For reasons discussed in Section 2.2, we assume all consumers face an
identical price schedule. Each consumer chooses to receive a set
that maximizes her value net of the payment 
.
To make consumer agent behavior more concrete, we follow a convenient
model of consumer preferences for information goods proposed in
[CS99]. When N items are offered, a consumer has a
strictly positive value for only a proportion 
. Suppose
these positively valued items have values distributed uniformly from
0 to 
. Consumer i ranks these items from n = 1 to 
with n = 1 being her most highly valued good. Then, (in
expectation) the value of the nth best article is given by
For convenience, we assume hereafter
that consumers value articles exactly at their expected values, given
by (1). Consumer i's surplus from reading the
most preferred articles is the aggregate value less any
payments made to the producer, 
, where 
for the article
indices 
defined over the list reordered by descending value for
consumer i.
For this paper we further limit consumer heterogeneity by assuming
that the value of the most preferred article for each consumer (which
will generally be different articles) is the same, so 
. Consumers differ in their values of 
, which are
distributed uniformly between 0 and 
. The
probability density of article values is thus 
. Someone with a higher values a greater
portion of the available items and also values each equally ranked
item at a higher level than someone with a lower . To simplify
the analysis in Section 2.2 we assume quantity is a continuous choice
variable for the consumer.
Our simulations in Section 3
respect the integer constraint.
For these consumer preferences, the socially efficient outcome would
be for each person to consume items. This would yield a
surplus of 
to each person for a total value of
for the entire population. A firm that
could observe each consumer's could perfectly price discriminate
by making a take-it-or-leave-it offer tailored to each individual and
extract this entire surplus. This serves as a baseline for the
maximum profit that a monopolist could earn.
Elsewhere, we studied the dynamics of an agent market in which consumers are incompletely informed and thus try to learn the parameters of the article value distribution [KDMM99]. In future work we will introduce multiple producers, so that consumers dealing with one producer may be able to learn, at a cost, whether other producers have better offerings.
