be charged if the associated expected payoff q
(1 ) is higher than 2fb (q) 1. As can be observed from the diagram, that
the pricing
strategy of the firm is to charge a separating price for all q q =
(1)2 +

(1)4 +2 (21)
(21)

For a low perception about the quality of the product the firm would not bother
10

2f0 (q) 1

q (1 )

[1 ]

Figure 1: Last period pricing strategy.


to enter the market. The monopolist can choose one of the three above mentioned
prices. An act of consumption at the separating price would imply that the agent
received a good signal and hence would favorably update the belief. This would
enable the seller to sell at a higher price, though with a lower probability of making
a sale. If on the other hand the monopolist chooses the pooling price it would ensure
a sale, though at a lower price. The difference in the expected utility of the two types
and hence the two prices decreases with the increase in the quality of the product.
Thus, the benefit of launching the product at a separating price would decrease as
q increases. Hence, there exists a threshold precision level such that the expected
profit from selling to all equals the profit from selling only to the high type. Thus,
the firm would sell only to the high type for intermediary beliefs and sell to all for
the high level beliefs.
Lemma 2
If the monopolist launches its product sequentially at time period t, i.e. t < N then
pt would be the separating price. i.e. pt = 2fg (q t1 ) 1
Proof. Let us assume that if t (q t1 , M t ) = n < N then pt (q t1 ) 6= PS (q t1 ) =
2fg (q t1 ) 1. Or else, if t (q t1 , M t ) = n < M t then pt (q t1 ) = PP (q t1 ) =

11

2fb (q t1 ) 1. The expected utility of the firm would then be


X
X
V (q t1 , M t ; t = n, pt = 2fb (q t1 )1) =
mi [2fb (q t1 )1]+V (q t1 , M t
mi )
in

in

However, if pt (q t1 ) = PP (q t1 )
V (q t1 , M t ) = M ax{0, M t (q (1 )), M t [2fb (q t1 ) 1]}
Thus V (q t1 , M t ; M t ) V (q t1 , M t ; t = n, pt = (2fb (q t1 ) 1)) = t [2fb (q t1 )
1] + V (q t1 , M t n). This however, contradicts the assumption that t (q t1 ) = n.
Hence, if t (q t1 , M t ) = n < N then pt (q t1 ) = PS (q t1 ) = 2fg (q t1 ) 1.
The lemma above shows that if the monopolist launches the product sequentially
it must do so at a separating price. Launching the product at a pooling price would
imply that the prior about the quality of the product would not change in the next
period. Launching the product sequentially comes at a cost of lower current expected
profit. Sequential launch is profitable only if it favorably changes the prior in the
next period. A sequential launch at a pooling price would not change the prior in the
next period. Thus, if the firm finds it profitable to launch the product at a pooling
price in less than M t regions, it would do better by launching the product simultaneously in all regions. Hence, the firm would always launch the product sequentially
at a separating price.
Before we proceed let us introduce some notation. The updated common prior at
the end of t periods, q t can be written as a function of the number of regions that
the product was launched, t (or the number of trials induced in period t, mt ), the
number of adoptions in period t, k t and the last periods common prior, q t1
q(t , k t , q t1 ) = P r(1 |t , k t , q t1 )
Proposition 1
The launch strategy of the firm for N 2 when quality is private information is the
following:

12

0
x Exit y

q12 (M )
1
x Seq,n=1 y
x

Seq,n 1 y

q22 (M )
q
1
x Sim. PS y x Sim. PP y

Proof. Let q0 be the common prior about the quality of the product at the beginning
of period 1. Also, let M be the number of regions left to launch at the beginning of
time period 1. Since, this is the first period M 1 = N .
The value function faced by the monopolist at the first time period is given by
the following expression
V 2 (q0 , M ) = M ax{0; n[q0 (1 )] + EV 1 [q, M n]n < M ; M [q0 (1 )]
; M [2fb (q0 ) 1]} (1)
Step 1 q0 1
The monopolists value function can be reduced to
V 2 (q0 , n, M ) = M ax{0; n[q0 (1 )] + EV 1 [q, M n]}
We know, V 1 (q, n, M ) = M ax{0, M [q (1 )]; M [2f0 (q) 1]}, q. V 1 (q, M ) and
hence EV 1 (q, M ) are increasing in q. Thus, L(q0 ) = n[q0 (1)]+EV 1 [q, M n] is
increasing in q0 . L(q0 = 0) 0, L(q+0 = 1) > 0. Thus q12 (n, M ) [0, 1] such
that L(q12 (n, M )) = 0. There exists q12 (M ) = min[q12 (M, n)]n such that q q12 (M )
the monopolist would exit the market.
Step 2 q [1 , q]
The monopolists value function is
V 2 (q0 , n, M ) = M ax{n[q0 (1 )] + EV 1 [q, M n], M [q0 (1 )]}
Define R(q0 , n, M ) = n[q0 (1 )] + EV 1 [q, M n] M [q0 (1 )]. R(q0 , n, M )
0 (q)
0.5q0 q. Also, R(q0 = 1 , n, M ) 0, R(q0 =
is decreasing in q0 as fq
2
q, n, M ) 0. Thus, there q2 (M, n) [1, q] such that R(q0 = q22 (M, n), n, M ) = 0
and q0 q22 (M, n) the monopolist wishes to launch the product simultaneously in
all M regions than to launch it in n < M regions. Thus there exists, q22 (M ) =
13

M ax[q22 (M, n)]n such that q q22 (M ) the monopolist would launch the product
simultaneously.
Step 3 q q
The monopolists value function is
V 2 (q0 , n, M ) = M ax{n[q0 (1 )] + EV 1 [q, M n], M [2f0 (q0 ) 1]}
Define, T (q0 , n, M ) = n[q0 (1 )] + EV 1 [q, M n] M [2f0 (q0 ) 1]. T (q0 , n, M )
0 (q)
is decreasing in q0 as fq
0.5q0 q. Also, T (q0 = q, M, n) = R(q0 = q, n, M )
0, T (q0 = 1, M, n) < 0. Thus, for all q q the firm would launch the product
simultaneously.
Step 4 q [q12 (M ), q22 (M )] The proof for this step is relegated to Proposition 2 where
the optimal launching sequence for the general finite horizon game is solved.
In the first period, for q < 1 the monopolist can either exit the market or
launch the product sequentially at a separating price. If the monopolist exits the
market it receives a payoff of 0. However, if it launches the product sequentially the
monopolist would bear a current loss by selling at price, PS (q) < 0. However, selling
at a separating price sequentially would lead to a possibility of an improved q in the
next period and hence a higher revenue. This is because an event of a sale at the
separating price would indicate that the current consumer received a high signal. A
successful experiment in the first period may ensure sales at a higher belief and price
in the next period. The benefit of an improved belief about the quality of the product in the next period and hence a higher future expected revenue increases with q.
However, the loss of making a sale in the current period at a negative price of a discount decreases with q. Thus, there exits a threshold q1N (M ) such that q q1N (M )
the firm is willing to bear loss in the current period to improve the belief favorably
in the next period. Hence, the firm launches the product sequentially only to one
region q q1N (M ).
For a belief greater than 1 the monopolist can either launch the product sequentially or simultaneously at a separating price. Launching the product sequentially
improves the belief about the quality of the product in the next period. However,
this advantage comes at a cost of a lower immediate discounted payoff in the first
period. Launching the product simultaneously would give an advantage of a greater
immediate discounted payoff. However, the firm would lose out on the possibility of
making a sale at a higher price to future consumers. This advantage of making a
sale at a higher price decreases with an increase in q. Thus, there exists a threshold,
q2N (M ) such that q q2N (M ) the firm would launch the product simultaneously.
14

Also, q [q1N (M ), 1 ] the firm launches the product in one region sequentially.
q [1 , q2N (M )] the firm launches the product in more than or equal to 1 region.

3.2

Finite Time Horizon

The monopolist solves a discounted problem with bounded returns per stage. Before
proceeding with the explicit solution of the problem for T < the properties of the
value function of the finite horizon problem are characterized. The value function
faced by the monopolist at time period N < T is given by the following expression
V N (q, M ) = M ax{0 + V N 1 (q, M ); n[q (1 )] + EV N 1 [q, M n];
M [q (1 )]; M [2f0 (q) 1]}

(2)

where q is the belief about the quality of the product and M is the number of regions where the product has not been introduced at the beginning of time period
t. For notational convenience we drop the time superscripts. Clearly, V N (q, M )
is the maximal return that can be obtained starting at q, M and proceeding for N
stages. The four arguments in the left hand side correspond to the four un-dominated
strategies available to the monopolist. The first argument in the value function corresponds to the expected payoff from exiting the market. If the monopolist decides
to sell the product in only n regions, the expected immediate payoff is equal to
n(q (1 )) from lemma 2 and the discounted expected continuation payoff is
EV N 1 [q, M n]. However, the firm can launch the product together in all M
regions at a pooling price, PP (q) = 2f0 (q) 1 which sells with certainty and gets
revenue M (2f0 (q) 1). Or the firm could sell the product simultaneously at a separating price,PS (q) = 2f1 (q) 1 which sells with probability, P r(s1 |q) and generates
the immediate payoff of M (q (1 ).
The value function is continuous and non-decreasing in q, M for any N , as it can be
immediately verified.
Lemma 3 (Continuity and Monotonicity in q, M of the value function for
the finite horizon problem)
The value function V N (q, M ) is a continuous function of q an M. V N (q, M ) is increasing in q and M
Proof. The value function is maximum of continuous and increasing functions in q
and M . Since, Max function is continuous and increasing the value function is a
continuous and increasing function of q and M .
15

Also, the value function of the problem with N periods left is larger than the
value function of the problem with N 1 periods left to go.
Lemma 4 (Monotonicity of the finite horizon problem)
The value function V N (.) of the N period to go problem is non decreasing in the
number of periods to go N . i.e.
V N (q) V N 1 (q)
Proof. For the N periods left to go, it is always possible to adopt the strategy optimal
for the problem with N 1 periods left and not to sell and get zero in the last period.
Thus, the value function of the N period problem is non-decreasing in the number
of periods left.
It is now shown that the value function of the finite-horizon problem is concave
in M .
Lemma 5 (Concavity of the Value function in M ,V N (q, M ) M V N (q, 1))
Proof. Proof by induction. Let N=1
V 1 (q, M ) = M ax{0; M [q (1 )]; M [2f0 (q) 1]}
V N (q, 1) = M ax{0, [q (1 )]; [2f0 (q) 1]}N

V 1 (q, M ) is linear in M as it is a max of two linear functions in M and 0. Therefore,


V 1 (q, M ) = M V 1 (q, M ) and (EV 1 (q, M ) M EV 1 (q, 1)) Now suppose, V N 1 (q, M )
M V N 1 (q, 1)
V N (q, M ) = M ax{0, n[q(1)]+EV N 1 (q, M n)n M, M [q(1)], M [2f0 (q)1]}
V N (q, 1) = M ax{0, [q (1 )]; [2f0 (q) 1]}N

It is obvious to see that V N (q, M ) M V N (q, 1)q, N, M

In the rest of the sub-section the Bellman equation is analyzed in order to establish that the optimal solution is a simple cutoff policy.
For q 1 the comparison is between the first and the second argument.
V N (q, M ) = M ax{0 + V N 1 (q, M ); n[q (1 )] + EV N 1 [q, M n]]}
16

This is because q 1 , M [q (1 )], M [2f0 (q) 1] < 0.


q [1 , q] the maximum is achieved by either the second or the third argument.
V N (q, M ) = M ax{n[q (1 )] + EV N 1 [q, M n]; M [q (1 )]}
This is due to the fact that q 1 , M [q (1 )] 0 and M [2f0 (q) 1] < 0.
On the other hand, q [
q , 1] the value function can be re-written as
V N (q, M ) = M ax{n[q (1 )] + EV N 1 [q, M n]; M [2f0 (q) 1]}
This can be inferred from Lemma 1.
Consider the optimal policy q 1 . The optimal policy of the monopolist with
N periods left to go is such that the monopolist does not sell at all to the current
consumer whenever the belief is lower than the cutoff level q1N (M ), and sells for
higher beliefs to only one region q [q1N (M ), 1 ]. The firm once exists never
reverses the decision to stop selling. This is because {q1N (M )} is decreasing in the
number of periods, N left to go. In the two period problem, for q < q12 (M ) it is
optimal for the firm not to sell in both the current and the next period rather than
quoting the separating price, PS (q) that sells at loss to the consumer in the current
period. The general result on the optimality and monotonicity of the cutoff policy
for exit is stated in the next result.

0
x Exit y

1
q1N (M )
x Seq,n=1 y
x

Seq,n 1 y

q
1
q2N (M )
x Sim. PS y x Sim. PP y

Proposition 2(a) (Existence and Monotonicity of cutoff levels for exit


in N)
In the problem with N periods to go and M regions left to launch the product in,
it is optimal to exit the market when the belief is below the cutoff level q1N (M ). In
addition, the cutoff levels q1N (M ) is non-increasing in the number of periods to go N ,
i.e. q1N (M ) q1N 1 (M ) and non-increasing in M , i.e. q1N (M ) q1N (M 1)
17

Proof. By induction.
Step 1
We know from Proposition 1 that q12 (M, n)n, M such that q q12 (M, n) the
monopolist prefers to exit the market in the second period when there are M regions
left to launch the product than to launch the product in n < M regions. We
show that if q1N (M, n)n such that q q1N (M, n) the monopolist prefers to exit
the market than sell sequentially in n markets then q1N +1 (M, n)n, M such that
q q1N +1 (M, n) the monopolist prefers to exit the market than sell in n < M
markets.
Step 2
Suppose, it is optimal to follow a cutoff policy of exit for N period q q1N (M, n).
i.e. q1N (M, n) satisfies the following equation
n[q1N (M, n) (1 )] + EV N 1 [q1N (M, n), M n] = 0
We know from Lemma 4, EV N [q, M ] EV N 1 [q, M ]. Therefore,
n[q1N (M, n) (1 )] + EV N [q1N (M, n), M n] 0
Step 4 Existence and Monotonicity of q1N (M ) in N
Define, LN +1 (q, n) = n[q (1)]+EV N [q, M n]. Since, q1N +1 (M n, n0 ) > 0n0 ,
LN +1 (q = 0, n) < 0 and LN +1 (q = q1N (M, n), n) 0. Since, LN +1 (q, n) is continuous and strictly increasing in q. q1N +1 (M, n) [0, q1N (M, n)] such that LN +1 (q =
q1N +1 (M, n), n) = 0. Also, q1N +1 (M, n) q1N (M, n)M, n. Define, q1N +1 (M ) =
min[q1N +1 (M, n)]n . Then, q1N +1 (M ) min[q1N (M, n)]n = q1N (M )M 0 M .
Step 5 Monotonicity in M
Suppose, it is optimal to follow a cutoff policy of exit for N period q q1N (M, n).
i.e. q1N (M, n) satisfies the following equation
n[q1N (M, n) (1 )] + EV N 1 [q1N (M, n), M n] = 0
We know from Lemma 3, EV N [q, M 0 ] EV N [q, M ]M 0 M . Therefore,
n[q1N (M, n) (1 )] + EV N 1 [q1N (M, n), M 0 n] 0
Define, LN (q, n) = n[q(1)]+EV N 1 [q, M 0 n]. Since, q1N 1 (M 0 n, n0 ) > 0n0 ,
LN (q = 0, n) < 0 and LN (q = q1N (M, n), n) 0. Since, LN (q, n) is continuous and
strictly increasing q1N (M 0 , n) [0, q1N (M, n)] such that LN (q = q1N (M 0 , n), n) = 0.
Also, q1N (M 0 , n) q1N (M, n)n. q1N (M 0 ) = min[q1N (M 0 , n)]n min[q1N (M, n)]n =
q1N (M ).
18

The intuition for the existence of q1N (M ) is similar to that given before. We also
find that q1N (M ) increases as N increases. Thus, as the horizon of the game increases
the firm is more likely to exit the market.
Now consider the optimal product launch strategy for q [1 , q]. The decision to
continue learning involves a current cost and a future benefit. To keep the learning
process going, it is necessary to price high, even though it would be myopically optimal to sell to all the consumers at the pooling price. The cost of learning is the loss in
current expected profit from inducing a high price and is equal to (M n)(q(1)).
However, the benefit in the future from higher profits is EV N 1 [q, M n]. In particular, it is optimal to stop the learning process q [q2N (M ), 1] for the problem
with N periods to go and M regions left to launch the product. The sequence of the
cutoff levels {q2N (M )} are increasing in the number of periods to go N . This derives
from the fact that experimentation has a larger value with a larger horizon N ahead.
Proposition 2(b)(Optimality and Monotonicity of cutoff levels for the region to launch simultaneously)
In the problem with N periods to go and M regions left to launch the product in, it is
optimal to launch the product simultaneously q [q2N (M ), 1]. In addition, the cutoff
levels q2N (M ) are increasing in the number of periods to go N , i.e. q2N (M ) q2N 1 (M )
and q2N (M ) q2N (M 0 )M M 0 . This implies that the region where the firm launches
its product simultaneously is decreasing in the number of periods to go N .
Proof. Appendix A
The firm gets a lower current expected payoff by launching the product sequentially. However, it gets a chance of selling at a higher price in future in the event of
a sale in the current period. The advantage of launching the product sequentially
decreases as q increases. If q is low, it is beneficial to give up some current expected
payoff in return for a possibility of a sale at a higher q in the future. But as q
increases the advantage of a higher future expected returns, decreases. Thus, there
exists a cutoff level q2n (M ) such that q q2N (M ) the firm would launch the product
simultaneously. The range where the firm launches the product simultaneously decreases with the increase in the horizon of the game. Also, this range increases with
the number of regions where the product has not been launched. This is because it
pays to encourage learning when there is a larger time horizon and greater number
of regions left.

19

Corollary : (Optimality and Monotonicity of cutoff levels for the region


to launch sequentially)
In the problem with N periods to go and M regions left to launch the product in,
it is optimal to launch the product sequentially, q [q1N (M ), q2N (M )]. In addition, the cutoff levels q2N (M ) are increasing in the number of periods to go N , i.e.
q2N (M ) q2N 1 (M ). This implies that the region where the firm launches its product
sequentially is increasing in the number of periods to go N .
Proof. From Proposition 2 we know that the monopolist would launch the product
sequentially q [q1N (M ), q2N (M )]. In this proof we show that the firm would launch
in only one region q [q1N (M ), 1 ].
q [q1N (M ), 1 ] launching the product in n regions gives an expected return of
nm(q(1))+EV N 1 [q, M n]. The difference between the expected utility from
launching the product in n + 1 and n regions is the following. Since, q, Ln+1 Ln =
k(q (1 )) + EV N 1 (q, M n 1) EV N 1 (q, M n)q [q1N (M ), 1 ],
q (1 ) 0 the monopolist is willing to introduce the product at a discount in
the current period so as to encourage learning in the next period. Also, a launch
in an additional period would imply greater losses in the current period and fewer
regions in the future to launch the product at a positive price.
Let Gl (q N +1 ) be the cumulative density distribution of the next period common
prior, q N +1 when l number of trials are allowed in the N th period. It can be shown
N +1
that Gl (q N +1 ) Gl+2 (q N +1 )q N +1 q N . Thus, gP
) gl+2 (q N +1 )q N +1 q.
l (q
Thus, EV N 1 (q, M n1)EV N 1 (q, M n) = qN +1 q [gn+1 (q N +1 )V N 1 (q t+1 , M
(n 1)k)] [gn (q N +1 )V N 1 (q N +1 , M (n)k)] 0. Hence, Ln+1 Ln 0n. Thus
the firm would launch the product in only one region for q 1 However, q
[1, q2N (M )] there exists a unique maximum number,H (q, M ) of launches that the
firm would initiate. H (q, M ) is the n maximizes kn(q(1))+EV (q, M n).
q [q1 , 1 ] the firm launches the product only in one region. This is because
launching the product sequentially q [q1N (M ), 1 ] implies selling the product
at a discounted price. Increasing the number of launches increases this loss. Also,
launching the product in greater number of regions, now implies that the product
would be launched in fewer number of regions in the future. However, the number of
launches q [1 , q2N (M )] is n (M ) 1. Thus, for some regions of q the number
of launches could be greater than 1. In the next proposition we show that the prices
decrease and number of launches increase on an average with time.
Proposition 3 (Prices Supermartingale, Launches Submartingale)
20

Prices, a function of the past beliefs are supermartingale. i.e. They decrease on
an average in time. Also, number of launches, a function of last period belief are
submartingale. i.e. They increase on an average in time.
Proof. The game moves to the next period,t + 1 only when the product is launched
sequentially in period t. Thus, the relevant price at time period
t, P t = 2f1 (q) 1.
P
There are three possible prices in the next period. P1t+1 = j P r(s = g)j P r(s =
b)nj (2fg (
q ) 1)C(n, j) where,Pn is the number of launches in period t and q is
the next period
belief; P2t+1 =
j KP r(s = g)j P r(s = b)nj (2fP
q ) 1)C(n, j)
g (
P
t+1
j
nj
and P3 =
j HP r(s = g) P r(s = b) (2fg (
q ) 1)C(n, j) + j HP r(s =
j
nj
g) P r(s P
= b) (2fb (
q ) 1)C(n, j). It can be seen that P2t+1 , P3t+1 P1t+1 . Consider,
P1t+1 = j P r(s = g)j P r(s = b)nj (2fg (
q ) 1)C(n, j). Since, fg is concave in q,
P
P
t+1
j
nj
P1 2fg [ j P r(s = g) P r(s0 ) q]1 2fg [ j j (1)nj q]1 2fg (q)1 =
P t . Since, P2t+1 , P3t+1 P1t+1 P t , prices are supermartingale.

3.3

Infinite Horizon

As it is clear from the analysis of the previous section, this model of monopoly
predicts that the monopolist would follow a cutoff policy for its product and pricing
strategy. For low values of the belief about the quality of the product the firm would
exit the market. However, it would launch the product sequentially for intermediary
levels of q and launch simultaneously for large values of q. In this section the infinite
horizon problem is solved. We solve for the stationary solution of the problem for N
very large. The value function of the infinite horizon problem of the monopolist can
be written as
V (q, M ) = max{0, n(q (1 )) + EV (q, M ), M (2f0 (q) 1), M (q (1 ))}
The value function of the infinite-horizon problem is continuous and, when strictly
positive, strictly increasing in q. The first argument corresponds to exiting the
market, the second argument to the separating price strategy that keeps the learning
process on, and the third to simultaneous launch strategy that allows the monopolist
to stop the learning process of the consumers.
Note that for q small enough it is optimal to post the separating price(even though
it is negative), and sell with positive probability at a current loss.
Proposition 3(a) (Exit threshold)
For belief lower than q1 (M ) (0, 1 ) the monopolist exists the market. q1 (M )
is non-increasing in M .
21

Proof. Let q1 (M, n) be such that n(q1 (M, n) (1 )) + EV (q1 (M, n), M ) = 0.
Such a value uniquely exists and is in the interval (0, 1 ), because Ln (q) = n(q
(1 )) + EV (q, M ) is a continuous, strictly increasing function of q, Ln (q = 0) 0
and Ln (q = 1 ) > 0. Therefore, q q1 (M, n) the firm would prefer to launch the
product sequentially in n regions than to exit. Let q1 (M ) = min[q1 (M, n)]. q1 (M )
exists because there are a finite number of q1 (M, n).
q1 (M, n) be such that n(q1 (M, n) (1 )) + EV (q1 (M, n), M ) = 0. By Lemma 3,
EV (q, M ) EV (q, M 0 )M M 0 . Hence, n(q1 (M, n)(1))+EV (q1 (M, n), M 0 )
0. Since, n(q (1 )) + EV (q, M 0 ) is a continuous, strictly increasing function
of q, Ln (q = 0) 0 and Ln (q = 1 ) > 0, there exists q1 (M 0 , n) [0, 1 ]
such that it is optimal to launch the product in n regions q q1 (M 0 , n). Moreover,
q1 (M 0 n) q1 (M, n)n. Since, q1 (M, n) is increasing in n, q1 (M 0 ) = min[q1 (M 0 , n)]
min[q1 (M, n)] = q1 (M ).
Learning will be eventually stopped if the belief q is between (q2 (M ), 1)
Proposition 3(b)(Simultaneous Launch)
For belief between (q2 (M ), 1) the monopolist would launch its product simultaneously. q2 (M ) are non-increasing in M .
Proof. Appendix B
The intuition of the threshold policy of the monopolist for the infinite horizon
problem is similar to that of the finite horizon problem.

Welfare Comparisons

In this section we discuss how the monopolists optimal product launch strategy is
different from that of a social welfare maximizer. Consider the problem of a social
planner that has to use the price system and the number of experiments to achieve
optimal information disclosure and maximize welfare. The value function faced by
the social planner is the following
W N (q, M ) = M ax{0, n[q (1 )] + EW N 1 [q, M n], M [q (1 )], M [2q 1]}
Lemma 6
The value function of the social planner is greater than that of the monopolist,
W N (q, M ) V N (q, M )

22

Proof. Consider, W 1 (q, M ) = M ax{0, M [q (1 )], M [2q 1]}. V 1 (q, M ) =


M ax{0, M [q(1)], M [2f0 (q)1]}. It is obvious to see that W 1 (q, M ) V 1 (q, M ).
Therefore, EW 1 (q, M ) EV 1 (q, M ).
Suppose, W N (q, M ) V N (q, M )M, q. W N +1 (q, M ) = M ax{0, M [q(1)], M [2q
1], n[q (1 )] + EW N [q, M n]} and V N +1 (q, M ) = M ax{0, M [q (1
)], M [2f0 (q)1], n[q(1)]+EV N [q, M n]}. Since, W N (q, M ) V N (q, M )M, q
and q f0 (q) we can conclude that W N +1 (q, M ) V N +1 (q, M )M, q
The social planner maximizes the problem of both the firm and that of the consumer. Therefore, the value function of the social planner is atleast as much as that
of the monopolist. The optimal policy of the social planner for the last period is

if q 1
Exit
PS (q) = (2f1 (q) 1) q [1 , ]
=

PP (q) = (2f0 (q) 1) q


Lemma 7
The social planner does not sell q q1N (M ) [0, 1 ], sells simultaneously in all
q1N (M ), q2N (M )].
regions q [
q2N , 1]. Also, the social planner sells sequentially q [

0
x Exit y

1
q1N (M )
x Seq,n=1 y
x

Seq,n 1 y

1
q2N (M )
x Sim. PS y x Sim. PP y

Proof. The proof of Lemma 7 follows from that of Proposition 3 with the W (q, M )
defined above as the value function.
The optimal product launch and pricing strategy of the social planner is similar
to that of a monopolist. Only the thresholds of the two policies vary. The following
proposition explains how the two policies are different. It also reflects that the monopolist discourages learning more than the social planner would optimally like to.
Proposition 4
23

The monopolist exists the market for a belief larger (q1 (M )) than the socially optimal
(
q1 (M )). Also, q2 (M ) q2 . Thus, the social planner encourages learning more than
the monopolist.
Proof. q 1 , W N (q, M ) = M ax{0, n[q (1 )] + EW N 1 [q, M n]},
V N (q, M ) = M ax{0, n[q (1 )] + EV N 1 [q, M n]}. From Lemma 7 we know
q1 (M ), q1 (M ) exist. W (q, M ), V (q, M ) are increasing in q and from Lemma 6 we
know that W N 1 (q, M ) V N 1 (q, M )q, M (hence, EW N 1 (q, M ) EV N 1 (q, M )).
Thus, q1 (M ) q1 (M ). Hence, the monopolist exists the market for a belief larger
than the socially optimal.
q2 (M, n) is defined such that
n[q2 (M, n) (1 )] + EV (q2 (M, n), M n) = M [q2 (M, n) (1 )]
From Lemma 6 we know that EW (q, M n) EV (q, M n). Thus,
n[q2 (M, n) (1 )] + EW (q2 (M, n), M n) M [q2 (M, n) (1 )]
q2 (M, n) is defined such that
n[
q2 (M, n) (1 )] + EW (
q2 (M, n), M n) = M [
q2 (M, n) (1 )]
For 0.5, n[q (1)]+EW (q, M n)M [q (1)] is decreasing in q. Thus,
q2 (M, n) q2 (M, n). Thus, q2 (M ) = M ax(q2 (M, n)) q2 (M ) = M ax(
q2 (M, n)).
The social planner encourages learning more than the monopolist. The monopolist exits the market earlier than the social planner. Also, the range for which the
social planner encourages a sequential launch is larger for a social planner than that
of a monopolist.

5
5.1

Extensions
Heterogeneity in population

In this section we solve the model where the regions have different population sizes.
Each region i has mi 0 number of consumers. L denotes an ordering of the
regions in the decreasing order of their population. Therefore, L(1) is a region where
mL(1) = M ax{mi }iN and mL(N ) = M in{mi }iN . At the beginning of each period, t
the monopolist decides on the number of regions, t N t1 t T , the identity
24

of these regions, rt and the price at which to sell them, pt Rt T .


For every t, i such that i M t denotes lit = 1 if the product is available in region
i at time period t and 0 otherwise. The history at any time period t, ht H is a
collection of all the actions taken by the monopolist, k , pk and that of the consumers,
k k t.
Monopolist
At any point in time t the monopolist has two decisions to take. First, it has to
decide the number of regions that it wishes to launch its product and the identity
of these regions. The firms utility does not depend upon the the number of regions
that the product was launched but only upon the total number of trials introduced
in the previous period. Thus, reinterpret M t as the number of consumers that the
product has not been introduced to, till period t. Since, product can be introduced
to each consumer only once M t M t+1 . The firm only decides on ,(ht ), the total
number of consumers that the product was introduced to. Also, it must also decide
the price, p(ht ) at which to launch it. The payoff relevant history for the monopolist
is q t1 , the last period belief about the quality of the product and the number of
consumers to which the product has not been introduced, M t . The strategy of the
monopolist is
(q t1 , M t ) = (q t1 , M t ) p(q t1 , M t ) {g (q g1 , M g ) pg (q g1 , M g )}
g=t+1 (3)
q g1 [0, 1], M t M

The monopolist maximizes the discounted expected payoff. The realized utility of
the monopolist at time period t is (q t1 , M t ) = p(q t1 , M t )c(q t1 , M t )t T ,
where,
c(q t1 , M t ) is the number of people that adopt the product. c(q t1 , M t ) =
P
t
i mi is.t.li = 1
After the reinterpretation of M t as the number of consumers to which the product
has not been introduced, the results for finite horizon and infinite horizon product
launch and pricing problem of the firm remain the same.

5.2

Heterogeneity in the reservation utilities

We only consider the case where the monopolist decides to launch its product in two
regions which differ in terms of their reservation utility. Let the reservation utilities
of the regions be r1 , r2 . Let without loss of generality r1 r2 . Both the regions
have the same population, m. At the beginning of each period, t the monopolist
decides on the number of regions, t N t1 t T and the price at which to sell
them, pt Rt T. For every t, i such that i M t denotes lit = 1 if the product
is available in region i at time period t and 0 otherwise. The history at any time
25

period t, ht H is a collection of all the actions taken by the monopolist, k , pk and


that of the consumers, k k t.
The monopolist can charge one of the four prices, separating price for region 1,
PS1 = 2f1 (q)1r1 ; separating price for region 2, PS2 = 2f1 (q)1r2 ; Pooling price
for region 1, PP 1 = 2f0 (q) 1 r1 ; Pooling price for region 2, PP 2 = 2f0 (q) 1 r2 .
The respective profits derived from these prices are, S1 = PS1 P r(s1 |q)(2m); S2 =
PS2 P r(s1 |q)(m); P 1 = PP 1 (2m); P 2 = PP 2 (m). The following lemma describes
the pricing strategy of the firm in the last period or when it launches the product
simultaneously.
Lemma 8
The pricing decision of a firm in a simultaneous launch is the following:

0
x

Exit

qs2
y x Screening in Reg 1 y

qs1s2
qp1p2
1
x Pool in Reg 2 y x Pool in Reg 1 y

Proof. The monopolist has a choice to launch the product at either of the prices,
PS1 , PS2 , PP 1 , PP 2 .
Note that PS2 PS1 , PS1 PP 1 , PS2 PP 2 . Denote, qS1 by the q such that
S1 = 0. Similarly, qs2 . It can be noted that qS1 qS2
Lemma 9
If the firm launches the product sequentially then it does so only with a separating
price
Proof. First we will prove that the monopolist would not launch its product at any
of the pooling prices. That is the optimal price is not PP 1 , PP 2 .
Let us suppose that the firm launches the product sequentially in only region 1
at price, PP 1 . Then the profits received would be
m[2f0 (q) 1 r1 ] + V 1 (q)

where, V 1 (q) = M ax{S1 , S2 , P 1 , P 2 }. The firm can do better by launching the


product together in both the regions at the price which maximizes V 1 (q). Thus, it
26

is never optimal for the firm to launch the product at the price, PP 1 . Similarly, for
the price PP 2 .

5.3

Private Information

This section analyzes the one sided problem where the monopolist is aware about the
quality of its product. Let us assume that the monopolists product can either be
of Good(G) or Bad(B) type. We only focus on the pure strategy of the monopolist.
If the utility from consuming the bad product is less than the reservation utility of
all agents, then all pure strategies equilibrium would be pooling equilibrium. That
is the bad type monopolist would never find it in its advantage to distinguish itself
from the good type and hence always adopt the same strategy as that of the high
type.
The launch strategy of the high type monopolist is also a threshold policy. The
proposition below characterizes the optimal launch strategy for the high type monopolist and shows that the firms allow for greater experimentation when it is aware
of its product quality than if it is not
Proposition 4:
The amount of experimentation allowed by the firm is greater when it is aware of the
product quality than when it is not.
Proof. Appendix C
The firm that knows that its product is of high quality is willing to take a greater
risk by encouraging greater learning. The firm that is unaware of the quality of its
product assigns a higher probability to a bad herd than the firm that is aware that
its product is good. This knowledge of the quality encourages the firm to take bolder
steps of introducing the product sequentially so as to obtain a higher expected future
payoff.

Conclusion

Prices and the number of previous sales of a product at any point in time reveal
information about the unknown quality of the product. Social learning literature so
far, has focused on the scenario where economic agents do not influence learning.
This paper sheds light on how a monopolist can influence the degree of learning
among consumers by strategically choosing the prices and the number of sales of its
products at every point in time.

27

We study a model where consumers and the monopolist are unaware of the quality
of the product and learn about it through the prices and number of sales. When
the belief of the product is extreme the firm either exits the market or launches the
product sequentially. However, for intermediate levels of the belief of the product the
firm launches the product sequentially in different regions. We find that for games
with greater time horizon the firm is more likely to introduce the product sequentially. Also, the price sequence is found to be supermartingale, i.e. the prices on an
average decrease with time. However, the number of sales or trials are found to be
submartingale, i.e. the number of launches increases on an average with time. We
also find that the amount of learning introduced by the monopolist is not efficient.

Appendix
A

Proposition 2(b)

Proof. We will first prove the existence of q2N (M )


(a)Existence of q2N (M )
Step 1: Proof by induction. We know from Proposition 1 that there exists
q22 (M, n)n < M such that
n[q22 (M, n) (1 )] + EV 1 [q22 (M, n), M n] = M [q22 (M, n) (1 )]n < M
Step 2:
q [1 , q], N, M
V N (q, M ) = M ax{n[q (1 )] + EV N 1 [q, M n], M [q (1 )]}
Suppose, q2N (M, n) such that n[q2N (M, n) (1 )] + EV N 1 [q2N (M, n), M n] =
M [q2N (M, n)(1)]. We know from Lemma 3 that EV N [q, M n] EV N 1 [q, M
n]q. Therefore,
n[q2N (M, n) (1 )] + EV N [q2N (M, n), M n] M [q2N (M, n) (1 )]
Step 3: RN +1 (q, n) is strictly decreasing in q
In this step we define RN +1 (q, n) = n[q (1 )] + EV N [q, M n] M [q (1 )]
28

and show that it is continuous and strictly decreasing in q.


Step 3(a):RN +1 (q, n) is a composite function of continuous function and hence is
continuous.
Step 3(b): To prove that RN +1 (q, n) is strictly decreasing,
EV N [q, M n]
RN +1
= (M n) +
q
q
N +1

We will show by induction that R q (q,n) 0.


We know, V 1 (q, M n) = M ax{0, (M n)[q (1 )], (M n)[2f0 (q) 1]}q
1
2
n]
0 (q)
[1 , q]. Since, 2 fq
1q q , EV [q,M
M n and hence, R q(q,n) < 0.
q
N 1

Now suppose, EV q[q,M n] M nM, n < M .


V N (q, M n) = M ax{0, n[q (1 )] + EV N 1 (q, M 2n), (M n)(q (1
)), (M n)(2f0 (q) 1)}.
N [q,M n]
=
If V N (q, M n) = n[q (1 )] + EV N 1 (q, M 2n) then EV q
[n + EV

N 1 (q,M 2n)

] n + M 2n. The last inequality follows from 1

q
EV N 1 [q,M n]
M n, M, n < M .
and
q
N [q,M n]
f0 (q)
Since, 2 q 1, q q, EV q
M

Step 5: RN +1 (q = q, n) 0, RN +1 (q = q2N (M, n), n) 0. Therefore from Step 2


and 3, q2N +1 (M, n) [q2N (M, n), q]n, M such that RN +1 (q = q2N +1 (M, n), n) = 0 or
n[q2N +1 (M, n)(1)]+EV N [q2N +1 (M, n), M n] = M [q2N +1 (M, n)(1)]n < M
Let, q2N +1 (M ) = M ax[q2N +1 (M, n)]n . Hence, q2N +1 (M ) q2N (M )M, N .
Monotonicity in M

Proposition 3(b)(Simultaneous Launch)

Proof. We will begin by first proving the existence and the monotonicity of q2 (M )
then that of q3 (M ).
(a) q2 (M )
q [1 , q] the value function can be written as
V (q, M ) = M ax{n[q (1 ) + (M n)EV (q, 1), M (q (1 ))}
29

Now, Rn (q) = n[q (1 )] + EV [q, M n] M [q (1 )] is continuous and


strictly decreasing in q, Rn (q = q) 0 and Rn (q = 1 ) > 0. Therefore, there
exists q2 (M, n) such that
n[q2 (M, n) (1 )] + EV [q2 (M, n), M n] M [q2 (M, n) (1 )] = 0
Therefore, q q2 (M, n) the firm would prefer to launch the product simultaneously
in all regions than to launch it sequentially in n regions. Let q2 (M ) = max[q2 (M, n)].
q2 (M ) exists because there are a finite number of q2 (M, n).
Now we would prove the monotonicity of q2 (M ). Let q2 (M, n) be such that
n[q2 (M, n) (1 )] + EV [q2 (M, n), M n] M [q2 (M, n) (1 )] = 0
It can be shown that (M M 0 )[q(1)] EV 1 [q, M n]EV 1 [q, M 0 n]M 0
M . Thus, 0 = n[q22 (M, n)(1)]+EV [q2 (M, n), M n]M [q2 (M, n)(1)]
n[q2 (M, n) (1 )] + EV [q2 (M, n), M 0 n] M 0 [q2 (M, n) (1 )]. Since,
n[q (1 )] + EV [q, M n] M [q (1 )] is decreasing in q. q2

Proposition 4:

Proof. The product launch strategy of the monopolist when it is aware of the quality
of the product is the following:
The pricing decision of the firm in a simultaneous launch is the following or the last
stage for a finite horizon problem:

0
x

Exit

xy

0.5
Screening

qP

1
xy

Pooling

In the problem with N periods to go and M regions left to launch the product in, it
N
is optimal to exit the market when the belief is below the cutoff level q1P
(M ). It is
N
optimal to launch the product simultaneously q [q2P (M ), 1]. In addition, the cutoff
N
N
levels q1P
(M ), q2P
(M ) is non-increasing and increasing, respectively in the number of
N 1
N 1
N
N
periods to go N , i.e. q1P
(M ) q1P
(M ), q2P
(M ) q2P
(M ) and non-increasing in
N
N
M , i.e. q1P
(M ) q1P
(M 1). This implies that the region where the firm launches
its product simultaneously is decreasing in the number of periods to go N .
30

The proof of the statements above are very similar to proposition 2(a)(b). In this
N
N
proof we would show that the thresholds q1P
M q1N (M ) and q2P
M q2N (M ).
N
N
N
First, q1P M q1 (M ). q1P (M, n) is defined such that
N
N
(M, n), M n] = 0
(M, n)) 1] + EV N 1 [q1P
n[2fg (q1P

Similarly, q1N (M, n) is defined such that


n(q1N (M, n)+(1)(1q1N (M, n)))[2fg (q1N (M, n))1]+EV N 1 [q1N (M, n), M n] = 0
Since, q1N (M, n) + (1 )(1 q1N (M, n)) and the function n(q (1 )) +
N
(M, n). Hence,
EV N 1 (q, M n) is increasing in q. We have that q1N (M, n) q1P
N
N
q1 (M ) q1P (M ).
N
(M, n) is defined such that
Similarly, q2P
N
N
(n M )[2fg (q2P
(M, n)) 1] + EV N 1 [q2P
(M, n), M n] = 0

And, q2N (M, n) is defined such that


(n M )[2fg (q2N (M, n)) 1] + EV N 1 [q2N (M, n), M n] = 0
Since, q2N (M, n) + (1 )(1 q2N (M, n)) and the function n(q (1 )) +
N
(M, n). Hence,
EV N 1 (q, M n) is increasing in q. We have that q2N (M, n) q2P
N
N
q2 (M ) q2P (M ).

References
[1] Ayal, Igal and Jehiel Zif, Market expansion strategies in multinational marketing. Journal of Marketing, Vol. 43(1979): 84-94
[2] Arthur, W.B. Competing Technologies, Increasing Returns and Lock-In by
Historical Events.The Economic Journal,Vol.99, No.394(March 1989):116-131.
[3] Banerjee, Abhijit.V. A simple model of herd behaviorThe Quarterly Joural
of Economics, Vol. 107, No. 3(Aug 1992): 797-817
[4] Bikhchandani, S, Hirshleifer, D. and Welch, I. A Theory of Fads, Fashion,Customs and Cultural Chnage as Informational Cascades. The Journal of
Political Economy, Vol.100, No. 5.(Oct 1992):9921026.

31

[5] Davidson, William H. and Harrigan, Richard, Key decisions in international


marketing. Introducing new products abroad. Columbia Journal of World
Business (Winter), 15-23.
[6] Ohmae, Kenichi, 1989. Managing in a borderless world. Harvard Business
Review 67 (May-June), 152-161.
[7] Riesenbeck, Hajo and Freeling, Anthon, 1991. How global are global brands?
The McKinsey Quarterly 4, 3-18.
[8] Ceminal, Ramon and Vives, Xavier, Why Market Shares Matter? RAND
Journal of Economics, Vol. 27, No. 2:221-239, 1996.
[9] Bar-Issac, Heski. Reputation and Survival: Learning in a Dynamic Signalling
Model Review of Economics Studies, Vol. 70, No.2 (2003):231-251
[10] Bergmann, D and Valimaki, J. Market diffusion with two-sided learning.
RAND Journal of Economics, Vol. 28(No.4):773-796, 1995.
[11] Bergmann, D and Valimaki, J. Experimentation in Markets. Review of Economic Studies, Vol. 67(No.2): 213-234, 2000.
[12] Sgroi, D. Optimizing Information in the Herd: Guinea Pigs, Profits and Welfare. Games and Economic Behavior, Vol. 39: 137-166, 2002
[13] Welch, Ivo, Sequential Slaes, Learning and Cascades, Journal of Finance, Vol.
47(2):695-732, 1992.

32

Das könnte Ihnen auch gefallen