Full text of Working Papers (Federal Reserve Bank of Richmond) : Large and Small Sellers : A Theory of Equilibrium Price Dispersion with Sequential Search, Working Paper 14-08

The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.

Working Paper Series

Large and Small Sellers: A Theory of
Equilibrium Price Dispersion with
Sequential Search

WP 14-08

Guido Menzio
University of Pennsylvania and NBER
Nicholas Trachter
Federal Reserve Bank of Richmond

This paper can be downloaded without charge from:
http://www.richmondfed.org/publications/

Large and Small Sellers:
A Theory of Equilibrium Price Dispersion
with Sequential Search
Guido Menzio

Nicholas Trachter

University of Pennsylvania and NBER

Federal Reserve Bank of Richmond

March 2014
Working Paper No. 14-08

Abstract
The paper studies equilibrium pricing in a product market for an indivisible
good where buyers search for sellers. Buyers search sequentially for sellers but do
not meet every seller with the same probability. Speci…cally, a fraction of the buyers’
meetings lead to one particular large seller, while the remaining meetings lead to
one of a continuum of small sellers. In this environment, the small sellers would like
to set a price that makes the buyers indi¤erent between purchasing the good and
searching for another seller. The large seller would like to price the small sellers
out of the market by posting a price that is low enough to induce buyers not to
purchase from the small sellers. These incentives give rise to a game of cat and
mouse, whose only equilibrium involves mixed strategies for both the large and the
small sellers. The fact that the small sellers play mixed strategies implies that there
is price dispersion. The fact that the large seller plays mixed strategies implies that
prices and allocations vary over time. We show that the fraction of the gains from
trade accruing to the buyers is positive and nonmonotonic in the degree of market
power of the large seller. As long as the large seller has some positive but incomplete
market power, the fraction of the gains from trade accruing to the buyers depends
in a natural way on the extent of search frictions.
JEL Codes: D21, D43.
Keywords: Imperfect competition, Search frictions, Price dispersion.
Menzio: Department of Economics, University of Pennsylvania, 3718 Locust Walk, Philadelphia,
PA 19013 (email: gmenzio@sas.upenn.edu). Trachter: Federal Reserve Bank of Richmond, 701 E. Byrd
Street, Richmond, VA 23219 (email: nicholas.trachter@rich.frb.org). The authors are grateful to Giuseppe
Moscarini for several discussions that motivated them to write this paper. The views expressed in this
paper are those of the authors and do not necessarily represent the views of the Federal Reserve Bank of
Richmond or the Federal Reserve System.

1

1

Introduction

We propose a novel theory of equilibrium price dispersion in product markets with search
frictions. As in Diamond (1971), buyers search for sellers sequentially. In contrast to
Diamond (1971), buyers do not meet all sellers with the same probability. Speci…cally, a
fraction of the buyers’meetings leads to one particular large seller, while the remaining
meetings leads to one of a continuum of small sellers. We prove that the unique equilibrium
of this model is such that sellers post a nondegenerate distribution of prices and buyers
capture a positive fraction of the gains from trade. The fraction of gains from trade
accruing to the buyers is hump-shaped with respect to the market power of the large
seller. However, for any degree of market power of the large seller, the fraction of gains
from trade accruing to the buyers converges to one when search frictions vanish, and to
zero when search frictions become arbitrarily large.
In a famous paper, Diamond (1971) analyzes a product market where buyers search
sequentially for sellers. He …nds that the equilibrium of this market is such that all sellers
post the monopoly price and buyers capture none of the gains from trade. Moreover, he
…nds that this is the only equilibrium independently of the extent of search frictions. This
result, which is popularly known as the Diamond Paradox, is problematic for several reasons. From the empirical point of view, the result ‡ies against the evidence documenting
the existence of a great deal of price dispersion for identical goods (see, e.g., Sorensen
2000 or Kaplan and Menzio 2014). From the theoretical point of view, the result implies
a puzzling discontinuity in outcomes since, without search frictions, all sellers post the
competitive price and buyers capture all of the gains from trade. Yet the logic behind
the Diamond Paradox is rather strong. Every seller wants to post the buyer’s reservation price, i.e., the price that makes a buyer indi¤erent between purchasing the good and
searching for another seller. But if every seller posts the same price, the option value of
searching for another seller is zero and the buyer’s reservation price must be equal to the
monopoly price.
In a series of closely related papers, Butters (1977), Varian (1980), and Burdett and
Judd (1983) identify an alternative search process that leads to equilibrium price dispersion and, in turn, to a resolution of the Diamond Paradox. In particular, they consider
a search market where some buyers contact multiple sellers simultaneously, while other
buyers contact one seller at a time. They …nd that, in equilibrium, sellers post di¤erent
prices and buyers capture a positive fraction of the gains from trade. The intuition behind
2

their result is clear. An equilibrium in which all sellers post the monopoly price (or any
common price above the cost of production) cannot exist, as an individual seller would
have an incentive to post a slightly lower price and trade not only with those buyers who
have only contacted him, but also with those buyers who have contacted him, as well
as other sellers. This undercutting process cannot lead to an equilibrium in which all
sellers post a price equal to the cost of production, as an individual seller can always
attain a strictly positive pro…t by charging a slightly higher price and trading only with
buyers who have failed to contact multiple sellers. Hence, the equilibrium must involve a
distribution of prices and buyers capture some of the gains from trade.
The assumption that some buyers contact multiple sellers simultaneously is quite
strong. The assumption does not mean that there are some buyers that can freely recall
previously contacted sellers. Indeed, even if some buyers could freely recall sellers, the
only equilibrium would be such that all sellers post the monopoly price and buyers capture
none of the gains from trade. The assumption really means that there are some buyers
that come into contact with multiple sellers before being able to decide whether to stop
searching. This observation motivates our paper, which advances a theory of equilibrium
price dispersion in markets where search is genuinely sequential, in the sense that buyers
have the option to stop searching after meeting any individual seller.
We consider a product market populated by buyers— each demanding one unit of an
indivisible good— and sellers— each producing the good at the same cost. At the beginning
of each day of trading, sellers post prices. Then buyers observe the price distribution and
start searching for sellers. As in Diamond (1971), buyers search for sellers sequentially.
However, in contrast to Diamond (1971), buyers do not meet all sellers with the same
probability. Speci…cally, a fraction of the buyers’meetings leads to one particular large
seller, while the remaining meetings lead to one of a continuum of small sellers. We
prove that an equilibrium in this market exists and is unique. In equilibrium, the large
seller randomizes over his price from a distribution whose support is a closed, convex,
non-empty subset of the interval between the sellers’ cost and the buyers’ valuation of
the good. The small sellers post a distribution of prices whose support is also a closed,
convex and non-empty subset of the interval between cost and valuation of the good. The
buyers capture a positive fraction of the gains from trade, which varies depending on the
realization of the large seller’s price.
The intuition behind these results is straightforward. The small sellers would like to
3

set a price that makes the buyers indi¤erent between purchasing the good and searching
for another seller. The large seller would like to price the small sellers out of the market by
posting a price that is low enough to induce buyers not to purchase from the small sellers.
These incentives give rise to a game of cat and mouse, whose only equilibrium involves
mixed strategies for both the large and the small sellers. The fact that the small sellers
play mixed strategies implies that there is price dispersion in equilibrium and, hence, the
buyers capture a strictly positive fraction of the gains from trade. The fact that the large
seller plays mixed strategies implies that prices and allocations vary over time.
The equilibrium outcomes in our model depend critically on two parameters: the
market power of the large seller— as measured by the fraction of meetings that involve
the large seller— and the extent of search frictions— as measured by the rate at which
buyers meet sellers. We …nd that the competitiveness of the market is nonmonotonic
with respect to the market power of the large seller. When the market power of the
large seller vanishes, the environment converges to the one studied by Diamond (1971)
and the equilibrium outcomes converges to those of a pure monopoly. In particular, the
distributions of prices posted by the large and the small sellers converge to the monopoly
price and the fraction of the gains from trade accruing to the buyers goes to zero. Similarly,
when the large seller has complete market power, the equilibrium outcomes converge to
those of a pure monopoly. However, as long as the market power of the large seller is
positive but …nite, there is price dispersion and buyers capture a positive fraction of the
gains from trade. Thus, the Diamond Paradox can be viewed as the limit of a paradoxically
nonmonotonic relationship between the market power of the large seller and the extent of
competition.
Whenever the market power of the large seller is positive and incomplete, the competitiveness of the market depends in a natural way on the extent of search frictions. When
the search frictions become arbitrarily small, the equilibrium outcomes converge to those
of a perfectly competitive market. In particular, the distributions of prices posted by the
large and the small sellers converge to the competitive price and the fraction of the gains
from trade accruing to the buyers converges to one. In contrast, when the search frictions
become arbitrarily large, the equilibrium outcomes converge to those of a pure monopoly.
Thus, whenever the market power of the large seller is positive and incomplete, the discontinuity in equilibrium outcomes highlighted by Diamond (1971) disappears: When search
frictions vanish, the equilibrium becomes competitive.

4

Our paper adds to the existing theories of equilibrium price dispersion in product
markets with search frictions. As mentioned above, Butters (1977), Varian (1982), and
Burdett and Judd (1983) generate price dispersion in the context of a model where search
is sometimes sequential and sometimes simultaneous. In contrast, we obtain price dispersion in a model where search is genuinely sequential. Albrecht and Axell (1984) obtain
price dispersion in a model of sequential search by introducing heterogeneity in buyers’
valuations. However, their model is such that the Diamond Paradox holds for the buyers
with the lowest valuation. Hence, in the presence of any entry cost, the lowest valuation
buyers would stay out of the market and price dispersion would unravel. In contrast, our
theory of price dispersion is robust to the introduction of entry costs. Curtis and Wright
(2004) and Gaumont, Schindler and Wright (2006) obtain price dispersion in a model of
sequential search by introducing heterogeneity in the gains from trade between buyers
and sellers. In contrast, our theory of price dispersion does not require such heterogeneity. Finally, Benabou (1989) obtains price dispersion in a model of sequential search by
introducing in‡ation and costs to adjust nominal prices. In contrast, our theory of price
dispersion does not rely on nominal rigidities.
There are also several theories that resolve the Diamond Paradox without generating
price dispersion. Pissarides (1984) and Mortensen and Pissarides (1994) show that, if
prices are determined as the outcome of a bargaining game between buyers and sellers
rather than being posted by sellers, then buyers will generally capture a positive fraction
of the gains from trade. Montgomery (1991), Shimer (1996), Moen (1997), and Burdett,
Shi and Wright (2001) show that, if buyers can direct their search toward particular
sellers, then sellers will compete for searchers and buyers will capture a positive fraction
of the gains from trade. Carrillo-Tudela, Menzio and Smith (2011) show that, if buyers
can recall past sellers and sellers can distinguish between buyers with single and multiple
contacts, then buyers extract some of the gains from trade in equilibrium.

2

Environment and de…nition of equilibrium

In this section, we describe the market for an indivisible good where buyers search sequentially for sellers and face a positive probability of meeting one particular seller. We
then state the problem of the buyers, the problem of the large seller, and the problem of
the small sellers in this market. Finally, we de…ne an equilibrium.

5

2.1

Environment

The market for an indivisible good opens daily. The market is populated by a continuum
of buyers with measure b > 0. Each buyer demands one unit of the good every day.1 The
market is also populated by one large seller and a continuum of small sellers with measure
one. Each seller produces the good at the same, constant unit cost.2
At the beginning of each day, every seller posts simultaneously and independently a
price for the good. We denote as x the price of the large seller and as y the price of a
small seller. We also denote as F (x) the probability that the large seller posts a price
smaller or equal to x and as F the support of the price distribution F .3 Similarly, we

denote as G(y) the probability that a small seller posts a price smaller or equal to y and

as G the support of the price distribution G. Both F and G are endogenous objects. We

assume that sellers cannot change their price during the day but can freely change their
price from one day to the next.4
After observing the distribution of posted prices, every buyer enters the market and
starts searching for sellers.5 A searching buyer meets a seller at the Poisson rate , where
> 0 is a parameter that controls the extent of search frictions in the market. Conditional
on meeting a seller, the buyer meets the large seller with probability
continuum of small sellers with probability 1

, where

and one of the

2 (0; 1) is a parameter that

controls the extent of market power of the large seller. A searching buyer is forced out
of the market at the Poisson rate , where

> 0 is a parameter that controls the cost of

searching.
Upon meeting a seller, the buyer decides whether to purchase the good. If the buyer
purchases the good at the price p, he enjoys a utility of u

p and exits the market for

the day, where u > 0 is the buyer’s valuation of the good. If the seller sells the good
1

Equivalently, one can think that the market is visited by a di¤erent group of buyers every day.
Our theory of price dispersion generalizes to the case in which there is more than one large seller, as
well as to the case in which all sellers are large. We choose to focus on the case in which there is one
large seller and a continuum of small sellers only for the sake of exposition.
3
Our theory of price dispersion would generalize to the case in which sellers can also adjust their prices
during the day, but only every once in a while.
4
The support of a cumulative distribution function is de…ned as the smallest closed set whose complement has probability zero.
5
We assume that buyers observe the distribution of prices in order to side-step issues of learning while
searching, which are notoriously di¢ cult and have been tackled only by a handful of papers (see, e.g.,
Rothschild 1974, and Burdett and Wishwanath 1988). Moreover, the assumption is not unreasonable.
For instance, buyers may learn about the price distribution on the Internet before leaving their home in
the morning and, later in the day, they may meet a sequence of sellers on the way to and from work.
2

6

at the price p, he enjoys a pro…t of p

c and keeps searching for additional customers,

where c 2 [0; u) is the seller’s cost of producing the good. If the buyer is forced to exit the

market before purchasing the good, he enjoys a utility of 0. In order to keep the problem
of the buyer stationary, we assume that every day is in…nitely long. We also assume that
buyers and sellers do not discount their payo¤s during the day.
Before de…ning an equilibrium, a few observations about the environment are in order.
First, notice that search is sequential, in the sense that a buyer has the option to purchase
the good and stop searching after every meeting with an individual seller. Thus, in
contrast to Butters (1977), Varian (1980), and Burdett and Judd (1983), our theory of
price dispersion does not hinge on the assumption of simultaneous search. Second, notice
that all buyers are identical, in the sense that they have the same valuation of the good.
Thus, in contrast to Albrecht and Axell (1984), our theory of price dispersion does not
rely on the assumption of buyer heterogeneity. Third, notice that all buyer-seller matches
are identical, in the sense that they all involve the same gains from trade. Thus, in
contrast to Curtis and Wright (2004) and Gaumont, Schindler and Wright (2006), our
theory of price dispersion does not rely on the assumption of heterogeneity in the gains
from trade across di¤erent buyer-seller meetings. Indeed, the only substantive di¤erence
between our environment and the one analyzed by Diamond (1971) is the existence of a
large seller, i.e., a seller that is contacted by buyers with positive probability and, hence,
whose pricing decision impacts the buyers’value of searching the market.

2.2

Individual problems and de…nition of equilibrium

We restrict attention to equilibria where the acceptance strategy of the buyers, the pricing
strategy of the large seller, F (x), and the pricing strategy of the small sellers, G(y),
are all independent of the history of prices and trades in previous days.6 Given this
restriction, the price of the large seller today does not a¤ect the seller’s expected pro…t
in the subsequent days. Thus, the large seller chooses today’s price so as to maximize
today’s expected pro…t. Similarly, the price of a small seller today does not a¤ect the
seller’s expected pro…t in the subsequent days. Thus, the small seller chooses today’s
6
As we shall prove in Section 3, there exists a unique history independent equilibrium. Therefore,
if the market opens on a …nite number of days, the history independent equilibrium is also the only
equilibrium on the last day in which the market is open. By induction, this implies that the history
independent equilibrium is the only equilibrium on any previous day as well. We have not ruled out the
existence of a non-history independent equilibrium when the market remains open forever.

7

price so as to maximize today’s expected pro…t. Finally, the buyer’s purchasing decision
today has no e¤ect on his expected utility in subsequent days. Hence, the buyer chooses
to accept or reject a price so as to maximize today’s expected utility.
Problem of the buyer. Let V (x) denote today’s expected utility for a buyer who is
searching, when the price posted by the large seller is x. The buyer’s expected utility
V (x) is such that
max fu

V (x) =

x

V (x); 0g + (1

)

Z

max fu

y

V (x); 0g dG(y):

(1)

The left-hand side of (1) is the buyer’s annuitized value of searching. The …rst term on the
right-hand side of (1) is the rate at which the buyer meets the large seller,
value of the option of purchasing the good from the large seller, max fv

, times the
x

V (x); 0g.

The second term on the right-hand side of (1) is the rate at which the buyer meets a small
seller, (1
small seller,
of searching.

R

), times the expected value of the option of purchasing the good from a
max fv

y

V (x); 0g dG(y). We shall refer to V (x) as the buyer’s value

Let R(x) denote the price that makes the buyer indi¤erent between purchasing the
good and continuing his search, when the price posted by the large seller is x. Formally,
R(x) is such that
R(x) = u

(2)

V (x):

We shall refer to R(x) as the buyer’s reservation price. Clearly, the buyer …nds it optimal
to purchase the good whenever he meets a seller posting a price strictly smaller than the
reservation price. Similarly, the buyer …nds it optimal to continue searching whenever he
meets a seller posting a price strictly greater than the reservation price. We assume that
the buyer purchases the good whenever he meets a seller charging the reservation price.
The buyer’s optimal purchasing strategy described above determines the evolution of
the population of buyers during the day, as well as the overall number of meetings taking
place between buyers and sellers in the day. The measure b(t; x) of buyers who are still
searching after t units of time since the beginning of the day is given by
b(t; x) = b exp f [ +
where

R(x)] + (1

1[x

)G(R(x))] tg ;

is the rate at which buyers are forced out of the market,

8

1[x

(3)

R(x)] is the rate

at which buyers meet the large seller and purchase the good7 and (1

)G(R(x)) is the

rate at which buyers meet a small seller and purchase the good. Similarly, the measure
m(x) of meetings that take place between buyers and sellers throughout the day is given
by
m(x) =

+

1[x

b
R(x)] + (1

)G(R(x))

(4)

:

As suggested by the notation, the price x posted by the large seller a¤ects the buyer’s
value of searching, V (x), the buyer’s reservation wage, R(x), the measure of searching
buyers, b(t; x), and the overall measure of meetings between buyers and sellers, m(x).
Intuitively, when the large seller posts a higher price, buyers have less to gain from
searching and, hence, they are willing to purchase the good at higher prices from the
small sellers. This implies that, when the large seller posts a higher price, buyers exit the
product market more quickly and, hence, they meet fewer sellers.
Problem of the large seller. Let L(x) denote the pro…t expected by the large seller
today if he posts a price of x. The pro…t L(x) is given by
L(x) = m(x) 1[x

R(x)](x

(5)

c):

The expression in (5) is easy to understand. The seller is contacted by a measure m(x)
of buyers. Each one of these buyers purchases the good with probability 1[x
Each unit of the good purchased by the buyers gives the seller a pro…t of x

R(x)].

c.

The large seller chooses the price x so as to maximize his expected pro…t L(x). Thus,
the large seller …nds it optimal to follow the mixed pricing strategy F (x) if and only if
L(x) = L for all x 2 F and L(x)

L for all x 2
= F, where F denotes the support of

the price distribution F and L denotes the maximum of L with respect to x.

Notice that, when the price x is greater than the buyer’s reservation price R(x), the
pro…t of the large seller L(x) is always equal to zero. When x is smaller than R(x), the
price has two countervailing e¤ects on the pro…t of the large seller. On the one hand, a
higher x increases the pro…t that the large seller enjoys every time it trades with a buyer.
On the other hand, a higher x tends to increase the buyer’s reservation price. In turn,
this tends to increase the fraction of buyers who purchase the good from the small sellers
and, consequently, it reduces the fraction of buyers who contact and purchase from the
7

Throughout the paper, we denote as 1[f (x) 0] the function that takes the value 1 if the condition
in square bracket is satis…ed, and the value 0 otherwise.

9

large seller. Because of these two countervailing e¤ects of x on L(x), it may be the case
that the large seller’s pro…t attains its maximum over a non-degenerate interval of prices
and, hence, it may be the case that the optimal strategy of the large seller is mixed.
Problem of the small seller. Let S(y) denote the pro…t expected by the small seller
today if he posts a price of y. The pro…t S(y) is given by
Z
S(y) = m(x)(1
)1[y R(x)](y c)dF (x):

(6)

The expression in (6) is easy to understand. The price posted by the large seller is drawn
from the cumulative distribution function F (x). Conditional on the large seller posting
the price x, the small seller is contacted by m(x)(1
purchases the good with probability 1[y
the buyers gives the seller a pro…t of y

) buyers. Each one of these buyers

R(x)]. Each unit of the good purchased by
c.

The small seller chooses the price y so as to maximize his expected pro…t S(y). Thus,
the small seller …nds it optimal to follow the mixed pricing strategy G(y) if and only if
S(y) = S for all y 2 G and S(y)

S for all y 2
= G, where G denotes the support of the

price distribution G and S denotes the maximum of S(y) with respect to y.

Notice that the price y has two countervailing e¤ects on the pro…t of the small seller
S(y). On the one hand, a higher y increases the pro…t that the small seller enjoys every
time it trades with a buyer. On the other hand, a higher y increases the probability
that the price posted by the large seller is such that the buyer’s reservation price R(x) is
lower than y. In turn, this implies that a higher y reduces the expected number of buyers
who purchase from the small seller. Because of these two countervailing e¤ects of y on
S(y), it may be the case that the pro…t of the small seller attains its maximum over a
nondegenerate interval of prices and, hence, it may be the case that the optimal strategy
of the small seller is mixed.
De…nition of equilibrium. We are now in the position to de…ne an equilibrium for our
frictional product market.
De…nition 1. A history independent equilibrium is a tuple (R; F; G) such that: (i) the
reservation price function R is such that R(x) = v

V (x) for all x; (ii) the cumulative

distribution function F is such that L(x) = L for all x 2 F and L(x)

L for all

x2
= F ; (iii) the cumulative distribution function G is such that S(y) = S for all y 2 G
and S(y)

S for all y 2
= G.

10

3

Characterization of equilibrium

In this section, we establish the existence and uniqueness of the equilibrium and we characterize its main features. First, we establish some properties of the buyer’s reservation
price, R, of the large seller’s mixing distribution, F , and of the small sellers’price distribution, G, that must hold in any equilibrium. These properties imply that any equilibrium
will feature price dispersion— in the sense that small sellers post di¤erent prices— and
price variation— in the sense that the large seller posts di¤erent prices on di¤erent days.
Second, we show that an equilibrium always exist and is unique. Finally, we analyze
the e¤ect on equilibrium outcomes of changes in the degree of market power of the large
seller— as measured by the parameter — and in the extent of search frictions— as measured by the parameter . We …nd that the fraction of the gains from trade accruing to
the buyers is nonmonotonic in the degree of market power of the large seller. However,
given any degree of market power of the large seller, the fraction of the gains from trade
accruing to the buyers depends in the natural way on the extent of search frictions. These
…ndings allow us to better understand the puzzling results in Diamond (1971).

3.1

Equilibrium price dispersion and variation

We begin the characterization of the equilibrium through a series of lemmas. First, we
establish some general properties of the buyer’s reservation price. Second, we use the
properties of the buyer’s reservation price in order to establish that all sellers attain
strictly positive pro…ts. Third, we use the properties of the buyer’s reservation price to
locate the boundaries of the support of the mixing distribution of the large seller and of
the price distribution of the small sellers. Fourth, we prove that the price distribution of
the small sellers does not contain any mass points or gaps. This result implies that any
equilibrium features price dispersion. Finally, we show that the mixing distribution of the
large seller is nondegenerate with a mass point at the top. This result implies that any
equilibrium features time-variation in prices.
The …rst lemma uses the equilibrium condition (i) to establish some general properties
of the buyers’reservation price R.
Lemma 1: (Reservation price). Let (R; F; G) be an equilibrium. (i) The reservation price
R(x) is continuous and such that R(x) > c and R(x)

u for all x 2 [c; u]; (ii) There

exists a unique x 2 (c; u] such that R(x ) = x , R(x) > x for all x 2 [c; x ) and R(x) < x
11

for all x 2 (x ; u]; (iii) For any x1 and x2 such that c

and R(x2 )
R(x2 )

R(x1 ) < x2

x1 < x2

x , R(x2 )

x1 . For any x1 and x2 such that x

R(x1 ) > 0

x1 < x 2

u,

R(x1 ) = 0.

Proof: In Appendix A.
The properties of the buyer’s reservation price are intuitive. First, it is clear that the
buyer’s reservation price, R(x), is strictly greater than the seller’s marginal cost c and
smaller than the buyer’s valuation u. To see why this is the case, notice that none of the
sellers post prices below the marginal cost c, as this would give them strictly negative
pro…ts. Therefore, buyers only …nd sellers posting prices greater or equal to c. Since
…nding sellers is time consuming, the buyer’s value of searching V (x) is strictly smaller
than u

c and, in turn, the reservation price is strictly greater than c. Inspection of

(1), immediately reveals that the buyer’s value of searching is positive and, in turn, the
reservation price is smaller than u.
Second, it is clear that the price posted by the buyer’s reservation price, R(x), is
increasing in the price posted by the large seller, x, and that it is greater than x when
x is relatively low and smaller than x when x is relatively high. If the large seller posts
a price x equal to the marginal cost c, R(x) is greater than x because R(x) > c. If the
large seller increases his price x above c, the buyer’s value of searching V (x) declines
and, consequently, the buyer’s reservation price R(x) increases. However, since buyers
encounter the large seller only in a fraction of their meetings, V (x) declines less than
one-for-one with x and, consequently, R(x) increases less than one-for-one with x. In
turn, this implies that there exists a price x such that, if x greater than x , the buyer’s
reservation price is lower than the price posted by the large seller. As we shall see, the
price x plays a key role in the characterization of the equilibrium.
The …rst part of Lemma 1 states that the buyer’s reservation price is always strictly
greater than the seller’s marginal cost. Given this feature of the buyer’s optimal strategy,
sellers can always achieve strictly positive pro…ts. This is formally established by the next
lemma.
Lemma 2: (Equilibrium pro…t). Let (R; F; G) be an equilibrium. The maximized pro…ts
of the large and small sellers are strictly positive, i.e. L > 0 and S > 0.
Proof: To prove L > 0, it is su¢ cient to note that the large seller can achieve a strictly
positive pro…t by posting the price x . In fact, by posting x , the large seller can trade
12

with all the buyers it meets— as R(x ) = x — and it can attain a strictly positive pro…t
on each trade— as x > c. Similarly, to prove S > 0, it is su¢ cient to note that the small
seller can achieve a strictly positive pro…t by posting the price R(c). In fact, by posting
R(c), the small seller can trade with all the buyers it meets— as R(x)

R(c) for all x on

the support of F — and it can make a strictly positive pro…t on each trade— as R(c) > c.

The next lemma uses the results in Lemma 1 and in Lemma 2 in order to locate the
boundaries of the support of the price distributions F and G.
Lemma 3: (Boundaries of F and G). Let (R; F; G) be an equilibrium. Denote as x and
x the lowest and the highest price in F . Similarly, denote as y and y the lowest and the

highest price in G. Then: (i) x = x , (ii) y = R(x); (iii) y = R(x).
Proof: In Appendix B.
The …rst part of Lemma 3 states that the highest price posted by the large seller, x,
must be equal to x . To see why this is the case, it is su¢ cient to show that, if x were
di¤erent from x , the pro…t of the large seller would not be maximized at x and, hence,
part (ii) in the de…nition of equilibrium would be violated.
First, suppose x > x . In this case, the large seller attains a pro…t of zero by posting
the price x because R(x) < x. In contrast, the large seller can attain a strictly positive
pro…t by posting the price x because R(x ) = x and x > c. Thus, x

x . Now, suppose

x < x . Notice that a small seller never posts a price y strictly greater than R(x), as
doing so means being priced out of the market by the large seller with probability one
and, consequently, making no pro…t. Hence, the distribution of prices among small sellers
is such that G(R(x)) = 1. In turn, G(R(x)) = 1 implies that the large seller does not
maximize his pro…t by posting the price x. Indeed, by posting the price x rather than
the price x, the large seller makes the same number of trades as both prices are below
the buyers’reservation price— R(x )

x and R(x) < x— and at both prices none of the

small sellers is priced out of the market— G(R(x )) = G(R(x)) = 1. On the other hand,
by posting the price x rather than the price x, the large seller enjoys a higher pro…t per
trade. Thus x

x . Overall, any equilibrium must be such that x = x .

The second part of Lemma 3 states the highest price posted by the small seller, y,
must be equal to R(x). To see why this is the case, it is su¢ cient to show that, if y were

13

di¤erent from R(x), the pro…t of the small seller would not be maximized at y and, hence,
part (iii) in the de…nition of equilibrium would be violated.
We have already argued that a small seller never posts a price strictly greater than
R(x). Thus, y

R(x). Now, suppose y < R(x). If this is the case, the pro…t of the large

seller is strictly increasing in x over the interval between R 1 (y) and x. For any price in
the interval [R 1 (y); x], the large seller makes the same number of trades as any such price
is below the buyers’reservation price— i.e., R(x)

x— and at any such price none of the

small sellers is priced out of the market— G(R(x)) = 1. However, for higher prices in the
interval [R 1 (y); x], the large seller enjoys a higher pro…t per trade. This implies that the
large seller does not post any price x 2 [R 1 (y); x), i.e. F (R 1 (y) ) = F (x ) 8 . In turn,

the fact that F (R 1 (y) ) = F (x ) implies that the small seller can attain a higher pro…t

by posting the price R(x) rather than the price y. In fact, by posting R(x) rather than
y, the small seller makes the same number of trades, as both prices are lower than the
buyers’reservation price with probability F (R 1 (y) ). However, by posting R(x) rather
than y, the small seller enjoys a higher pro…t per trade. Thus, y

R(x). Overall, we

have shown that any equilibrium must be such that y = R(x).
The last part of Lemma 3 states that the lowest price posted by the seller, y, must be
equal to R(x). The proof of this result is also intuitive. First, suppose y < R(x). In this
case, the small seller can attain a higher pro…t by posting the price R(x) rather than y.
By posting R(x) rather than y, the small seller makes the same number of trades as both
prices are below the buyers’reservation price with probability one. However, by posting
R(x) rather than y, the small seller would enjoy a higher pro…t per trade. Therefore, the
equilibrium condition (iii) is violated. Now, suppose y > R(x). In this case, the large
seller could attain a higher pro…t by posting the price R 1 (y)

rather than the price

x, for some small > 0. By posting either price, the large seller prices out of the market
all the small sellers and, hence, enjoys the same number of trades. However, by posting
R 1 (y)

, the large seller enjoys a larger pro…t per trade. Therefore, the equilibrium

condition (ii) is violated. Overall, we have established that the equilibrium conditions (ii)
and (iii) can only be ful…lled if y = R(x).
The next lemma establishes some additional properties of the distribution of prices
among the small sellers.
8

Throughout the paper, we denote limx!x f (x) as f (x ) and limx!x+ f (x) as f (x+).
0

0

14

Lemma 4: (Properties of G) Let (R; F; G) be an equilibrium. (i) The support of the
distribution G is some interval [y; y], with c < y < y

u. (ii) The distribution G has no

mass points.
Proof: In Appendix C.
The …rst part of Lemma 4 states that there are no gaps in the price distribution of the
small sellers. This intuition behind this result is simple. Suppose that G(y1 ) = G(y2 )
for some y1 , y2 2 G and y1 < y2 . If this is the case, the pro…t of the large seller is strictly

increasing in x over the interval [R 1 (y1 ); R 1 (y2 )). For any price x 2 [R 1 (y1 ); R 1 (y2 )),

the large seller makes the same number of trades, as it prices out of the market a fraction
of 1

G(y1 ) of small sellers. However, for higher prices in [R 1 (y1 ); R 1 (y2 )), the large

seller enjoys a higher pro…t per trade. Therefore, the large seller does not …nd it optimal
to post any price x 2 [R 1 (y1 ); R 1 (y2 )), i.e. F (R 1 (y1 ) ) = F (R 1 (y2 ) ). In turn, the

fact that F (R 1 (y1 ) ) = F (R 1 (y2 ) ) implies that the small seller can attain a higher
pro…t by posting the price y2 rather than the price y1 . By posting y2 rather than y1 , the
small seller makes the same number of trades as, for both prices, the probability of being
priced out of the market is equal to F (R 1 (y1 ) ). However, by posting y2 rather than y1 ,
the small seller enjoys a higher pro…t per trade. Therefore, y1 cannot be on the support
of the distribution G.
The second part of Lemma 4 states that there are no mass points in the price distribution of the small sellers. To see why, suppose there is a mass point at some price y1 < y.

If this is the case, the pro…t of the large seller is such that L(R 1 (y1 ) ) > L(R 1 (y1 )).
Indeed, the large seller makes a discretely larger number of trades at a price x in…nitesimally smaller than R 1 (y1 ), as this price cuts out of the market the mass of small
sellers posting the price y1 . The fact that L(R 1 (y1 ) ) > L(R 1 (y1 )) implies the large
seller does not …nd it optimal to post any price in some interval [R 1 (y1 ); R 1 (y2 )], i.e.
F (R 1 (y1 ) ) = F (R 1 (y2 ) ). In turn, the fact that F (R 1 (y1 ) ) = F (R 1 (y2 ) ) implies that the pro…t of the small seller is strictly increasing in y over the interval [y1 ; y2 ).
Indeed, for any y 2 [y1 ; y2 ), the small seller makes the same number of trades as the
probability of being priced out of the market is always equal to F (R 1 (y1 ) ). However,

for higher y’s in [y1 ; y2 ), the small seller enjoys a higher pro…t per trade. Therefore, the
pro…t of the small seller does not attain its maximum at the price y1 and, consequently,
the distribution G cannot have a mass point at y1 < y. A similar argument can then be
used to rule out the possibility of a mass point at y1 = y.
15

Having ruled out the existence of gaps and mass points, we can conclude that the
support of the price distribution G must be some nondegenerate interval [y; y]. Moreover,
notice that the lowest price on the distribution, y, must be strictly greater than the seller’s
marginal cost c, as y = R(x), x

c and R(x) > c for all x

c. Similarly, notice that

the highest price on the distribution, y, must be smaller than the buyer’s valuation u, as
y = R(x), R(x) = x and x

u.

The next lemma establishes some additional properties of the distribution from which
the large seller draws his price.
Lemma 5: (Properties of F ) Let (R; F; G) be an equilibrium. (i) The support of the
distribution F is some interval [x; x], with c < x < x

u. (ii) The distribution F has

one and only one mass point at the price x.
Proof: In Appendix D.
The …rst part of Lemma 5 states that the distribution F has no gaps. The second
part of Lemma 5 states that the distribution F has one (and only one) mass point at the
price x. It is easy to understand why this is the case. First, suppose F does not have a
mass point at x. In this case, the pro…t of the small seller does not attain its maximized
value at the price y and, hence, the equilibrium condition (iii) is violated. In fact, if the
small seller posts the price y, he is priced out of the market with probability one and,
hence, it attains a pro…t of zero. In contrast, if the small seller posts the price R(c), he
can attain a strictly positive pro…t. Next, suppose that F has a mass point at some price
x1 < x. In this case, the pro…t of the small seller is discontinuous at R(x1 ). Speci…cally,
S(R(x1 )+) < S(R(x1 )) since the small seller would make a discretely larger number of
sales by posting a price equal to R(x1 ) rather than by posting any price strictly greater
than R(x1 ). In turn, S(R(x1 )+) < S(R(x1 )) implies that the small seller never posts a
price in some interval (R(x1 ); R(x2 )). However, this is not consistent with equilibrium
because the distribution of prices among small sellers cannot have gaps.
The above observations imply that the support of the distribution F is some interval
[x; x]. The lower bound of the support, x, is strictly greater than the sellers’ marginal
cost c because of search frictions. The upper bound of the support, x, is strictly smaller
than the buyers’valuation u because of price dispersion. Moreover, the upper bound is
strictly greater than the lower bound because, as established in Lemma 3, x = R 1 (y)
and x = R 1 (y) and, as established in Lemma 4, y is strictly greater than y. Therefore,
the distribution from which the large seller draws his price is nondegenerate.
16

The results in Lemmas 4 and 5 already o¤er a rather sharp characterization of equilibrium. Lemma 4 guarantees that, in any equilibrium, there will be variation in the prices
posted by di¤erent sellers for the same good in the same period of time and, hence, the
Diamond paradox will not hold. Lemma 5 guarantees that, in any equilibrium, there will
be variation over time in the large seller’s price and, consequently, in the buyers’reservation price, in the number of transactions, and in the distribution of prices across di¤erent
transactions. We have thus established the following theorem.
Theorem 1: (Price dispersion and price variation). In any equilibrium, the distribution
of prices among sellers is nondegenerate and varies over time.
The intuition behind Theorem 1 is straightforward. The small sellers would like to set
a price that makes the buyers indi¤erent between purchasing the good and searching for
another seller. The large seller would like to price the small sellers out of the market by
posting a price that is low enough to induce buyers not to purchase from the small sellers.
These incentives give rise to a game of cat and mouse, whose only equilibrium involves
mixed strategies for both the large and the small sellers. The fact that the small sellers
play mixed strategies implies that there is price dispersion in equilibrium. The fact that
the large seller plays mixed strategies implies that prices and allocations vary over time.

3.2

Existence and uniqueness of equilibrium

Using the equilibrium conditions in De…nition 1 and Lemmas 1 through 5, we can solve
for the reservation price of buyers, R, the distribution of prices for the large seller, F , and
the distribution of prices for the small sellers, G.
The equilibrium condition (ii) states that the pro…t of the large seller is maximized
at every price on the support of the distribution F . That is, L(x) = L for all x 2 F.

Lemma 5 establishes that the support of the distribution F is some interval [x; x]. That
is, F =[x; x]. Therefore, for all x 2 [x; x], we have
L

= L(x)
=

+

b (x
+ (1

c)
:
)G(R(x))

(7)

Equation (7) pins down the maximized pro…t of the large seller. In fact, Lemma 3
establishes that the highest price posted by the large seller is x , i.e. x = x . Also,
17

Lemma 4 establishes that, when the large seller posts the price x, none of the small sellers
is priced out of the market, i.e. G(R(x)) = 1. Using these facts, we can solve for the
equation L = L(x) with respect to L and obtain
L =

b

(x
+

c)

(8)

:

Having solved for the maximized pro…t of the large seller, we can use equation (7) to
pin down the distribution of prices among small sellers. Intuitively, equation (7) states
that— as the price of the large seller falls from x to x— the decline in the pro…t that he
enjoys on each trade must be exactly compensated by an increase in the number of trades
that he makes. And since the increase in the number of trades made by the large seller
depends on the distribution of small sellers, equation (7) pins down the distribution of
prices among small sellers. Formally, using (8), we can solve (7) with respect to G(R(x))
and obtain

( + )(x
(1
)(x

G(R(x)) = 1

x)
:
c)

(9)

Having solved for the distribution of prices among the small sellers, we can solve for
the lowest price posted by the large seller. In fact, Lemma 4 establishes that, when the
large seller posts the price x, none of the small sellers is priced out of the market, i.e.
G(R(x)) = 0. Using this fact, we can solve equation (9) with respect to x and obtain
+
+

x=c+

(x

c):

(10)

The equilibrium condition (iii) states that the pro…t of the small seller is maximized
at every price on the support of the distribution G. That is, S(y) = S for all y 2 G.
Lemma 4 establishes that the support of the distribution G is the interval [y; y]. That is,

G =[y; y]. Hence, for all y 2 [y; y], we have
S

= S(y)
= b (1

)(y

c)

(x)
+
+

Z

x

R

1 (y)

+

F 0 (x)
dx ;
+ (1
)G(R(x))

(11)

where (x) denotes the probability that the large seller posts the price x.
Equation (11) can be used to pin down the mixing distribution of the large seller.
Intuitively, equation (11) states that— as the price of the small seller falls from y to y—
18

the increase in the pro…t that he enjoys on each trade must be exactly compensated by
an increase in the expected number of trades that he makes. And since the number of
trades made by the small seller depends on the probability that the large seller prices him
out of the market, equation (11) pins down the mixing distribution of the large seller.
Formally, equation (11) implies that S 0 (y) = 0 for all y 2 [y; y]. Since, Lemma 4

establishes that y = R(x) and y = R(x), it follows that S 0 (R(x)) = 0 for all x 2 [x; x] or,
equivalently,

0

(x) =

(x)

R0 (x)
;
R(x) c

(12)

F 0 (s)
ds:
+ (1
)G(R(s))

(13)

where (x) is de…ned as
(x)
+
+

(x) =

Z

x

x

+

The di¤erential equation in (12) can be solved for (x) to obtain
(x) = (x)

R(x)
R(x)

c
:
c

(14)

Using the solution for (x) in (14) and the de…nition of (x) in (13), we can recover the
mixing distribution F . In particular, after equating the right-hand side of (14) to the
right-hand side of (13) and di¤erentiating with respect to x, we …nd that the derivative
of F is given by
F 0 (x) = (x)( + )

R0 (x)(x c)
:
(R(x) c)2

(15)

Similarly, after equating the right-hand side of (14) to the right-hand side of (13) evaluated
at x = x, we …nd that the mass point (x) is given by
(16)

(x) = (x)( + ):
Since the mixing distribution F must integrate up to one, (x) must be given by
(x) =

( + ) 1+

Z

x

x

R0 (x)(x c)
(R(x) c)2

1

:

(17)

The equilibrium condition (i) can be used to solve for the buyer’s reservation price.
In fact, the condition states that the buyer’s reservation price, R(x), equals the di¤erence
between the buyer’s valuation of the good, u, and the buyer’s value of searching, V (x).
Using R(x) = u

V (x) and the fact— implied by Lemma 1 and Lemma 3— that x
19

R(x)

for all x 2 [x; x], we can solve the Bellman Equation (1) with respect to V (x). We …nd
that, for all x in the interval [x; x], the buyer’s value of searching is given by
(u
V (x) =

x) + (1
+

)

+ (1

R R(x)
y

(u

y)dG(y)
(18)

:

)G(R(x))

Di¤erentiating (18) with respect to x and using the fact that V 0 (x) =

R0 (x), we …nd

that, for all x in the interval [x; x], the derivative of the buyer’s reservation price is given
by
R0 (x) =

(x
c)
.
( + )(x c)

(19)

Then, using the expression for R0 (x) in (19) and the fact that R(x ) = x , we …nd that,
for all x in the interval [x; x], the buyer’s reservation price is given by
Z

R(x) = x

x

R0 (s)ds.

(20)

x

Finally, we solve for the equilibrium value of x . Since x = R(x ) and R(x ) =
u

V (x ), it follows that x = u V (x ) or, equivalently,
"
Z R(x )
1
x =u
(u
(u x ) + (1
)
+
y

#

y)dG(y) :

(21)

The integral on the right hand side of (21) is such that
Z

R(x )

(u

y)dG(y)

y

= u

y+

Z

R(x )

G(y)dy = u

x +

y

= u

x +

1

Z

x

(22)

G(R(x))R0 (x)dx

x

(x

x) +

+

(x

1

c) log

x
x

c
,
c

where the …rst equality follows from integration by parts, the second equality follows from
y = R(x), y = R(x) = R(x ) = x and from a change of variable, while the last equality
follows by substituting G(R(x)) with its equilibrium value in (10), by substituting R0 (x)
with its equilibrium value in (19), and by solving the integral.
After substituting (22) into (21) and solving for x , we obtain
x =

+

u+
20

+

c,

(23)

where

is de…ned as
=

+
+

(1
+

)

log

+
+

.

(24)

We have now identi…ed a unique candidate equilibrium. The price x is uniquely
determined by (23). Given x , the buyer’s reservation price R(x) is uniquely determined
by (20). Given x and R(x), the lowest price on the support of the distribution F , x, is
uniquely determined by (10), the highest bound on the support of the distribution F , x, is
uniquely determined by x = x , and the shape of the distribution F is uniquely determined
by (15) and (16). Similarly, given x , R(x) and x, the lowest price on the distribution
G, y, is uniquely determined by y = R(x), the highest price on the distribution G, y,
is uniquely determined by y = R(x), and the shape of the distribution G is uniquely
determined by (9). Hence, there is a unique candidate equilibrium.
The candidate equilibrium described above is indeed an equilibrium. First, F is a
proper cumulative distribution function. In fact, the coe¢ cient
and, hence, x > c, which in turn implies R(x)
Since R(x)

in (23) is strictly positive
0

c > 0 and R (x) > 0 for all x 2 [x; x].

c > 0 for all x 2 [x; x], F 0 (x) is strictly positive for all x 2 [x; x]. Moreover,

it is immediate to verify that F (x) = 0 and F (x) = 1. Second, G is a proper cumulative
distribution function. Since R0 (x) > 0 for all x 2 [x; x], G0 (y) is strictly positive for all
y 2 [y; y]. Moreover, it is immediate to verify that G(y) = 0 and G(y) = 1. Third,

the candidate equilibrium satis…es all the equilibrium conditions in De…nition 1. The
candidate equilibrium satis…es condition (ii). Indeed, from the construction of G, it
follows that the pro…t of the large seller takes the same value L for all prices on the
interval [x; x]. Moreover, it is straightforward to verify that the pro…t of the large seller
is smaller than L for all prices outside of the interval [x; x]. The candidate equilibrium
satis…es condition (iii). Indeed, from the construction of F , it follows that the pro…t of
the small seller takes the same value S for all prices on the interval [y; y]. Moreover, it is
straightforward to verify that the pro…t of the small seller is smaller than S for all prices
outside of the interval [y; y]. Finally, from the construction of R(x) and x , it follows that
the candidate equilibrium satis…es condition (i).
The above observations complete the proof of the following theorem.
Theorem 2: (Existence and Uniqueness). An equilibrium exists and is unique.

21

F '(x)

µ (x )

c

x

x

v

x

Figure 1: Large seller’s price distribution

3.3

Properties of equilibrium and comparative statics

It is useful to summarize the features of the equilibrium. The large seller draws its price
from the distribution F, whose properties are illustrated in Figure 1. The support of
the distribution is the interval [x; x], where x = x is a weighted average of the seller’s
marginal cost c and the buyer’s valuation u, while x is strictly greater than c and strictly
smaller than x. The distribution has a mass point of measure (x) at x. Moreover, using
(15), it is straightforward to verify that the distribution has a strictly positive and strictly
decreasing density over the interval [x; x).
The distribution of prices among small sellers is G, whose properties are illustrated in
Figure 2. The support of the distribution is the interval [y; y], where y is given by R(x)
and y is given by R(x). Using (20), it is immediate to see that y is equal to x and y is
strictly greater than x and strictly smaller than x. Moreover, using (9) and (19), one can
show that the distribution G has a strictly positive and strictly increasing density over
the interval [y; y].
The properties of the reservation price of buyers, R, are illustrated in Figure 3. From
Lemma 1 and Lemma 3, it follows that the reservation price is strictly greater than the
22

G '(y)

y

c

y

v

y

Figure 2: Small seller’s price distribution
price posted by the large seller for x 2 [x; x) and it is equal to the price posted by the

large seller for x = x = x . For x 2 [x; x], the reservation price is strictly increasing in x,

but less than one for one. Moreover, using (20), one can show that the reservation price
is strictly concave in x. The reservation price is informative about the buyer’s expected
surplus since Ex [V (x)] = u

Ex [R(x)]. Since R(x) is strictly increasing in x and x takes

values between x and x, it follows that buyer’s expected surplus is bounded below by
u

R(x) and bounded above by u

R(x). Since R(x) = x 2 (c; u) and R(x) 2 (c; u),

it follows that the buyer’s expected surplus is strictly greater than 0 and strictly smaller
than v

c. That is, the equilibrium of our market with search frictions is somewhere

between a perfectly competitive equilibrium and a pure monopoly equilibrium.

At the beginning of every market day, the large seller draws its price from the distribution F and the small sellers post prices according to the distribution G. In a positive
fraction of days, the large seller draws a price x equal to x. When x = x, the buyer’s
reservation price takes its highest value R(x) and, hence, the buyer’s surplus takes its
lowest value u

R(x). Moreover, when x = x, none of the small sellers is priced out of

the market and, hence, every meeting between a buyer and a seller leads to a trade. Since
all meetings between buyers and sellers are associated with the same positive gains from
23

R(x)

R(x )

R(x)
45o

x

c

v

x

x

Figure 3: Buyer’s reservation price
trade u

c > 0, the market outcome is e¢ cient when x = x. From time to time, the

large seller draws a price x below x. When x < x, the buyer’s reservation price takes the
value R(x) and the buyer’s surplus takes the value u

R(x), which is higher the lower

is x. Moreover, when x < x, a fraction of small sellers is priced out of the market and,
hence, not all meetings between buyers and sellers lead to trade. Hence, when x < x, the
market outcome is ine¢ cient and the extent of the ine¢ ciency is higher the lower is x.
The behavior of the equilibrium depends critically on the degree of market power of
the large seller— as measured by the parameter — and on the extent of search frictions—
as measured by the parameter . First, consider the behavior of the equilibrium as the
market power of the large seller vanishes, in the sense that

goes to zero. In this case,

the price distributions of the large and small sellers are such that
lim x = u;
!0

lim (x) = 1;

(25)

lim y = u:

(26)

!0

lim y = u,
!0

!0

The buyer’s expected surplus is such that
0=u

lim R(x)
!0

lim Ex [V (x)]
!0

24

u

lim R(x) = 0.
!0

(27)

The limits in (25)-(27) reveal that, as the market power of the large seller vanishes, all
the sellers post a price equal to the buyers’valuation and the buyers’expected surplus is
equal to zero. That is, as the market power of the large seller vanishes, the equilibrium
outcomes converge to those of a pure monopoly. This …nding is not surprising because,
as

! 0, our environment converges to the one studied by Diamond (1971). Indeed, as

! 0, the large seller has no impact on the buyer’s reservation price and, in turn, on the

fraction of small sellers that are priced out of the market. For this reason, the large seller
has no incentive to post any price below the buyer’s reservation price. Moreover, as

! 0,

the small sellers face no uncertainty about the realization of the reservation price because
this price is una¤ected by the large seller. For this reason, the small sellers also have no
incentive to post any price below the buyer’s reservation price. And, as pointed out by
Diamond (1971), when all the sellers post the reservation price, the buyers’reservation
price must be equal to u and their surplus must be equal to zero.
Next, consider the behavior of the economy when the market power of the large seller
becomes complete, in the sense that

goes to one. In this case, the price distributions of

the large and small sellers are such that
lim x = u;

lim x = u;

(28)

lim y = u,

lim y = u:

(29)

!1

!1

!1

!1

The buyer’s expected surplus is such that
0=u

lim R(x)
!1

lim Ex [V (x)]
!1

u

lim R(x) = 0.
!1

(30)

The limits in (28)-(30) reveal that, when the market power of the large seller becomes
complete, all the sellers post prices equal to the buyers’valuation and the buyers capture
none of the gains from trade. That is, when the market power of the large seller becomes
complete, the equilibrium outcomes converge to those of a pure monopoly. This …nding
is intuitive. As

! 1, the large seller has an impact on the buyer’s reservation price and

on the fraction of small sellers that are priced out of the market. However, the fraction of

small sellers that are priced out of the market has no impact on the number of trades that
the large seller makes. For this reason, the large seller has no incentive to post any price
below the buyers’reservation price. Similarly, the small sellers have no incentive to post
any price below the buyers’reservation price because the large seller has no incentive to
price them out of the market. Again, as all sellers post the reservation price, the buyers’
25

0.5
0.45

E [ V (x)| F (x)]

0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

α

Figure 4: Buyer’s surplus and market power
reservation price is u and their surplus is zero.
Overall, we have shown that the buyer’s surplus is strictly positive for any
and converges to zero for both

! 0 and

! 1. For

2 (0; 1)

! 0, the fraction of the gains

from trade accruing to the buyers goes to zero because our environment converges to the
one studied by Diamond (1971). For

! 1, the fraction of the gains from trade accruing

to the buyers goes to zero because the large seller e¤ectively becomes a monopolist. For
2 (0; 1), the incentives of the large seller to price the small sellers out of the market and

the incentive of the small sellers to avoid being priced out of the market by the large seller
lead to lower prices and allow buyers to capture some of the gains from trade. Therefore,
the fraction of the gains from trade captured by the buyers is nonmonotonic in the market
power of the large seller. In this sense, one can view the Diamond Paradox as the limit of a
paradoxical nonmonotonic relationship between the extent of competition between sellers
and the degree of market power of the large seller. Figure 4 illustrates this nonmonotonic
relationship for u = 1, c = 0:2,

For any

= 0:1 and

= 2.

2 (0; 1), the extent of competition varies in a natural way depending on

the extent of search frictions. Consider the behavior of the economy as search frictions
26

vanish, in the sense that
small sellers are such that

! 1. In this case, the price distributions of the large and
lim x = c;

lim x = c;

(31)

lim y = c,

lim y = c:

(32)

!1

!1

!1

!1

The buyer’s expected surplus is such that
u

c=u

lim R(x)
!1

lim Ex [V (x)]
!1

u

lim R(x) = u
!1

c.

(33)

The limits in (31)-(33) reveal that, as search frictions vanish, all the sellers post prices
equal to the marginal cost and the buyers capture all of the gains from trade. That is,
as search frictions vanish, the equilibrium outcomes converges to those that would obtain
under perfect competition in a frictionless market. Intuitively, as

! 1, buyers meet

sellers faster and faster and, for any nondegenerate distribution of prices, the reservation
price gets closer and closer to the lowest posted price. In turn, this induces sellers to push
prices towards the marginal cost c. Notice that this argument does not hold when
Indeed, when

= 0.

= 0, the price distribution is degenerate and, hence, the velocity at which

buyers meet sellers has no e¤ect on the buyers’reservation price.
Finally, consider the behavior of the economy as search frictions become in…nitely
large, in the sense that
sellers are such that

! 0. In this case, the price distributions of the large and small
lim x = u;

lim x = u;

(34)

lim y = u,

lim y = u:

(35)

!0

!0

!0

!0

The buyer’s expected surplus is such that
0=u

lim R(x)
!0

lim Ex [V (x)]
!0

u

lim R(x) = 0.
!0

(36)

The limits in (34)-(36) reveal that, as search frictions become in…nitely large, all the
sellers post prices equal to the buyers’valuation and the buyers capture none of the gains
from trade. That is, as search frictions become in…nitely large, the equilibrium outcomes
converge to those of a pure monopoly. Intuitively, as search frictions become larger,
buyers meet sellers at a lower rate and, for any nondegenerate distribution of prices, their
reservation price increases. In turn, this induces the large and the small sellers to push
their prices toward the buyer’s valuation u.
Figure 5 illustrates the relationship between the fraction of the gains from trade accru27

0.8

α
α
α
α
α

0.7

E [ V (x)| F (x)]

0.6

=
=
=
=
=

0. 9
0. 75
0. 5
0. 25
0. 1

0.5

0.4

0.3

0.2

0.1

0

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

1 − 1/(1 + λ)

Figure 5: Buyer’s surplus and search frictions
ing to the buyers and the extent of search frictions for di¤erent values of , given u = 1,
c = 0:2, and

= 0:1. For any value of

greater than zero and smaller than one, buyers

capture all of the gains from trade when search frictions vanish. Only for
or

equal to 0

equal to 1, buyers do not capture any of the gains from trade irrespective of how

small search frictions might be. Diamond (1971) stressed the existence of a discontinuity
in equilibrium outcomes between the case in which search frictions are arbitrarily small
and the case in which the market is frictionless. Our results show that this discontinuity
is not a robust feature of the equilibrium. It only exists when the large seller has either
no market power or complete market power.

4

Conclusions

This paper studied equilibrium pricing in a product market with search frictions. As in
Diamond (1971), buyers search sequentially for sellers. In contrast to Diamond (1971),
buyers do not meet all sellers with the same probability. In particular, a fraction of the
buyers’meetings involves one particular large seller, while the remaining fraction involves
one out of a continuum of small sellers. We established the existence and uniqueness of the
28

equilibrium, and we fully characterized its properties. We found that the equilibrium is
such that both the large and the small sellers play mixed pricing strategies. The fact that
small sellers play mixed strategies implies that there is price dispersion in equilibrium.
The fact that the large seller plays mixed strategies implies that there is variation in prices
and quantities from one day to the next. Buyers capture a positive fraction of the gains
from trade as long as the market power of the large seller— as measured by the parameter
— is positive but incomplete. When this is the case, the fraction of the gains from trade
accruing to the buyers depends in a natural way on the extent of search frictions— as
measured by the parameter . Speci…cally, buyers capture all of the gains from trade
when search frictions vanish, and they capture none of the gains from trade when search
frictions become in…nitely large. It is only when the large seller has no market power at
all (the case considered in Diamond 1971) or when he has complete market power that
buyers do not capture any of the gains from trade independently of how small search
frictions might be. Therefore, the Diamond Paradox is a non-generic outcome in markets
with sequential search.
The only di¤erence between the environment considered in this paper and the one
analyzed by Diamond is the presence of a large seller. The large seller’s price has an
impact on the buyers’decision of whether or not to purchase from the small sellers and,
hence, the number of buyers who meet and purchase from the large seller. This mechanism
is su¢ cient to break the Diamond Paradox. In fact, if all sellers posted the monopoly price,
the large seller could price them out of the market by lowering his price in…nitesimally.
The mechanism is also su¢ cient to generate price dispersion. In fact, because the large
seller tries to price the small sellers out of the market and the small sellers try not to
be priced out of the market, the sellers are playing a game of cat and mouse whose only
equilibrium involves mixed strategies and, hence, price dispersion.
The theory of price dispersion advanced in this paper seems quite general. As the
discussion above suggests, our theory of price dispersion follows immediately from the
presence of a large seller whose pricing decision impacts the buyers’search strategy and,
in turn, the number of buyers who visit his store. Hence, our theory should generalize
to an environment in which there are several large seller and a continuous fringe of small
sellers, as well as to an environment in which all sellers are large. Similarly, our theory
should generalize to an environment in which sellers are ex-ante identical and choose
their size through an ex-ante investment. However, one might still wonder whether it is
reasonable to assume that there are some sellers whose pricing decisions can a¤ect the
29

buyers’ reservation strategy and, in turn, the fraction of buyers who visit their stores.
First, for a seller to impact the buyers’ reservation price, it has to be the case that
(some) buyers meet that seller with positive probability. This condition seems reasonable.
Second, for a seller to impact the buyers’reservation price, it has to be the case that (some)
buyers know the price distribution when they embark in their search. This condition seems
at least plausible. For example, it may be the case that some buyers learn the distribution
of prices on the Internet before leaving home in the morning and, afterwards, they meet
a sequence of sellers on their way to and from work. Third, for a seller’s price to impact
the number of buyers who visit his store, it has to be the case that some other sellers
cannot respond to his pricing decision. This condition seems plausible, as any delay in
price adjustment would su¢ ce.

References
[1] Albrecht, J., and B. Axell. 1984. “An Equilibrium Model of Search Unemployment.”
Journal of Political Economy, 92: 824-840.
[2] Benabou, R. 1988. “Search, Price Setting and In‡ation.”Review of Economic Studies,
55: 353-376.
[3] Burdett, K., and K. Judd. 1983. “Equilibrium Price Dispersion.” Econometrica, 51:
955-970.
[4] Burdett, K., S. Shi, and R. Wright. 2001. “Pricing and Matching with Frictions.”
Journal of Political Economy, 109: 1060-1085.
[5] Burdett, K. and T. Wishwanath. 1988. “Declining Reservation Wages and Learning.”
Review of Economic Studies, 65: 655-665.
[6] Butters, G. 1977. “Equilibrium Distributions of Sales and Advertising Prices.”Review
of Economic Studies, 44: 465-491.
[7] Carrillo-Tudela, C., G. Menzio, and E. Smith. 2011. “Job Search with Bidder Memories.”International Economic Review, 52: 639-655.
[8] Curtis, E., and R. Wright. 2004. “Price Setting, Price Dispersion, and the Value of
Money: or, the Law of Two Prices.”Journal of Monetary Economics, 51: 1599-1621.

30

[9] Diamond, P. 1971. “A Model of Price Adjustment.”Journal of Economic Theory, 3:
156-168.
[10] Gaumont, D., M. Schindler, and R. Wright. 2006. “Alternative Theories of Wage
Dispersion.”European Economic Review, 50: 831-848.
[11] Kaplan, G., and G. Menzio. 2014. “The Morphology of Price Dispersion.” Working
Paper 19877, NBER, Cambridge, MA.
[12] Moen, E. 1997. “Competitive Search Equilibrium.” Journal of Political Economy,
105: 694-723.
[13] Mortensen, D., and C. Pissarides. 1994. “Job Creation and Job Destruction in the
Theory of Unemployment.” Review of Economic Studies, 61: 397-416.
[14] Pissarides, C. 1985. “Short-Run Equilibrium Dynamics of Unemployment, Vacancies
and Wages.” American Economic Review, 75: 676-690.
[15] Rothschild, M. 1974. “Searching for the Lowest Price when the Distribution of Prices
is Unknown.”Journal of Political Economy, 82: 689-711.
[16] Shimer, R. 2006. “Contracts in a Frictional Labor Market.”Manuscript, M.I.T.
[17] Sorensen, A. 2000. “Equilibrium Price Dispersion in Retail Markets for Prescription
Drugs.” Journal of Political Economy, 108: 833-850.
[18] Varian, H. 1980. “A Theory of Sales.”American Economic Review, 70: 651-659.

31

Appendix
A

Proof of Lemma 1

(i) A small seller never posts a price strictly lower than c, as this would give him strictly
negative pro…ts. Hence, G(c ) = 0. For the same reason, the large seller never posts a
price strictly lower than c. Hence, x

c. Now, …x x

c. The buyer’s value of searching

V is such that
(V ) +
(V )

V =
where

(V ) =

Z

(u

x

maxfu

V)

y

if V < u
if V
u

x,
x,

(A1)

(A2)

V; 0gdG(y):

Let LHS(V jx) denote the left-hand side of (A1). The function LHS(V jx) is a continuous,
strictly increasing function of V such that LHS(0jx) = 0 and LHS(u

cjx) = (u

c) >

0. Let RHS(V jx) denote the right-hand side of (xx). The function RHS(V jx) is a
continuous function of V because (V ) is a continuous function. The function RHS(V jx)
is decreasing in V because both
RHS(0jx)
(u

0 because

(0)

(V ) and

(u

0 and and RHS(u

x

V ) are decreasing in V . Moreover,

cjx) = 0 because G(c ) = 0 implies

c) = 0.

The properties of LHS(V jx) and RHS(V jx) can be used to characterize the solution to

(A1). First, there exists a unique V that solves (A1) since LHS(V jx) is strictly increasing
in V and RHS(V jx) is decreasing in V . Let us denote as V (x) the solution to (A1).
Second, V (x) 2 [0; u c) since LHS(0jx)

RHS(0jx) and LHS(u cjx) > RHS(u cjx).

Third, V (x) is decreasing in x since LHS(V jx) is independent of x and RHS(V jx) is
decreasing in x.

Notice that V (x) 2 [0; u

c) implies that V (x) < (u

x) for x = c and V (x)

(u

x)

for x = u. Since V (x) is continuous in x, there exists at least one x 2 (c; u] such that
V (x ) = u

x . Let x1 be equal to one of these x s. Then, LHS(V (x1 )jx1 ) = V (x1 );

RHS(V (x1 )jx1 ) =

(V (x1 )) and LHS(V (x1 )jx1 ) = RHS(V (x1 )jx1 ). Now, consider an

arbitrary x2 > x1 . Then, LHS(V (x1 )jx2 ) = V (x1 ) and RHS(V (x1 )jx2 ) =
since V = u

x1 implies V > u

(V (x1 ))

x2 . Therefore, LHS(V (x1 )jx2 ) = LHS(V (x1 )jx1 ) and

RHS(V (x1 )jx2 ) = RHS(V (x1 )jx1 ), which implies LHS(V (x1 )jx2 ) = RHS(V (x1 )jx2 )
and V (x2 ) = V (x1 ). Since x1 = x and x2 was chosen arbitrarily, it follows that V (x) =
V (x ) for all x

x . Since V (x ) = (u

x ) and V (x) = V (x ) for all x > x , it follows
32

that V (x) > u

x for all x > x . Since V (x ) = (u

x > x , there is a unique x such that V (x ) = (u
Now, consider x1 and x2 such that c

x ) and V (x) > (u

x) for any

x ).
x . Since x1 < x , V (x) < u

x1 < x 2

x

and
V (x1 ) = (V (x1 )) +
Similarly, since x2 < x , V (x) < u

(u

x1

V (x1 )) :

(A3)

(u

x2

V (x2 )) :

(A4)

x and

V (x2 ) = (V (x2 )) +

After subtracting (A3) from (A4) and grouping terms, we obtain
V (x1 )

(V (x1 ))

V (x2 ) =

(V (x2 )) +
+

Since (V ) is decreasing in V and V (x) is decreasing in x,

[x2

x1 ]

(A5)

:

(V (x1 )) < (V (x2 )). There-

fore, we have
V (x1 )

V (x2 )

Since V (x) is decreasing in x, V (x1 )

(x2

+

x1 ) < x2

(A6)

x1 :

V (x2 ). If V (x1 ) = V (x2 ), then

(V (x1 )) =

(V (x2 )) and the right-hand side of (xx) is strictly positive. Hence, we must have
0 < V (x1 )

(A7)

V (x2 ).

Parts (i), (ii) and (iii) in Lemma 1 follow immediately from the properties of V (x) and
R(x) = u

B

V (x).

Proof of Lemma 3

(i) We want to establish that x = x in any equilibrium. Let (R; F; G) denote an equilibrium. First, suppose x > x . The pro…t of the large seller is given by
L(x) =

+

b 1[x R(x)](x
1[x R(x)] + (1

Notice that L(x) = 0 because x > x implies 1[x
because 1[x

c)
:
)G(R(x))

(B1)

R(x)] = 0. In contrast, L(x ) > 0

R(x )] = 1 and x > c. Therefore, L(x) < L(x )

L . However, since

(R; F; G) is an equilibrium, L(x) = L . Thus, we have reached a contradiction and x

33

must be smaller or equal to x .
Next, suppose x < x . The pro…t of the small seller is given by
Z
b (1
)1[y R(x)](y c)
S(y) =
dF (x):
+ 1[x R(x)] + (1
)G(R(x))

(B2)

Notice that S(y) = 0 for all y > R 1 (x). In contrast, S(R(c)) > 0 because F (c ) = 0
S for all y > R 1 (x). Since S(y) = S for

and R(c) > c. Therefore, S(y) < S(R(c))
all y 2 G, it follows that G(R 1 (x)) = 1.

For all x 2 [x; x ], G(R(x)) = 1 and, hence, the pro…t of the large seller is such that
L(x) =

b

(x
+

c)

The above expression implies that L(x) < L(x )

(B3)

:

L . However, since (R; F; G) is an

equilibrium, L(x) = L . Thus, we have reached a contradiction and x must be greater or
equal to x .
(ii) We want to establish that y = R(x) in any equilibrium. Let (R; F; G) denote an equilibrium. We have already shown above that y > R(x) is not consistent with equilibrium.
Now, suppose y < R(x). For all x 2 [R 1 (y); x], G(R(x)) = 1 and 1[x

R(x)] = 1.

Hence, for all x 2 [R 1 (y); x], the pro…t of the large seller is given by
b
(x
+

c):

(B4)

The above expression implies that L(R 1 (y)) < L(x)

L . Since L(x) = L for all

L(x) =

x 2 F, L(R 1 (y)) < L implies that F (R 1 (y) ) = F (x ).

Since F (R 1 (y) ) = F (x ), the pro…t of the small seller is such that
S(y) =

b (1
+
b (1
+

)
)

c) (x)

(y

(R(x)

(B5)

c) (x) = S(R(x));

where (x) denotes the probability that the large seller posts the price x. If (x) > 0,
S(y) < S(R(x))

S . If (x) = 0, S(y) = S(R(x)) = 0 < S . However, since (R; F; G) is

an equilibrium, S(y) = S . Thus, we have reached a contradiction and y must be greater
or equal to R(x).
(iii) We want to establish that y = R(x) in any equilibrium. Let (R; F; G) denote an
equilibrium. First, suppose y < R(x). For all y 2 [y; R(x)], F (R 1 (y) ) = 0 and, hence,
34

the pro…t of the small seller is given by
S(y) =

b (1
+

)

(y

(B6)

c):

The above expression implies that S(y) < S(R(x))

S . However, since (R; F; G) is an

equilibrium, S(y) = S . Thus we have reached a contradiction and y must be greater or
equal to R(x).
Next, suppose y > R(x). For all x 2 [x; R 1 (y)), G(R(x)) = 0 and, hence, the pro…t of

the large seller is given by

L(x) =

b
+

(x

(B7)

c):

The above expression implies that L(x) < L(R 1 (y)

) for any

However, since (R; F; G) is an equilibrium, L(x) = L

L(R 1 (y)

2 (0; R 1 (y)

x).

). Thus we have

reached a contradiction and y must be smaller or equal to R(x).

C

Proof of Lemma 4

(i) We want to establish that the distribution G has no gaps in equilibrium. Let (R; F; G)
denote an equilibrium. On the way to a contradiction, suppose that there is a gap in
the distribution G between the prices y1 and y2 , with y1 2 G, y2 2 G and y1 < y2 . For
all x 2 [R 1 (y1 ); R 1 (y2 )), G(R(x)) = G(y1 ) and 1[x

R(x)] = 1. Therefore, for all

x 2 [R 1 (y1 ); R 1 (y2 )) the pro…t of the large seller is given by
L(x) =

+

b (x
+ (1

c)
)G(y1 )

:

(C1)

The above expression implies that L(x) is strictly increasing over the interval [R 1 (y1 ); R 1 (y2 )).
In turn, this implies that F (R 1 (y1 ) ) = F (R 1 (y2 ) ).
Since F (R 1 (y1 ) ) = F (R 1 (y2 ) ), for all y 2 [y1 ; y2 ] the pro…t of the small seller is
given by

S(y) =

Z

x

R

1 (y

1)

If F (R 1 (y1 ) ) < 1, S(y1 ) < S(y2 )

b (1
)(y c)
dF (x):
+
+ (1
)G(R(x))

(C2)

S . If F (R 1 (y1 ) ) = 0, S(y1 ) = S(y2 ) = 0 <

S . However, since (R; F; G) is an equilibrium, S(y1 ) = S . Thus we have reached a
contradiction and there cannot be a gap in the distribution G.
(ii) We want to establish that the distribution G has no mass points in equilibrium. Let
(R; F; G) denote an equilibrium. On the way to a contradiction, suppose that there is a
35

mass point in the distribution G at the price y1 , with y1 < y. In this case, the pro…t of
the large seller has a discontinuity at R 1 (y1 ). In fact, since G(y1 ) < G(y1 ), there is an
> 0 such that
1

L(R (y1 )

b
+
b
+

) =
>

(R
+
(R
+

1

(y1 )
c)
(1
)G(y1
)
1
(y1 ) c)
= L(R 1 (y1 )):
(1
)G(y1 )

The discontinuity in the pro…t of the large seller implies that L(x) < L(R 1 (y1 )

(C3)

) for

all x 2 [R 1 (y1 ); R 1 (y2 )] for some y2 2 (y1 ; y). In turn, this implies that F (R 1 (y1 ) ) =

F (R 1 (y2 ) ).

Since F (R 1 (y1 ) ) = F (R 1 (y2 ) ), for all y 2 [y1 ; y2 ] the pro…t of the small seller is

given by

S(y) =

Z

x
1 (y

R

b (1
)(y c)
dF (x):
+
+ (1
)G(R(x))

1)

If F (R 1 (y1 ) ) < 1, S(y1 ) < S(y2 )

(C4)

S . If F (R 1 (y1 ) ) = 0, S(y1 ) = S(y2 ) = 0 <

S . However, since (R; F; G) is an equilibrium, S(y1 ) = S . Thus we have reached a
contradiction and there cannot be a mass point at y1 < y . A similar argument can be
used to show that there cannot be a mass point at y1 = y .

D

Proof of Lemma 5

(i) We want to establish that the distribution F has no gaps in equilibrium. Let (R; F; G)
denote an equilibrium. On the way to a contradiction, suppose that there is a gap in the
distribution F between the prices x1 and x2 , with x1 2 F, x2 2 F and x1 < x2 . For all

y 2 [R(x1 ); R(x2 )), F (x1 ) = F (x2 ). Therefore, for all y 2 [R(x1 ); R(x2 )], the pro…t of
the small seller is given by

S(y) =

Z

x

x1

b (1
)(y c)
dF (x):
+
+ (1
)G(R(x))

(D1)

The above expression implies that S(y) is strictly increasing over the interval [R(x1 ); R(x2 )].
In turn, this implies that G(R(x1 ) ) = G(R(x2 ) ). That is, the distribution G has a gap
between the prices R(x1 ) and R(x2 ) Since, this contradicts Lemma 4, the distribution F
cannot have a gap.
(ii) We want to establish that the distribution F has a mass point only at x. Let (R; F; G)
36

denote an equilibrium. To establish that F has a mass point at x, suppose it does not.
Then, the probability that the large seller posts a price strictly smaller than x is one,
i.e. F (x ) = 1. In turn, this implies that the pro…t of a small seller posting the price
y = R(x) is zero, i.e. and S(y) = 0. However, since (R; F; G) is an equilibrium and y 2 G,

S(y) = S > 0. Thus, we have reached a contradiction and F must have a mass point at
x.
To establish that F does not have any other mass points, suppose there is a mass point
at some price x1 < x. Then, the pro…t of the small seller is discontinuous at R(x1 ). In
particular, S(R(x1 )) > S(R(x1 )+). Hence, there exists an R(x2 ) 2 (R(x1 ); y) such that

S(y) < S(R(x1 ))

S for all y 2 (R(x1 ); R(x2 )). Since S(y) = S for all y 2 G, the

previous inequality implies G(R(x1 )) = G(R(x2 )) That is, the distribution G has a gap
between the prices R(x1 ) and R(x2 ). Since, this contradicts Lemma 4, the distribution F
cannot have a mass point at x < x.

37