NET Institute*

www.NETinst.org

Working Paper #04-02

October 2004

Adware, Shareware, and Consumer Privacy

Nataly Gantman and Yossi Spiegel

Tel Aviv University

* The Networks, Electronic Commerce, and Telecommunications (“NET”) Institute,

http://www.NETinst.org, is a non-profit institution devoted to research on network
industries, electronic commerce, telecommunications, the Internet, “virtual networks”
comprised of computers that share the same technical standard or operating system, and
on network issues in general.
 Adware, Shareware, and Consumer Privacy¤

Nataly Gantmanyand Yossi Spiegelz

September 30, 2004

Abstract

Programmers can distribute new software to online users either for a fee as shareware
or bundle it with advertising banners and distribute it for free as adware. In this paper we
study the programmers’ choice between these two modes of distribution in the context of
a model that take explicit account of the strategic interaction between programmers who
develop software, …rms that advertise their products through ad banners, and consumers
who buy software and consumer products. Adware allows advertisers to send targeted in-
formation to speci…c consumers and may therefore improve their purchasing decisions. At
the same time, adware also raises privacy concerns. We study the e¤ect of programmers’
choice between shareware and adware on consumers’ welfare through its e¤ect on the ben-
e…cial information that consumers receive about consumers products on the one hand and
their loss of privacy on the other hand. We also examine the implications of improvements
in the technology of ad banners and the desirability of bans on the use of adware.
JEL Classi…cation: L12, L13, M37
Keywords: adware, shareware, advertising, privacy, ad banners

¤
The …nancial support of the NET insititute (http://www.NETinst.org) is gratefully acknowledged.
y
Berglas School of Economics, Tel Aviv University, Ramat Aviv, Tel Aviv, 69978, Israel. Email:
<[email protected]>
z
Recanati Graduate School of Business Adminstration, Tel Aviv University, email: [email protected],
http://www.tau.ac.il/~spiegel

1
1 Introduction
In recent years, an increasing number of software providers began to distribute their software
online by allowing consumers to download the software from the provider’s website. Until the
end of the 1990’s online software was mainly distributed as shareware: consumers are free to
download the software for a trial period but must pay for using the software once the trial period
expires. By the end of the 1990’s some software provides began to distribute their software as
adware which is distributed for free but is supported by advertisers who pay the software provider
for displaying ad banners or pop-up ads while the software is being used.1 In this paper we study
the choice of programmers between these two modes of distribution in the context of a model
that takes explicit account of the strategic interaction between programmers, …rms that sell
consumer products and may advertise their products through ad banners, and consumers who
buy software and products.
A main premise of our analysis is that the targeted information about consumer products
sent to speci…c adware users through as banners is potentially bene…cial as it may improve their
purchasing decisions. This premise is based on the observation that …rms are willing to pay
programmers considerable amounts for displaying ad banners on the desktops of adware users.
A case in point is the Claria Corporation which is one of the pioneers in online behavioral
marketing with over 900 advertisers, including 85 of the Fortune 1,000 and an annual revenue
of 90.5 million in 2003 and a pro…t of $35 million. In our model, consumers di¤er in their
preferences over products but may not know at the outset which …rm sells which product. Ad
banners allow …rms to send targeted ads to consumers that inform them about products that
match their tastes. Consequently, …rms are willing to pay for ad banners as those raise the
likelihood that consumers will purchase their products.
While adware allows advertisers to send potentially useful targeted information to speci…c
consumers, it also raises privacy concerns. Recent surveys show that many consumers are
concerned about the misuse of their private information collected online and refrain from buying

1
There are several de…nitions of adware. In this paper we only consider “legitimate” ad-supported software
which is installed with the end-user consent. We do not consider “spyware” which is often installed without
the end-user consent and tracks and collects personal information without consent (see e.g., Urbach and Kibel
(2004)). For a discussion on the histories of shareware and adware, see for example, Knopf (2000) and Stern
(2004).

2
online due to uncertainty regarding the usage of personal data (see e.g., IBM Global Services,
1999).2 De…nitions of privacy vary widely according to context and environment. Posner (1981)
discusses several possible de…nitions, including the “concealment of information,” “peace and
quite,” and “freedom and autonomy.” In this paper we consider the second de…nition, namely
privacy as the right for “peace and quite.”3 This right is a main reason behind the “do-not-call
list” that is enforced in the US by the FTC and FCC, and is intended to prevent telemarketers
from violating consumers’ privacy at home.4 We capture the desire of consumers for “peace and
quite” by assuming that, in addition to potentially useful information about consumer products,
adware users also get a disutility from targeted ads and that this disutility increases with the
number of ad banners that are displayed on their desktops.
We begin the paper by considering a model in which a single programmer has developed
a new software and needs to decide whether to distribute online as an adware or as a shareware.
Under the …rst option, consumers can download the software for free and derive utility from its
usage, but in addition, they face a trade o¤ between the bene…cial information about consumer
products conveyed by ad banners against the disutility from privacy loss. In equilibrium, con-
sumers with large privacy concerns will not adopt the adware while those with relatively small
privacy concerns will adopt it. In turn, the number of adopters determines the willingness of
…rms to pay for ad banners and hence the programmer’s pro…t from adware. If the programmer

2
Interesingly, the IBM study also shows that in the U.S., Germany, and the UK “A minority of consumers ...
say they are interested in receiving marketing material. Yet, in far greater numbers, consumers view personalized
marketing as a good thing and, in practice, almost half of all consumers in each country have purchased something
from a catalog mailed to their home in the past year.” This suggests that there may be a gap between rhetoric
and actual behavior.
3
Our notion of privacy is also consistent with Smith (2000) who de…nes privacy as “the desire by each of
us for physical space where we can be free of interruption, intrusion, embarrassment, or accountability and the
attempt to control the time and manner of disclosures of personal information about ourselves.”
4
In a decision from February, 17, 2004, the U.S. Court of Appeals for the Tenth Circuit held that “the
do-not-call registry” targets speech that invades the privacy of the home, a personal sanctuary that enjoys a
unique status in our constitutional jurisprudence” (Mainstream Marketing Services, Inc., TMG Marketing Inc.,
and American Teleservices Association v. Federal Trade Commission, et al., U.S. Court of Appeals for the 10th
Circuit, No. 03-1429, and consolidated cases). Likwise, in Frisby v. Schultz, 487 U.S. 474, 484 (1988), the
Supreme Court of the U.S. held that “One important aspect of residential privacy is protection of the unwilling
listener. ... [A] special bene…t of the privacy all citizens enjoy within their own walls, which the State may
legislate to protect, is an ability to avoid intrusions. Thus, we have repeatedly held that individuals are not
required to welcome unwanted speech into their own homes and that the government may protect this freedom.”
And, in FCC v. Paci…ca Found., 438 U.S. 726, 748 (1978) the Supreme Court of the U.S. held that “[I]n the
privacy of the home ... the individuals right to be left alone plainly outweighs the First Amendment rights of an
intruder.”

3
chooses to distribute the software as a shareware he sets a price for the software and sells it to
consumers.
We show that in equilibrium, the programmer will distribute his software as an adware
provided that its innate quality is relatively low. When the software’s innate quality increases
above a certain threshold, the programmer will switch to shareware. This pattern is consistent
with the experience of several popular software (e.g., Gozilla and GetRight which are the second
and the third most popular download managers on download.com with over 17 million and
11 million downloads respectively) that were …rst distributed as adware, but then, newer and
improved versions were distributed as shareware.
Given the fact that adware is still in its infancy, it is expected that in the near future,
the technology of context-based advertising will improve. A case in point is Gmail which is
a new free webmail service that Google plans to o¤er. Gmail will place text ads and links to
related web pages adjacent to relevant email messages by quickly analyzing their content and
determining which ads are most relevant to them.5 Such improvements have raised concerns
about the increasing loss of privacy on the Internet.6 In our model such improvements a¤ect
both consumers’ privacy as well as their information on consumers’ products. We show that such
improvements will induce the programmer to o¤er adware for a wider set of parameters. Hence,
consumers with large privacy concerns may be worse o¤ since in order to obtain the software
they would have to receive ad banners which lower their utility. Yet, our analysis shows that in
the aggregate, consumers become better o¤.
Unlike adware, “spyware” (or even “malware”) is often installed without the end-user’s
knowledge and tracks and collects personal information without consent (see e.g., Urbach and
Kibel (2004)). The rapid growth of spyware has become a serious problem. Apart from vio-
lations of privacy, spyware also causes various technical problems including slow performance
of computers, inability to access the Internet, extra icons and pop up ads, Internet or system
freezes, and so on.7 This rapid growth has prompted legislators in the U.S. to consider legislation

5
See http://gmail.google.com/gmail/help/about.html
6
For example, a recent article in MSNBC.com Science and Technology, “Privacy advocates target Google
mail” http://msnbc.msn.com/id/4679359/ argues among other things that Gamil “is spooky enough to rankle
privacy advicates, who say Google is going too far by poking through individual e-mails.”
7
In a workshop on Spyware held at the FTC in April 2004, Bryson Gordon from McAfee Security, argued that
spyware related problems are right now “the single largest issue that we are seeing,” and Maureen Cushman from

4
that would either ban or substantially restrict the use ad-supported software. Utah has already
passed such legislation that, among other things, prohibits any party from installing software
that monitors computer usage, uses context-based triggering mechanisms, and also prohibits
the use of context-based pop-ups that obscure the underlying content. Similar legislation are
currently pending in California, the U.S. Senate (Spy Block Act, S.2145) and House of Repre-
sentatives (Safeguard Against Privacy Invasions Act H.R. 2929), and several other states. Our
paper shows that imposing such bans on legitimate ad-supported software may harm consumers
by preventing them from receiving useful targeted ads.
We also consider competition in the software market between two programmers: one who
provides adware and another who provides shareware. Here we consider the equilibrium in the
software market when the adware provider sets the price of ad banners that …rms pay when
they display ad banners and the shareware provider sets the price of the shareware. We show
that our main conclusions from the single programmer case continue to hold even when there is
competition in the software market.
Our paper contributes among other things to the small but growing literature on the eco-
nomics of privacy. Several papers in this literature have identi…ed the loss of privacy with the
disclosure of information on the consumers’ preferences (Acquisti and Varian, 2004; Calzolari
and Pavan, 2004; Dodds 2003; Taylor 2002, 2004; and Wathieu, 2002). Such information allows
…rms to use personalized prices that hurt consumers by extracting their consumers’ surplus. But
as Varian (1996) points out, when …rms learn information about consumers’ preferences, they
can also o¤er them products that better meet their needs and thereby lower their search costs.
Hence, disclosure of information on consumers’ preferences involves a tradeo¤ between a reduc-
tion of search costs and extraction of consumers’ surplus. A di¤erent approach to consumers’
loss of privacy is taken by Hann et al (2003), who consider a model in which some consumers
are privacy guardians and are not interested in buying products. They then consider a game
in which …rms invest resources in identifying potential consumers (though cannot tell whether
they are privacy guardians or not) while privacy guardians invest resources in trying to avoid

Dell argued that spyware become ”a huge technical support issue for us,” and that “Spyware related technical
support calls have been as high as 12 percent of all technical support requests to the Dell technical support
queue.” See http://www.ftc.gov/bcp/workshops/spyware/transcript.pdf

5
solicitations by …rms. They show that competition between …rms raises both types of invest-
ments (which are socially excessive) and hence raises the cost of privacy protection.8 Hann et
al. (2002) empirically examine individuals’ tradeo¤s between the bene…ts and costs of providing
personal information to websites. They …nd that the bene…ts in terms of monetary reward and
future convenience signi…cantly a¤ect individuals preferences over websites with di¤ering privacy
policies. Among U.S. subjects, protection against errors, improper access, and secondary use of
personal information is worth $30:49 ¡ $44:62.
The rest of the paper is organized as follows: Section 2 presents the model. In Section
3 we solve for the equilibrium when there is a single programmer who needs to choose whether
to distribute his software online as an adware or as a shareware. Section 4 examines the e¤ect
of a technological improvement in the ad banners technology on the equilibrium and also con-
siders the policy implications of the model. Section 5 considers the equilibrium when there is
competition between adware and shareware. We conclude in Section 6.

2 The model
There are three types of agents in our model: a programmer, a continuum of consumers, and
n ¸ 2 …rms that sell consumer products. The programmer develops a software and distributes
it to consumers online. The software is either sold to consumers as a shareware or is bundled
with ad banners and is distributed to consumers for free. In the latter case, the programmer
collects fees from …rms that use ad banners that are displayed on the desktops of adware users
to advertise their products.

2.1 The timing of the model

The model evolves in three stages. In the …rst stage, the programmer chooses whether to
distribute his software online as a shareware or as an adware. Under the …rst option, the
programmer sets a price for the shareware that consumers need to pay for using it once the

8
McAndrews and Morgan study a related model in which phone service users can buy caller ID service in
order to block unwanted calls (e.g., from telemarketers) while callers can buy ID blocking which overrides caller
ID. They show that without both a well established right to identify callers, and well functioning markets for
the services, ine¤cient allocations emerge.

6
trial period expires. Under the second option, the programmer sets a per-viewer advertising fee
that …rms must pay in order to display ad banners on the desktops of adware users. Then, in
the second stage, each consumer decides whether or not to get the software. In the third stage
which is reached only if the programmer chooses to distribute his software as an adware, the n
…rms choose how many ad banners to display. Finally, consumers buy products and all payo¤s
are realized.

2.2 Consumers

There is a continuum of potential consumers with a total mass of one. Each consumer is
interested in buying one software and one out of n consumer products, each of which is produced
by a di¤erent …rm. Consumers belong to n di¤erent and equal sized groups: consumers who
belong to group i get a utility s if they buy product i and s ¡ t if they buy any other product,
where s ´ v ¡ p is the di¤erence between the gross utility of buying the “right” product, v, and
the product’s price, p. For simplicity we assume that p is the same for all products and treat
it as an exogenous parameter. Consumers, however, do not necessarily know of all …rms at the
outset and can buy from a given …rm only if they know about it. Speci…cally, a consumer who
belongs to group i learns about product i only with probability '. With probability 1 ¡ ', the
consumer ends up buying some other product. The expected utility of a consumer who chooses
not to buy software at all is therefore given by

U = 's + (1 ¡ ') (s ¡ t) : (1)

Shareware users: Using q to denote the shareware’s innate quality and ps its price, a shareware
user gets a net utility q ¡ ps from buying a shareware. Hence, the expected utility of a shareware
user who belongs to group i is

U s = q ¡ ps + U: (2)

To ensure that we obtain an interior solution for our model, we make the following
assumption:

7
Assumption 1: q · (1 ¡ ') t.

Assumption 1 implies that the direct utility from using the software, q, does not exceed
the utility loss due to choosing the “wrong” consumer product. This assumption will ensure
that in equilibrium, …rms will agree to pay for ad banners.

Adware users: If a consumer gets an adware, then in addition to the software he is interested
in, he also gets a software that tracks his behavior and enables the programmer to identify his
preferences. In our context, that means that the programmer can send adware users in group i
targeted ads that inform them about …rm i.
Let ki be the number of impressions that …rm i pays for (i.e., the number of times
that …rm i’s ad banners are displayed on the adware user’s desktop) and let m 2 [0; 1] be the
probability that an ad captures the user’s attention. The parameter m captures the e¤ectiveness
of ad banners in informing adware users of products they need. Assuming that the probability
of noticing each impression is independent across impressions, the probability that a consumer
in group i notices at least one of the ki impressions is

¹i = 1 ¡ (1 ¡ m)ki : (3)

With probability 1 ¡ ¹i = (1 ¡ m)ki ; the consumer ignores all ki impressions. The function ¹i
represents the intensity of online advertising.
In what follows it will be easier to express the model in terms of ¹i instead of ki : To this
end, note from equation (3) that

1
ki = z ln (1 ¡ ¹i ) ; z´ ; (4)
ln(1 ¡ m)

where z < 0. Equation (4) represents the number of impressions that …rm i needs to send in
order to ensure that its product is noticed by adware users in group i with probability ¹i . Since
ln (1 ¡ ¹i ) < 0, ki decreases with z, implying that as z increases towards 0, the ad banners
technology improves, fewer impressions are needed to attract the same level of attention from
consumers. Hence, z which increases with m, serves as a measure of how e¤ective ad banners
are in attracting the attention of adware users.

8
 Apart from ad banners, adware users in group i also learn about product i with prob-
ability ' just like shareware users. Assuming that the probability of learning about products
from an adware is independent of the probability of learning about it from other sources, the
probability that an adware user in group i buys product i is ¹i + (1 ¡ ¹i )' (the probability that
he learns about product i from the adware plus the probability that he does not learn about it
from the adware but does learn about it from other sources), while the probability that he buys
product j 6= i is (1 ¡ ¹i ) (1 ¡ '). We assume that if the consumer does not learn about product
1
i he learns about one of the n ¡ 1 “wrong” products with probability n¡1
.
While ad banners provide adware users with potentially useful information about con-
sumer products, they also violate their privacy by intruding on their right “to be left alone.”
We assume that the resulting disutility that adware users bear is directly related to the number
of impressions that are displayed on their desktops and is given by ¯ki , where ¯ is uniformly
distributed in the population on the support [0; B]. That is, consumers vary with respect to the
extent of their disutility from privacy loss.
Assuming that the innate quality of the adware is also q and using equation (4), the
expected utility of an adware user in group i is given by

Uia (¯) = q ¡ ¯z ln (1 ¡ ¹i ) + (¹i + (1 ¡ ¹i )') s + (1 ¡ ¹i ) (1 ¡ ') (s ¡ t) (5)

= q ¡ ¯z ln (1 ¡ ¹i ) + ¹i (1 ¡ ') t + U :

The third term on the right hand side of equation (5) represents the added information that
an adware user gets about consumer products which increases his chance to …nd the “right”
product. The second term is the disutility from lost privacy. This term decreases with z since
an improvement in the ad banners’ technology allows …rms to send adware users fewer ads in
order to get the same level of attention from them. Hence, the privacy of adware users is violated
to a lesser extent.

2.3 Firms

Firms’ decisions matter in our model only when the programmer o¤ers an adware. Otherwise,
each consumer (whether he owns a shareware or not) either learns about the “right” product

9
with probability ' or else picks one of the n ¡ 1 “Wrong” products at random. By symmetry
1
then, each …rm serves a mass n
of consumers, each of whom is interested in buying one unit.
Now, suppose that the programmer o¤ers an adware and suppose that a fraction ® of
all consumers adopt it. The remaining consumers do not adopt the adware; as in the shareware
1
case, each …rm end ups serving a fraction n
of these consumers. Hence, each …rm i faces a mass
1¡®
of n
of consumers who do not own any software.
Next, we turn to the demand for …rm i’s product among adware users. The total mass
®
of adware users is ®, of which n
are in group i. The probability that adware users in group i
buy from …rm i is ¹i + (1 ¡ ¹i )' while the probability that adware users in group j 6= i buy
(1¡¹j )(1¡')
from …rm i is n¡1
(the probability they fail to learn about product j times the probability
they learn about product i which is one of the n ¡ 1 “wrong” products). By the law of large
h P (1¡¹j )(1¡') i
numbers, …rm i serves a total of ®n ¹i + (1 ¡ ¹i )' + j6=i n¡1
adware users.
The expected demand that …rm i is facing, given the probability that the ad banners of
the n …rms are noticed, ¹1 ; :::; ¹n , is:
à !
1 ® X (1 ¡ ¹j ) (1 ¡ ')
Qi (¹1 ; : : : ; ¹n ) = (1 ¡ ®) + ¹i + (1 ¡ ¹i )' + : (6)
n n j6=i
n¡1

Assuming that the pro…t per unit of consumer product is ¼ ´ p ¡ c, where p is the price
and c is the (constant) marginal cost, using r to denote the per-impression per-user price that
…rms pay adware programmers, and using equation (4), the expected pro…t of …rm i is

®
¦i (¹1 ; : : : ; ¹n ) = Qi (¹1 ; : : : ; ¹n )¼ ¡ rz ln (1 ¡ ¹i ) : (7)
n

2.4 Programmers

For simplicity we begin the model at the point in time after the programmer has already success-
fully developed his software. Hence, the development costs are already sunk at this point and
the programmer’s problem is how to distribute the software online. If the programmer chooses
to distribute it as a shareware, then given equations (1) and (2), it is clearly optimal to set a
price, ps = q, for the shareware. The programmer’s pro…t in this case is q.
If the programmer chooses to distribute the software as an adware then he allows users

10
to download it for free but then collects money from …rms that advertise their products through
ad banners. The programmer then sets a price r for each ad that reaches each consumer and
P
lets …rms choose how many for ads to display. The programmer’s pro…t in this case is r ni=1 ki ,
or using equation (4),

n
X ®
rz ln (1 ¡ ¹i ) :
i=1
n

The programmer will then choose r to maximize his pro…t from selling ad banners to …rms.
In the …rst stage of the game, the programmer will compare the resulting pro…t from selling
adware to the pro…t from selling shareware, q, and depending on which is larger will decide how
to distribute his software.

3 Equilibrium
In this section we solve for the subgame perfect equilibrium in our three-stage model. The
equilibrium in the shareware subgame is simple: the programmer sets the price of shareware
equal to q and all consumers buy the software. Consequently, the programmer’s pro…t is q.
Hence we only need to solve stages 2 and 3 in the adware subgame. In order to solve stage
1 of the game, we will then compare the programmer’s pro…ts from shareware and adware to
determine which mode of the distribution the programmer will choose.

3.1 Stage 3: the choice of targeted ads

We begin with the third stage in which, assuming that the programmer chooses to distribute
his software as an adware, each …rm i chooses ¹i . The …rst order condition for ¹i is:

@¦i (¹1 ; : : : ; ¹n ) ® ® 1
= (1 ¡ ') ¼ + rz = 0:
@¹i n n 1 ¡ ¹i

The …rst term in the derivative is the marginal bene…t from ad banners: without ad banners,
…rm i cannot capture the attention of the 1 ¡ ' adware users in group i who are not aware of its
product from other sources. The associated loss of revenue per each such user is ¼. The second

11
term is the marginal cost of ad banners (since z < 0 this term is negative).
Solving the …rst order condition for ¹i , reveals that in equilibrium,

rz
b (r) ´ 1 +
¹i = ¹ ; i = 1; 2; :::; n: (8)
(1 ¡ ') ¼

b (r) ¸ 0.
Assumption 1 ensures that in stage 1 of the model, the programmer will set r such that ¹
b (r) is the
If Assumption 1 fails, …rms will prefer not to pay for any ad banners. The function ¹
demand function of each of the n …rms for ad banners given the price per-impression per-user,
r, that the programmer sets at the …rst stage of the game. Recalling that z < 0, equation (8)
b (r) is a linearly decreasing function of r.
shows that ¹
b (r) is increasing with ¼ and z but is decreasing with '. In-
Equation (8) shows that ¹
tuitively, …rms demand more ad banners when the market is more pro…table (¼ is larger) and
when ad banners are more e¤ective in attracting consumers’ attention (z increases towards 0),
but demand fewer ad banners when adware users are more likely to know about the “right”
products anyway (' is larger). It is interesting to note that each …rm’s demand for ad banners
is independent of ® which is the fraction of consumer who choose to buy adware. The reason for
this is that …rms pay per-viewer fees to the programmer and hence if there is a smaller number
of adware users, the …rms payments to the programmer decrease proportionally.

3.2 Stage 2: consumer’s demand for adware

When the programmer chooses to distribute the software as an adware, each consumer needs
b(r) from equation (8) into equation (5)
to decide whether or not to adopt it. Substituting for ¹
and simplifying, the di¤erence between the utility from having an adware and not having it is

U a (¯) ¡ U = q ¡ ¯z ln (1 ¡ ¹
b(r)) + ¹
b(r) (1 ¡ ') t: (9)

That is, relative to a non user, an adware user gets a utility q from using the software but also
b(r)) from privacy loss. And, due to targeted ads, the adware user
bears a disutility ¯z ln (1 ¡ ¹
b (r) (1 ¡ ') t from buying the “right” product when absent adware
gets an additional utility ¹
b (r) (1 ¡ ').
he will buy one of the “wrong” products. The probability of this event is ¹

12
 Consumers will adopt the adware if and only if U a (¯) ¡ U ¸ 0. Since U a (¯) ¡ U is
decreasing with ¯, the equation U a (¯) = U de…nes a unique value of ¯, denoted b̄ (q; ¹
b (r)),
below which the consumer will get an adware. Using equation (9),

b̄ (q; ¹ q+¹ b (r) (1 ¡ ') t

b (r)) = : (10)
z ln (1 ¡ ¹b(r))

Since ¯ is uniformly distributed in the population on the interval [0; B], the fraction of consumers
who choose to get an adware, ®, is
( )
b̄ (q; ¹
b (r))
b (q; r) ´ min
®=® ;1 : (11)
B

Equation (11) shows that the fraction of adware adopters is increasing with the bene…ts from
b (r) (1 ¡ ') t, and decreasing with the number of impressions that each
having the adware, q + ¹
b(r)).
adware user faces, z ln (1 ¡ ¹
The aggregate demand of …rms for ad banners is

b (q; r)z ln (1 ¡ ¹
Q(q; r) = ® b (r)) ;

b (q; r) is the mass of adware users and z ln (1 ¡ ¹

where ® b (r)) is the equilibrium number of
impressions that each …rm chooses to send each adware user. Using equations (10) and (11),
8
> q+b
¹(r)(1¡')t b̄ (q;b
¹(r))
>
> ; · 1;
< B B

Q(q; r) = (12)
>
>
>
: z ln (1 ¡ ¹ b̄ (q;b
¹(r))
b (r)) ; B
> 1:

The expression in the …rst line of (12) is the aggregate demand for ad banners when at least
b (q; r) < 1). The expression in the second line is
some consumers do not adopt the adware (i.e., ®
b (q; r) = 1).
the aggregate demand for ad banners when all consumers adopt the adware (i.e., ®
b (r) is a decreasing function of r, the aggregate demand of …rms for ad banners
Notice that since ¹
is a downward sloping function of the ad banners’ price. Intuitively, an increase in r creates three
b (r))
e¤ects. First, each …rm chooses to pay for fewer ad banners per adware user (i.e., z ln (1 ¡ ¹

13
is smaller). Second, the decrease in the number of ad banners lowers the privacy loss of adware
users and therefore tends to boost their number (the denominator in b̄ (q; ¹
b (r)) falls). Third,
the decrease in the number of ad banners means that adware users obtain less information
about consumer products and hence are less inclined to adopt the adware (the numerator in
b̄ (q; ¹
b (r)) falls). In our formulation, the …rst and second e¤ects just cancel each other out when
b (q; r) < 1. Given that an increase in r leads to fewer adware users, the aggregate demand for
®
ad banners, which is proportional to the number of adware users, falls.
b (q; r) = 1, all consumers already have an adware so only the …rst e¤ect is at work.
When ®
Hence, in this case an increase in r lowers the aggregate demand for ad banners by making it a
more costly means of advertising from the viewpoint of …rms.

3.3 Stage 1: the programmer’s problem

In the …rst stage of the game, the programmer needs to choose whether to distribute his software
online as a shareware or as an adware. To determine his choice, we need to compare the
programmer’s pro…t under each alternative. In the shareware case, the programmer sets the price
of the shareware equal to q, all consumers buy and the programmer’s pro…t is q. To determine
the programmer’s pro…t from adware, we …rst needs to determine the pro…t maximizing per-
impression price, r. To this end, notice that given the aggregate demand for ad banners, the
programmer’s pro…t from adware is

Oa (q; r) = rQ(q; r); (13)

where Q(q; r) is given by equation (12). Noting from equation (12) that Q(q; r) is increasing
b (r) is decreasing with r, it follows that Oa (r)
b (r) and recalling from equation (8) that ¹
with ¹
is strictly concave in r (both when B is small and when B is large). Hence, the optimal value
dOa (r)
of r is uniquely de…ned by the equation dr
= 0. Solving this equation, the pro…t maximizing

14
value of r is
8
>
> ¡ (q+(1¡')t)¼ ; B ¸ B;
>
< 2zt

r¤ = (14)
>
>
>
: ¡ (1¡')¼
ze
; B < B:

where

q + (1 ¡ ') t
B ´ b̄ (q; ¹
b (r¤ )) = ³ ´; (15)
2z ln q+(1¡')t
2(1¡')t

is the highest consumer type who still adopts an adware.

Note that r¤ in the …rst line of equation (14) exceeds r¤ in the second line of equation
(14). Intuitively, when B is large, some consumers do not adopt the adware. As we saw earlier,
an increase in r lowers the number adopters further and hence lowers the aggregate demand for
ad banners. The programmer must take this negative e¤ect of r into account and hence sets a
lower r than in the case where all consumers adopt the adware.
When B < B, all consumers get an adware. Substituting for r¤ from the …rst line of
equation (14) into equation (8), the equilibrium probability that a consumer in group i notices
at least one of …rm i’s impressions is
8
>
> (1¡')t¡q
; B ¸ B;
>
< 2(1¡')t
¤ ¤
b (r ) =
¹ ´¹
>
>
>
: 1 ¡ 1;
e
B < B;

Clearly, ¹¤ < 1. Assumption 1 ensures that ¹¤ ¸ 0.

When B < B, all consumers get an adware. On the other hand, when B ¸ B, the
fraction of consumers who get an adware is

B
b (q; r¤ ) =
® :
B

b (q; r¤ ) > 0. Since B ¸ B, ®

By Assumption 1 and since z < 0, ® b (q; r¤ ) · 1.

15
 Given r¤ , the programmer’s pro…t from adware is
8
>
> ¡ (q+(1¡')t)
2
¼
; B ¸ B;
>
< 4ztB

Oa (q; r¤ ) = : (16)
>
>
>
: (1¡')¼
e
; B < B:

Notice that the expression in the second line of equation (16) exceeds the expression in the …rst
line of equation (16): the programmer is clearly better o¤ when the support of the disutility
parameter, ¯, is small because then, in equilibrium, all consumers adopt the adware and hence
the demand for ad banners is maximal.9
When B < B (the support of the distribution of ¯ is “narrow”), O a (q; r¤ ) is a linearly
increasing function of the probability that the consumers will pick a “wrong” product, 1 ¡ ',
and the pro…t per-unit of consumer product, ¼, but is independent of the software’s quality
q. Intuitively, …rms value ad banners more when 1 ¡ ' and ¼ are large and this enables the
programmer to charge high per-viewer fees for ad banners. And, since all consumers adopt the
software anyway, the innate quality of the software is immaterial.
When B ¸ B (the support of the distribution of ¯ is “wide”), equation (16) shows that
Oa (q; r¤ ) is increasing with the software’s quality q, with the probability that consumers will
pick a “wrong” product, 1 ¡ ', with the pro…t per-unit of consumer product, ¼, and with the
e¤ectiveness of ad banners, z. Intuitively, an increase in q and 1¡' make adware more attractive
so more consumers adopt it. Consequently, …rms are willing to pay more money for ad banners
as those become more e¤ective means of advertising. Likewise, an increase in ¼ implies that
…rms are more eager to advertise. When z increases, …rms need to send fewer ad banners to
get consumers’ attention. This boosts the demand for ad banners and hence the programmer’s
pro…t.
In what follows, we shall restrict attention to the (more interesting) case where B is
relatively large by making the following assumption:

9
To see that the expression in the second line of equation (16) is larger than
³ the expression
´ in the …rst line
(q+(1¡')t)2 ¼ (q+(1¡')t)2 ¼ (q+(1¡')t)¼ q+(1¡')t
of equation (16), note that ¡ 4ztB ·¡ 4ztB
= ¡ 2t ln 2(1¡')t : The last expression
¡ ¢
is decreasing with q and hence is largest when q = 0. Its value at q = 0, is ¡ (1¡')¼
2 ln 12 < (1¡')¼
e : Hence,
2
¡ (q+(1¡')t)
4ztB
¼
< (1¡')¼
e for all q.

16
Assumption 2: B ¸ B.

Having found the pro…t from adware, we now turn to the programmer’s choice between
adware and shareware. To this end we need to compare the pro…t from adware, Oa (r¤ ), with
the pro…t from shareware, q:

Proposition 1: The solution to the programmer’s problem is as follows:

(i) If B < ¡ (1¡')¼

z
, the programmer will o¤er adware for all values of q.

(ii) If B ¸ ¡ (1¡')¼
z
, the programmer will o¤er adware if q · q1 and will o¤er shareware if
q > q1 , where
µ q ¶
(1¡')¼ (1¡')¼
t 1+ 2zB
¡ 1+ zB
q1 ´ ¼ :
¡ 2zB

Proof: Suppose that B < ¡ (1¡')¼

z
and let

(q + (1 ¡ ') t)2 ¼
¢(q) ´ O a (q; r¤ ) ¡ q = ¡ ¡ q:
4ztB

¼
Note that ¢"(q) = ¡ 2ztB > 0 so that ¢(q) has a unique minimum, qmin . Evaluated at qmin ,

ztB
¢(qmin ) = + (1 ¡ ') t:
¼

Since z < 0, it follows that ¢(qmin ) > 0 whenever B < ¡ (1¡')¼

z
. Hence when B < ¡ (1¡')¼
z
,
Oa (r¤ ) > q for all q.
Next, suppose that B ¸ ¡ (1¡')¼
z
. Then, the equation O a (r¤ ) = q yields two solutions:
µ q ¶ µ q ¶
(1¡')¼ (1¡')¼ (1¡')¼ (1¡')¼
t 1+ 2zB
¡ 1+ zB
t 1+ 2zB
+ 1+ zB
q1 ´ ¼ ; q2 ´ ¼ ;
¡ 2zB ¡ 2zB

where Oa (r¤ ) > q if q < q1 or q > q2 , and O a (r¤ ) < q if q1 · q · q2 . However, by Assumption

17
1, q · (1 ¡ ')t. Comparing q2 with (1 ¡ ')t reveals that
µ q ¶
(1¡')¼ (1¡')¼
t 1+ zB
+ 1+ zB
q2 ¡ (1 ¡ ')t = ¼ ¸ 0;
¡ 2zB

where the inequality follows because the assumption that B ¸ ¡ (1¡')¼

z
implies that 1 + (1¡')¼
zB
¸
0. Likewise, comparing q1 with (1 ¡ ')t reveals that
µ q ¶
¼(1¡') ¼(1¡')
t 1+ zB
¡ 1+ zB
q1 ¡ (1 ¡ ')t = ¼ :
¡ 2zB
q
Since z < 0 and since B ¸ ¡ (1¡')¼
z
, 0 · 1+ (1¡')¼
zB
< 1; hence, 1 + (1¡')¼
zB
· 1+ (1¡')¼
zB
,
implying that q1 · (1 ¡ ')t: ¥

Proposition 1 is illustrated in Figure 1. When q is low, shareware users derive little bene…t
from the software and hence are not willing to pay much for it. Hence, the programmer’s pro…t
from shareware is small. On the other hand, O a (0; r¤ ) > 0, implying that the programmer’s
pro…t from adware is positive even when q = 0. Hence, for small values of q, the programmer
prefers to distribute his software as an adware. Intuitively, the programmer’s pro…t from adware
comes from …rms who advertise their products through ad banners. Obviously, …rms will pay
for ad banners only if at least some consumers adopt the adware. The question then is whether
consumers will or will not adopt the adware when q is small. To address this question, recall
that the bene…t to adware users comes both from the software they use as well as from the
information about consumer products that they get from ad banners. When q = 0, adware
users do not bene…t from the software itself, but since they do not pay for it either, they may
still adopt it provided that the bene…t from ad banners exceeds the associated disutility from
privacy loss. As equation (8) shows, the intensity of advertising per viewer is independent of the
mass of adware users. Hence, adware users get useful information about consumers’ products
even when q = 0. Although adware users also bear a disutility from privacy loss, equations (10)
and (11) reveal that for users with relatively small values of ¯, this disutility is small relative
to the bene…ts. Consequently, the programmer can make money from adware even when q = 0.

18
 Oa(r*) q

Feasible values of q

Adware Shareware

(1-ϕ)2tY/2

q1 (1-ϕ)t q2 q

Figure 1: Illustrating the Proposition 1

By continuity, the same is true for small values of q. But as q increases, this preference may
eventually be reversed provided that B ¸ ¡ (1¡')¼
z
. If B < ¡ (1¡')¼
z
, the pro…t from adware
exceeds the pro…t from shareware for feasible values of q (by Assumption 1, q · (1 ¡ ')t). The
reason why the pro…t from shareware may exceed the pro…t from adware when q is large is that
the pro…t from shareware increases one-to-one with q. By contrast, an increase in q a¤ects the
pro…t from adware only indirectly by inducing more consumers to adopt adware. The rate of
increase in Oa (q; r¤ ) is less than 1 for low values of q.
Casual observation suggests that many popular software are …rst distributed as adware,
but then, newer and improved versions are distributed as shareware. Examples for this pat-
tern include Gozilla and GetRight which are the second and the third most popular download
managers on download.com with over 17 million and 11 million downloads respectively. Propo-
sition 1 provides a possible explanation for this pattern. It should be noted that q need not
represent the “true” quality of the software but rather its perceived quality by potential users.
If potential users believe that q is lower than it really is and the programmer has no way of
credibly convincing them otherwise, then it pays the programmer to …rst distribute the software
as an adware. As more consumers use the software and learn its true quality, the perceived
quality of the software increases and the programmer bene…ts from distributing newer versions
as shareware.

4 Comparative statics and policy implications

As we already mentioned in the Introduction, adware technology is still relatively new. It is
therefore expected that in the near future, the technology of sending context-based targeted ads
to speci…c consumers through ad banners will improve. For instance, Google plans to o¤er Gmail
which is a free webmail service that will analyze the contents of emails and will place relevant
text ads and links to related web pages adjacent to email messages. Such improvement have
raised concerns about the increasing loss of privacy on the Internet. It is therefore interesting to
…nd out how such improvements will a¤ect consumers in the context of our model where both
privacy loss and the bene…ts from improved information on consumers’ products are explicitly
taken into account. We begin this section by studying this issue. To this end, we assume that an

19
improvement in the technology of ad banners raises the value of z towards 0 and enables …rms
to attract the attention of prospective consumers more e¤ectively. We are interested in …nding
how the equilibrium outcome and especially consumers’ surplus respond to such an increase in
z. Throughout this section we will maintain Assumption 2; that is we assume that B ¸ B.
To study the e¤ect of an increase in z, note from the …rst lines in equations (14) and (16)
that both r¤ and Oa (r¤ ) increase with z. The reason for this is that the demand of …rms for ad
banners increases which in turn enables the programmer to charge higher prices for ad banners
and make more money by distributing the software as adware. Consequently, the programmer
o¤ers adware instead of shareware for a wider set of parameters.
Does an increase in z bene…ts consumers as well? To address this question, recall that
consumers adopt an adware if and only if ¯ · b̄ (q; ¹
b (r)); the utility of each adware user in
1
group i is Uia (¯) while the utility of each non user is U . Recalling that there are n
consumers
in each group and using equations (5) and (10), consumers’ surplus is given by

Xn Z b̄ (q;b¹(r)) a Z B
1 Ui (¯) U
CS(q; r) = d¯ + d¯ (17)
i=1
n 0 B b̄ (q;b
¹ (r)) B
b (r) (1 ¡ ') t)2
(q + ¹
= + U:
2zB ln (1 ¡ ¹b(r))

Substituting for r¤ from the …rst line of equation (14), the value of consumers’ surplus in equi-
librium is

(q + (1 ¡ ') t)2
CS(q; r¤ ) = ³ ´ + U:
8zB ln q+(1¡')t
2(1¡')t

³ ´
q+(1¡')t
By Assumption 1, ln 2(1¡')t
· 0 implying that CS(q; r¤ ) is increasing with z. Hence,
consumers are made better o¤ by a technological improvement in the e¤ectiveness of ad banners.
The reason for this is as follows. While consumers can download adware for free, each consumer
will get an adware if and only if the bene…t from improved information about consumers products
exceeds the disutility from privacy loss. Since the utility of consumers who do not adopt an
adware, U , is independent of z, it follows by revealed preferences that an increase in z boosts
consumers’ surplus if and only if it increases the fraction of adware users. The latter is given by

20
B
B
, which by equation (15) is increasing with z.
b(r) …xed,
Intuitively, an increase in z has two e¤ects on consumers. First, holding ¹
an increase in z means that fewer impressions are needed to get the same level of attention
from adware users. Hence, adware leads to a smaller loss of privacy and hence becomes more
b(r) (the intensity with which …rms use ad banners)
attractive. Second, an increase in z a¤ects ¹
both directly, as well as indirectly through its e¤ect on r¤ : The direct e¤ect of z on ¹
b(r) is
positive because an increase in z makes ad banners more e¤ective in attracting consumers and
this boosts the …rms’ demand for ad banners. At the same time, an increase in z induces the
programmer to raise r¤ and this depresses the …rms’ demand for ad banners. In our model
b(r¤ ) is independent of z.
however, the direct and indirect e¤ects cancel each other out so ¹
Consequently, only the …rst, positive, e¤ect is at work, implying that an increase in z makes
adware more attractive to consumers and hence more consumers adopt the adware.
b(r¤ ) is independent of z, the number of impressions
It should also be noted that since ¹
b (r¤ )), is decreasing with z. This implies in turn that an
that each adware user gets, z ln (1 ¡ ¹
improvement in the ad banners technology will lead to less violation of privacy rather than more
as some technological experts expect.
We summarize this discussion in the following proposition:

Proposition 2: An improvement in the ad banners technology that increases z induces the

programmer to adopt adware for a larger set of parameters and to raise the price of ad banners.
In equilibrium, more consumers adopt the adware and both the programmer and consumers
become better o¤. Moreover, following the improvement, fewer impression are sent and hence,
adware users face smaller loss of privacy.

We proceed by evaluating the policy implications of Proposition 1. To this end, we …rst

need to characterize the socially e¢cient outcome. As before, we shall impose Assumption 2
and restrict attention to cases in which B ¸ B. Our measure of social welfare is the sum of
the consumers’ surplus plus pro…ts, which in our context consist of the pro…ts of …rms and the
programmer’s pro…t.
In the shareware case, all consumers buy the shareware at a price ps = q. Hence, by
equation (2), consumers’ surplus equals U. Noting that the programmer’s pro…t is ps = q, and

21
the aggregate pro…ts of …rms are equal to ¼, social welfare in the shareware’s case is

W s (q) = U + ¼ + q: (18)

In the adware case, consumers’ surplus is given by equation (17), while the aggregate prof-
P P
its of …rms are ni=1 ¦i (b b (r)), and the programmer’s pro…t rz ni=1 ®n ln (1 ¡ ¹
¹ (r) ; : : : ; ¹ b (r)),
b̄ (q;b
¹(r))
where ® = B
is the fraction of adware users. Since the payments of …rms to the program-
mer are washed out, the sum of the aggregate pro…ts of …rms and the programmer’s pro…t equal
¼. Hence, social welfare in the adware’s case is

a b (r) (1 ¡ ') t)2

(q + ¹
W (q; r) = U + ¼ + : (19)
2zB ln (1 ¡ ¹b(r))

The socially optimal value of r, denoted r¤¤ , maximizes W a (q; r). In the following result
we compare r¤¤ with the equilibrium value of r, r¤ , which is given in the …rst line of equation
(14):

Proposition 3: The equilibrium value of r is excessive: r¤ > r¤¤ for all values of q for which
the programmer o¤ers an adware. Consequently, in equilibrium, there is too few ad banners.

Proof: The socially optimal per-viewer price of ad banners, r¤¤ , is implicitly de…ned by the
following …rst order condition:
µ ¶
@W a (q; r) b(r)) ln (1 ¡ ¹
(1 ¡ ¹ b(r)) (1 ¡ ') t 1 CS (q; r) ¹b0 (r)
=2 + = 0: (20)
@r q+¹ b (r) (1 ¡ ') t 2 b(r)) ln (1 ¡ ¹
(1 ¡ ¹ b(r))

@W a (q;r)
Noting that ln (1 ¡ ¹ b0 (r) < 0, the sign of
b(r)) · 0 and recalling from equation (8) that ¹ @r

depends on the sign of the bracketed term in equation (20). Evaluating this term at r¤ and
(1¡')t¡q
b (r¤ ) =
recalling that ¹ 2(1¡')t
whenever B ¸ B; yields

µ ¶ µ µ ¶ ¶
@W a (q; r¤ ) q + (1 ¡ ') t 1
sign = sign ln + ;
@r 2 (1 ¡ ') t 2
³ ´ ³ ´
where ln q+(1¡')t + 1
< 0 for all q · (1 ¡ ') t p2 ¡ 1 . Hence, r ¤ > r ¤¤ for all q ·
³ 2(1¡')t´ 2 e
2
(1 ¡ ') t pe ¡ 1 : Now, recall from Proposition 1 that the programmer o¤ers an adware if

22
and only if q · q1 and note that
µ ¶
2
(1 ¡ ') t p ¡ 1 ¡ q1
e
"r µ ¶#
t (1 ¡ ')¼ (1 ¡ ') ¼
= ¼ 1+ ¡ 1+ p
¡ 2zB zB ezB
· µ ¶¸
t (1 ¡ ')¼ (1 ¡ ') ¼
> ¼ 1+ ¡ 1+ p
¡ 2zB zB ezB
· ¸
t (1 ¡ ')¼ (1 ¡ ') ¼
= ¼ ¡ p > 0;
¡ 2zB zB ezB
³ ´
where the …rst inequality follows because z < 0. Hence, q1 < (1 ¡ ') t p2 ¡ 1 . This implies
e

in turn that for values of q for which the programmer o¤ers adware, r¤ > r¤¤ : The conclusion
b0 (r) < 0.
regarding ad banners follows by noting that ¹ ¥

Proposition 3 indicates that whenever the programmer o¤ers adware, the equilibrium
price of ad banners, r, is excessive; as a result, there is too little usage of ad banners by …rms.10
Intuitively, when the programmer sets r, he fails to take into account the resulting e¤ect on
adware users. Given that an increase in r induces …rms to buy fewer ad banners, it is clear
that r will end up being excessive if on average, the bene…t of adware users from information
on consumers’ products exceeds their disutility from privacy loss. For adware users with low
values of ¯, the bene…t clearly exceeds the cost. For adware users with high values of ¯, the cost
exceeds the bene…t since adware users also get a utility q from using the software itself and hence
are willing to adopt the adware even when their disutility from privacy loss exceeds the bene…t
from information about consumers’ products. As q increases, there are more adware users for
whom the net e¤ect of ad banners is negative. But since the programmer switches to shareware
when q increases above q1 , the fraction of adware users for whom the e¤ect of ad banners is
negative must be limited. Proposition 3 shows that, indeed, so long as the programmers o¤ers
an adware, the aggregate bene…t of consumers from ad banners exceeds the associated disutility.
The programmer fails to take into account this net bene…t and hence sets an excessive price for
ad banners.

10
Again, it is worth emphasizing that we only consider “legitimate” ad-supported software, but not spyware
and malware, which, from personal experience, are a major nuisance.

23
 Proposition 3 compares the equilibrium value of r with its socially optimal value. How-
ever, this comparison is relevant only if the programmer distributes his software as an adware.
Our next task is to compare the equilibrium choice between adware and shareware with the
socially optimal choice.

Proposition 4: The programmer o¤ers an adware for a smaller set of parameters than is so-
1
cially e¢cient whenever r¤ z ln (1 ¡ ¹
b(r¤ )) · 2
and o¤ers an adware for a wider set of parameters
b(r¤¤ )) ¸ 12 .
than is socially e¢cient whenever r¤¤ z ln (1 ¡ ¹

Proof: It is socially e¢cient to distribute the software as an adware if and only if W a (q; r¤¤ ) ¸
W s (q): Using equations (18) and (19), this condition is equivalent to

b (r¤¤ ) (1 ¡ ') t)2

(q + ¹
¸ q: (21)
2zB ln (1 ¡ ¹ b(r¤¤ ))

Using the …rst line in equation (16), it follows that in equilibrium, the programmer will distribute
his software as an adware if and only if

b (r¤ ) (1 ¡ ') t)2 r¤

(q + ¹
¸ q: (22)
B

Hence, to prove the proposition it is su¢cient to prove that the left side of (21) exceeds the left
b(r¤ )) · 12 . Then,
side of (22). Now, suppose that r¤ z ln (1 ¡ ¹

b (r¤¤ ) (1 ¡ ') t)2

(q + ¹ b (r¤ ) (1 ¡ ') t)2
(q + ¹
¸
2zB ln (1 ¡ ¹ b(r¤¤ )) 2zB ln (1 ¡ ¹ b(r¤ ))
b (r¤ ) (1 ¡ ') t)2 r¤
(q + ¹
¸ ;
B

where the …rst inequality follows by revealed preferences (r¤¤ maximizes the left side of (21)).
b(r¤¤ )) ¸ 12 , then,
On the other hand, if r¤¤ z ln (1 ¡ ¹

b (r¤ ) (1 ¡ ') t)2 r¤

(q + ¹ b (r¤¤ ) (1 ¡ ') t)2 r¤¤
(q + ¹
¸
B B
b (r¤¤ ) (1 ¡ ') t)2
(q + ¹
¸ ;
2zB ln (1 ¡ ¹ b(r¤¤ ))

24
where the …rst inequality follows by revealed preferences (r¤ maximizes the left side of (22)).
¥

b(r)) is the
To interpret Proposition 4, note from the second line of (12) that z ln (1 ¡ ¹
b(r))
aggregate demand for ad banners when all consumers adopt the adware. Hence, rz ln (1 ¡ ¹
is the programmer’s pro…t when the adware market is covered. Proposition 4 states that the
programmer will underprovide adware if, evaluated at the pro…t maximizing price of ad banners,
r¤ , this pro…t is su¢ciently small, but will overprovide adware if at the socially optimal price
of ad banners, r¤¤ , this pro…t is su¢ciently large. To go deeper into the su¢cient conditions in
Proposition 4, note that
µ ¶
¤ (q + (1 ¡ ') t) ¼
¤ q + (1 ¡ ') t
b(r )) = ¡
r z ln (1 ¡ ¹ ln :
2t 2 (1 ¡ ') t
³ ´
q+(1¡')t
By Assumption 1, ln 2(1¡')t
· 0, so this expression increases with ¼. Hence, underprovision
of adware is especially likely when the pro…t of …rms on consumers’ products, ¼, is small. By
contrast,
µ ¶
¤¤ ¤¤ ¤¤ r¤¤ z
b(r )) = r z ln 1 ¡
r z ln (1 ¡ ¹ ;
(1 ¡ ') ¼

where (20) shows that r¤¤ is independent of ¼. Hence, r¤¤ z ln (1 ¡ ¹

b(r¤¤ )) is decreasing with
¼, implying that the su¢cient condition for overprovision of adware is especially likely to hold
when ¼ is large.
As we mentioned in the Introduction, currently, U.S. legislators are considering legislation
that would regulate or even completely ban any software that monitors usage of the internet and
transmits information back from a location. Such legislation is mainly motivated by the rapid
growth of spyware and malware. However, the legislation may also e¤ectively make it impossible
to distribute legitimate adware on line. Using our model we can examine the desirability of bans
on adware.

Proposition 5: A ban on adware hurts both consumers and the programmer when q · q1 and
is irrelevant otherwise.

25
Proof: When q > q1 , the programmer o¤ers shareware so a ban on adware is irrelevant. When
q · q1 , the programmer prefers to o¤er adware. By revealed preferences, it is obvious that a ban
on adware would hurt the programmer in this range. As for consumers, note that consumers’
surplus is U in the shareware case and CS(q; r¤ ) in the adware case. The di¤erence between the
two expressions is

(q + (1 ¡ ') t)2
U ¡ CS(q; r¤ ) = ¡ ³ ´ < 0:
8zB ln q+(1¡')t
2(1¡')t

Hence, a ban on adware hurts consumers as well. ¥

That a ban on adware hurts the programmer is obvious. Less obvious is why such a ban
hurts consumers. The reason for this is simple however: when the programmer o¤ers shareware,
he charges consumers a price equal to their bene…t from using the software. Hence, consumers
get no surplus from the shareware. On the other hand, when the programmer o¤ers adware,
consumers with low privacy concerns get a positive surplus from the adware while those with
high privacy concerns do not adopt it. Hence, a ban on adware hurts consumers.

5 Competition in the software market

In this section we consider competition between two programmers who develop software of
equal quality. Given that the quality of the software is the same, it is clear that in order to
avoid Bertrand competition, one programmer will choose to distribute his software as shareware
and the other will choose to distribute it as adware. In what follows, we study the ensuing
competition between the two programmers by solving the following three stage game. In the
…rst stage, the two programmers simultaneously choose their prices: the shareware programmer
chooses the price of shareware, ps , while adware programmer chooses the per-impression price,
r. Given r, …rms decide in the second stage of the game how many ad banners to pay for. In the
third stage, each consumer decides whether to get a shareware or an adware. Finally, consumers
buy products and all payo¤s are realized.

26
 The pro…t of the shareware programmer is

(1 ¡ ®)ps ;

where 1 ¡ ® is the fraction of consumers who decide to purchase shareware. The pro…t of the
adware programmer is

n
X
rz ln (1 ¡ ¹i ) :
i=1

b (r), is given by equation

Given r, the intensity of advertising through ad banners, ¹
(8), just as in the single programmer’s case. Notice that each …rm i’s decision about ¹i is
independent of the number of consumers who will eventually get an adware. Hence, we can
solve for the second stage of the game independently of the third stage. The equilibrium prices
will be denoted by pbs and rb:

5.1 Shareware vs. adware

Consumers decide whether to get adware or shareware by comparing their utilities from the
b(r) for all i, it follows that
two options. Using equations (2) and (5), and noting that ¹i = ¹
consumers will choose to get adware if

Uia (¯) ¡ U s = ps ¡ ¯z ln (1 ¡ ¹
b(r)) + ¹
b(r) (1 ¡ ') t ¸ 0;

b(r)) · 0 and z < 0, this expression

and will choose to get shareware otherwise. Since ln (1 ¡ ¹
is decreasing with ¯. Hence, the equation U a (¯) = U s de…nes a unique value of ¯, denoted
b̄ (ps ; ¹
b (r)), below which the consumer will choose to get an adware and above which he will
choose to get a shareware:

b̄ (ps ; ¹ ps + ¹b (r) (1 ¡ ') t

b (r)) = : (23)
b(r))
z ln (1 ¡ ¹

That is, consumers with small privacy concerns will purchase adware while those with large
concerns will purchase shareware. Notice that b̄ (ps ; ¹
b (r)) is an increasing function of ps : the

27
higher the price of shareware, the more consumers will adopt a adware rather than shareware.
In other words, adware and shareware are substitutes for consumers.
In order to avoid corner solutions, we shall make the following assumption:

Assumption 3: B > b̄ (b
ps ; ¹
b (b
r)) :

Assumption 3 modi…es Assumption 2 by setting a new lower bound on B. This assump-

tion ensures that B is su¢ciently large to ensure that in equilibrium, at least some consumers
will …nd the disutility from privacy loss so large that they will prefer to adopt shareware instead
of adware. Of course, the shareware programmer can always lower ps to induce some consumers
to adopt shareware. Assumption 3 ensures that if ps is lowered to 0, the shareware programmer
can attract at least some consumers.
Since ¯ is uniformly distributed in the population on the interval [0; B], the fraction of
consumers who adopt an adware is

b̄ (ps ; ¹
b (r))
b (ps ; r) =
® : (24)
B

b (ps ; r), the demand for shareware is 1 ¡ ®

Given ® b (ps ; r), so the pro…t of the shareware program-
mer is
µ ¶
s s s s ps + ¹
b (r) (1 ¡ ') t s
b (p ; r)) p = 1 ¡
O (p ; r) = (1 ¡ ® p; (25)
zB ln (1 ¡ ¹b(r))

b (r) is given by equation (8). The pro…t of the adware programmer is

where ¹

(ps + ¹
b (r) (1 ¡ ') t) r
Oa (ps ; r) = ®
b (ps ; r)z ln (1 ¡ ¹
b (r)) r = : (26)
B

5.2 Shareware prices and per-impression prices

In the …rst stage of the game, the shareware programmer chooses the price of shareware, ps ,
while adware programmer chooses the per-impression price, r.

ps ; rb); where pbs > 0 and ¡ (1¡')¼

Lemma 1: There exists a unique Nash equilibrium, (b 3z
< rb <
¡ (1¡')¼
2z
:

28
 @ 2 Os (ps ;r) 2 @ 2 Oa (ps ;r) 2
Proof: It is straightforward to verify that @(ps )2
= ¡ zB ln(1¡b
¹(r))
< 0 and @r2
= ¡ z¼ <
0. Hence, the best response functions of the shareware and adware programmers, respectively,
@Os (ps ;r) @Oa (ps ;r)
are uniquely de…ned by the …rst order conditions, @ps
= 0 and @r
= 0. Using equation
(8) and simplifying, the best response functions of the two programmers are given by:
³ ´
¡rz rzt
zB ln (1¡')¼
¡ (1 ¡ ') t ¡ ¼
s
BR (r) = ; (27)
2

and

(ps + (1 ¡ ') t) ¼
BRa (ps ) = ¡ : (28)
2zt

To prove the existence of a Nash equilibrium, we must prove that (27) and (28) intersect in the
ps ; rb) space. Substituting for ps = BRs (r) from (27) into (28) yields
(b
µ ¶
3rt ¡rz (1 ¡ ') t
+ B ln = : (29)
¼ (1 ¡ ') ¼ ¡z

The left hand side of (29) increases with r, while the right hand side is independent of r.
Moreover, as r goes to 0, the left hand side of (29) goes to ¡1 while as r goes to 1, the hand
side of the right hand side of (29) goes to 1. Hence, equation (29) has a unique solution that
we denote by rb.
By (28), it follows that in equilibrium,

2b
rzt
pbs = ¡ ¡ (1 ¡ ') t: (30)
¼
³ ´
s
Since rb is unique, so is pb . Moreover, substituting for ln ¡rz
(1¡')¼
from (29) into b̄ (b
ps ; ¹
b (b
r)) and
simplifying terms, reveals that Assumption 3 ensures that ¡ 2br¼zt ¡ (1 ¡ ') t > 0. Hence, (30)
implies that pbs > 0.
Finally, ¡ 2br¼zt ¡ (1 ¡ ') t > 0 implies that

(1 ¡ ') ¼
rb < ¡ :
2z

29
 ³ ´
¡b
rz 3b
r (1¡')
On the other hand, since ln (1¡')¼
< 0, (29) implies that ¼
+ z
> 0, or

(1 ¡ ') ¼
rb > ¡ :
3z

Next we examine how an improvement in the ad banners’ technology that increases z

towards 0 a¤ects the equilibrium prices, pro…ts, and consumers’ surplus.

Proposition 6: An increase in z leads to an increase in rb and a decreases in pbs . Moreover,

Os (b
ps ; rb) decreases while Oa (b
ps ; rb) and consumers’ surplus increase.

Proof: rb is de…ned implicitly by (29). Di¤erentiating this equation with respect to rb and z and
rearranging terms,

@b
r (1 ¡ ') t ¡ zB
= ¡ ¢ > 0: (31)
@z z 2 3t
¼
+ B
b
r

Di¤erentiating (30) with respect to pbs and z , using (31), and rearranging terms,
µ ¶
@ pbs 2t @br
= ¡ rb + z
@z ¼ @z
à (1¡')t
!
2t 3b¼rt + z
= ¡ 3t B
< 0; (32)
¼ ¼
+ rb

³ ´
3b
rt (1¡')t ¡b
rz
where the last equality follows because by (29), ¼
+ z
= ¡B ln (1¡')¼
> 0:
Di¤erentiating (26) with respect to z, using the envelope theorem, and substituting from
@ pbs
(32) for @z
:

@Oa (b
ps ; rb) rb2 t rb @ pbs
= +
@z ¼B B @z · ¸
rbt 3b
rt 2 (1 ¡ ') t
= ¡ ¢ B¡ ¡
¼B 3t ¼
+ Brb ¼ z
· ¸
rbt (1 ¡ ') t
> ¡ 3t B ¢ B ¡ > 0;
¼B ¼ + rb 2z

30
where the …rst inequality follows because by Lemma 1, rb < ¡ (1¡')¼
2z
, and the second inequality
follows because z < 0.
Di¤erentiating (25) with respect to z; using the envelope theorem, substituting for pbs
from (30), and rearranging,
³¡ ¢ ³ ¡brz ´ ´
t @b
r @b
r
@Os (b
ps ; rb) r + z @z ln (1¡')¼ + rb + z @z pbs
2b ¼
=¡ ³ ³ ´´2 :
@z ¡b
rz
zB ln (1¡')¼

Since z < 0,
µ ¶ µµ ¶ µ ¶ ¶
@Os (b
ps ; rb)s @b
r rz
¡b @b
r
sign = sign 2b
r+z ln + rb + z :
@z @z (1 ¡ ') ¼ @z
³ ´
¡b
rz @b
r
Substituting for ln (1¡')¼
from (29) and for @z
from (31) and rearranging terms,

µ ¶ "Ã !µ ¶ #
@Os (b
ps ; rb)s (1 ¡ ') t ¡ zB rt (1 ¡ ') t
3b (1 ¡ ') t ¡ zB
sign = sign 2b
r+ ¡ ¢ ¡ ¡ + rb + ¡ ¢
@z z 3t ¼
+ Brb ¼B zB z 3t¼
+ Brb
" ¶#
6b
rt (1¡')t µ
+ 3b
r t (1 ¡ ') t
= sign ¡ ¼ z
+ :
B ¼ z

The proof is completed by noting that the bracketed expression is negative since equation (29)
³ ´ ³ ´
implies that 3b¼rt + (1¡')t
z
= ¡B ln ¡rz
(1¡')¼
> 0 and 6b
rt
¼
+ (1¡')t
z
= 3b
rt
¼
¡ B ln ¡rz
(1¡')¼
> 0.
To examine the e¤ect of z on consumers, recall that consumers get an adware if ¯ ·
b̄ (b
ps ; ¹ r)) and shareware if ¯ > b̄ (b
b (b ps ; ¹
b (b
r)). The utility of each adware user in group i is
Uia (¯) while the utility of each shareware user is U s . Using equations (2), (5), (23), (8), and
(30), it follows that consumers’ surplus is given by

Xn Z b̄ (bps ;b¹(br)) a Z B
s 1 Ui (¯) Us
p ; rb) =
CS(b d¯ + d¯ (33)
i=1
n 0 B b̄ (b
ps ;b r)) B
¹(b

ps + ¹
(b b (br) (1 ¡ ') t)2
= + q ¡ pbs + U
2zB ln (1 ¡ ¹ b(b
r))
rb2 zt2
= ³ ´ + q ¡ pbs + U:
2 ¡b
rz
2¼ B ln (1¡')¼

31
Di¤erentiating (33) with respect to z,
· µ ¶µ ¶ µ ¶¸
ps ; rb)
@CS(b rbt2 rz
¡b @b
r @b
r @ pbs
= ³ ´2 ln rb + 2z ¡ rb + z ¡ :
@z ¡b
rz (1 ¡ ') ¼ @z @z @z
2¼ 2 B ln (1¡')¼

³ ´
¡b
rz @b
r
Substituting for ln (1¡')¼
from (29) and for @z
from (31) and rearranging terms

ps ; rb)
@CS(b rbt2 G (b
r)
= ¡ ¢ ³ ´;
@z 2¼ 3 z 2 3b¼rt + B 3b r
+ (1¡')
¼ z

where

r) ´ 4 (1 ¡ ')2 ¼ 2 + 22 (1 ¡ ') ¼b
G (b r2 z2 :
rz + 33b

³ ´
ps ;b
@CS(b r) r (1¡')
3b
The denominator in @z
is positive since equation (29) implies that ¼
+ z = ¡ Bt ln ¡rz
(1¡')¼
>
0. To determine the sign of G (b r), note that G (br) is convex function of rb and attains a unique
2 2
³ 2 2
´
(1¡') ¼ (1¡') ¼
minimum point at rb = 3
. Since G 3
= 66z 2 > 0, it follows that the G (b
r) > 0 for
ps ;b
all rb. Hence, @CS(b
@z
r)
> 0: ¥

Proposition 6 shows that the conclusions from Section 4 about the impact of an improve-
ment in the technology of ad banners carry over to the case where there is competition in the
software market. In particular, the technological improvement bene…ts the adware programmer
as well as consumers. Proposition 6 shows that the improvement lowers the price of shareware
and hurts the shareware programmer.
Finally, we reexamine the e¤ect of a ban on the distribution of adware on the programmers
and on consumers when there is competition in the software market.

Proposition 7: A ban on adware hurts both consumers and the programmers.

Proof: A ban on adware induces both programmers to distribute their software as shareware.
Since the software of both programmers is assumed to have the same quality, competition
in the software market drives ps to 0. Hence, consumers’ surplus is q + U. Absent a ban,
³ ´
¡b
rz
ps ; rb). Substituting for pbs from (30) and for B ln (1¡')¼
consumers’ surplus is CS(b from (29)

32
and simplifying, the di¤erence between the two expressions is

s rb2 zt2
p ; rb) =
q + U ¡ CS(b ³ ´ ¡ pbs
¡b
r z
2¼ 2 B ln (1¡')¼
tM (b r)
= ¡ ³ ´;
(1¡')
2¼ 2 z 3b
¼
r
+ z

where

M (b
r) ´ 11b rz (1 ¡ ') ¼ + 2 (1 ¡ ')2 ¼ 2 :
r2 z 2 + 10b

The sign of q + U ¡ CS(b r) since 29) implies that 3b

ps ; rb) depends on the sign of M (b ¼
r
+ (1¡')
z
=
³ ´
¡rz
¡ Bt ln (1¡')¼ r) is a convex function of rb and M (b
> 0. M (b r) = 0 has two solutions, rb1 =
¡ 0:297(1¡')¼
z
and rb2 = ¡ 0:612(1¡')¼
z
such that a ban on adware harms consumers if and only if
rb1 · rb · rb2 : Given Lemma 1, rb is indeed in this range: ¥

Propositions 7 shows that the welfare implications of a ban on adware derived in Section
4 is robust to the introduction of competition in the software market: the ban harms the
programmers as well as consumers.

6 Conclusion
We have developed a framework for studying the choice of programmers between distributing
their software online as shareware and adware Our model takes explicit account of the strategic
interaction between the programmers, advertisers, and consumers and highlights the tradeo¤
that adware users face between improved information on consumer products and the violation
of their privacy. Given this tradeo¤, consumers choose to get adware only if their privacy
concerns are small. Otherwise, consumers will either not adopt a software, when only adware is
available, or will get a shareware when there is competition in the market for software.
Our model reveals that programmers will prefer to distribute their software to consumers
for free as adware when its perceived quality is low but will prefer to sell it as shareware when its
perceived quality increases. The model also reveals that improvements in the technology of ad

33
banners will lead to less violation of privacy and will bene…t consumers. By contrast, although
a ban on adware will protect the privacy of software users, it will also prevent them from getting
targeted information about consumer products and will therefore make them worse o¤
Finally, although our model considers the market for software it can also be applicable to
media markets. In this context, our model can be used to study the choice of content providers
between selling their content to consumers for a fee or distributing it to consumers for free and
supporting it with advertising. With this interpretation in mind, our paper is related to some of
the literature on media markets. In particular, Hansen and Kyhl(2001) compare two alternatives
for …nancing the broadcasts of special events on TV (like sport events): pay-per-view and TV
commercials. They …nd that advertising leads to a higher consumers’ surplus than pay-per-view.
This result is similar to our result that bans on adware (which are akin to bans on advertising
in their model) will hurt consumers.

7 References
Acquisti Alessandro and Hal Varian (2004), “Conditioning Prices on Purchase History,” Mimeo

Calzolari Giacomo and Alessandro Pavan (2004), “On the Optimality of Privacy in Sequential
Contracting,” Mimeo

Dodds Stefan (2003), “Con…dentiality Rules and Welfare: A Dynamic Contracting Approach
with Two Principals,” Mimeo

Hann Il-Horn, Hui Kai-Lung , Lee Tom S., and Png Ivan (2002), “The Value of Online Infor-
mation Privacy: Evidence from the USA and Singapore,” Mimeo

Hann Il-Horn , Hui Kai-Lung , Lee Tom S., and Png Ivan (2003), “Direct Marketing: Privacy
and Competition,” Mimeo

Hansen Claus Thustrup and Soren Kyhl (2001), “Pay-per-view Broadcasting of Outstanding
Events: Consequences of a Ban,” International Journal of Industrial Organization, 19,
589-609.

IBM Global Services (1999), IBM Multi-National Consumer Privacy Survey.

34
Knopf J. (2000), “The Origin of Shareware,” Public Software Library, Ltd. http://www.asp-
shareware.org/users/history-of-shareware.asp

McAndrews James J. and Donald P. Morgan (2003), “Who is Calling, Please? On Anonymous
Calling and Caller ID,” Mimeo

Posner Richard. (1981), “The Economics of Privacy,” American Economic Review, Papers and
Proceedings, 71, 405-409.

Smith H. Je¤, Sandra J. Milberg and Sandra J. Burke (1996) “Information Privacy: Measuring
Individuals’ Concerns About Organizational Practices,” MIS Quarterly, 20, 167-196.

Smith E. Robert (2000), “Ben Franklin’s Web Site: Privacy and Curiosity from Plymouth Rock
to the Internet,” Privacy Journal.

Stern James. (2004), “Spyware, Adware, Malware, Thief,” Association of Shareware Profes-
sional, Inc. http://www.ftc.gov/os/comments/spyware/040318assocofsharewarepro¤.pdf

Taylor Curtis R. (2002), “Private Demands and Demands for Privacy: Dynamic Pricing and
the Market for Customer Information,” Mimeo.

Taylor Curtis R. (2004), “Privacy and Information Acquisition in Competitive Markets,”

Mimeo.

Varian Hal (1996), “Economic Aspects of Personal Privacy,” In U.S. Department. of Commerce,
Privacy and Self-Regulation in the Information Age.

Urbach R. and G. Kibel (2004), “Adware/Spyware: An Update Regarding Pending Litigation

and Legislation,” Intellectual Proprty and Technology Law Journal, 16(7), 12-16.

Wathieu Luc (2002), “Privacy, Exposure, and Price Discrimination,” HBS Marketing Research
Paper No. 02-03

35