American Economic Association

Optimal Auctions
Author(s): John G. Riley and William F. Samuelson
Source: The American Economic Review, Vol. 71, No. 3 (Jun., 1981), pp. 381-392
Published by: American Economic Association
Stable URL: [Link]
Accessed: 22/01/2009 00:44

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at
[Link] JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at
[Link]

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.

JSTOR is a not-for-profit organization founded in 1995 to build trusted digital archives for scholarship. We work with the
scholarly community to preserve their work and the materials they rely upon, and to build a common research platform that
promotes the discovery and use of these resources. For more information about JSTOR, please contact support@[Link].

American Economic Association is collaborating with JSTOR to digitize, preserve and extend access to The
American Economic Review.

[Link]
 OptimalAuctions
By JOHN G. RILEY AND WILLIAM F. SAMUELSON*

In the two decades since the seminal paper dures? In turn, how is a sale price de-
by William Vickrey, literature on the theory termined? Second, by what means can the
of auctions has developed at a rapid though seller best exploit his monopoly position?
uneven pace.' Much of this literature is frag- For example, would it be more profitable for
mentary, varies widely in scope, and is not the seller to require payment not only by the
easily accessible to economists. As a result, high bidder, but also by those with lower
the implications of different auction rules in ranked bids?2
various settings remain relatively unknown. As one might expect, any change in the
This paper provides a systematic examina- rules of the auction results in different bid-
tion of alternative forms of auctions. In so ding strategies on the part of the buyers. In
doing it presents a general characterization particular, if the auction rules posit a mini-
of the implications for resource allocation of mum payment by one or more of the bidders
different auction designs within the model (determined by rank), those with sufficiently
originally proposed by Vickrey. low reservation values will be discouraged
The auction model is a useful description from entering a bid. Our analysis will dem-
of "thin markets" characterized by a funda- onstrate that, in a risk-neutral setting, it is
mental asymmetry of market position. While the reservation value below which a buyer
the standard model of perfect competition opts to remain out of the auction which is
posits buyers and sellers sufficiently numer- crucial. To be precise, if the lowest reserva-
ous that no economic agent has any degree tion value for which it is worthwhile bidding
of market power, the bare bones of the auc- is the same for two different auction rules,
tion model involves competition on only one then the expected return to the seller is also
side of the market. In this setting a single the same.
seller of an indivisible good faces a number Throughout the paper we shall retain the
(n) of potential buyers. Competition among following basic assumption.
the (possibly small number of) buyers takes a) A single seller with reservation value
place according to a well-defined set of auc- v0 faces n potential buyers, where buyer i
tion rules calling for the submission of price holds reservation value vi, i 1,..., n.
offers from the buyers. Most commonly, the b) The reservation values of the parties
choice of auction method employed rests with are independent and identically distributed,
the monopolistic seller. drawn from the common distribution F(v)
These brief observations suggest two natu- with F(v) = O, F(v-) = 1 and F(v) strictly in-
ral questions for analysis: First, what form creasing and differentiable over the interval
does the competition among the few buyers [v,v-. We will refer to this as the IID as-
take under the most common auction proce- sumption.
The IID assumption was first presented by
*Professor of economics, University of California-Los Vickrey, and has been frequently employed
Angeles, and assistant professor, School of Manage-
ment, Boston University, respectively. Riley's research 2Vickrey's comparison of the open "ascending bid"
was supported by National Science Foundation grant auction and the sealed "high bid" auction has been
SOC79-07573. Helpful discussions with Howard Raiffa, generalized in unpublished dissertations by Armando
Jerry Green, Eric Maskin, and Paul Samuelson, and the Ortega- Reichert, Gerard R. Butters, and William
valuable comments of a referee are gratefully acknowl- Samuelson. Milton Harris and Artur Raviv (1979) pro-
edged. vide the first discussion of optimal auction design. Em-
'A current bibliography by Robert Stark and Michael ploying very different methods they consider the special
Rothkopf lists nearly 500 papers written over this period. case in which each buyer has flat (uniform) prior prob-
For a recent survey of this literature, see Richard abilistic beliefs about the amount others are willing to
Engelbrecht-Wiggans. pay.

381
382 THE A MERICAN ECONOMIC RE VIE W JUNE 1981

in the bidding literature. In practical terms, maximum of the reservation values of the
each party is uncertain about the others' other n - 1 buyers. As Vickrey noted, this is
reservation values, believing that each indi- equivalent to a sealed bid auction in which
vidual decides the maximum amount he is each buyer submits a bid and the high bidder
willing to pay independently of the others. In pays the second highest rather than the
addition, the parties share common priors highest bid.3 To see this, suppose the ith
with respect to the possible reservation val- buyer considers shading his bid, bi, below his
ues of each individual. With the IID assump- reservation value vi. If the largest of all the
tion, the bidding procedures we outline be- other bids, b*, exceeds vi, another buyer is
low belong to the class of games of incom- the high bidder so that buyer i's gain re-
plete information first formulated by John mains zero. If b*<b1, buyer i remains the
Harsanyi. high bidder and continues to gain vi-b*.
The paper is organized as follows: First we However, if bi <b* <vi, the shading yields a
present our central result demonstrating that zero gain, whereas without shading the gain
expected seller revenue from quite different is vi-b*. A parallel argument establishes
auctions can be very easily compared. As an that there is no advantage in making a bid,
immediate implication, the equivalence of the b1, greater than vi. The optimal strategy of
"English" or "ascending bid" auction and each buyer is therefore to submit his reserva-
the "Dutch" or "high bid" auction is estab- tion value. It follows that, just as in the
lished. More important, it is shown that, for English auction, the high bidder ends up
a broad family of auction rules, expected paying the second highest reservation value.
seller revenue is maximized using either of This equivalence greatly simplifies the com-
the two common auctions if the seller an- parison between the English and sealed high
nounces that he will not accept bids below bid auctions4 since it implies that we need
some appropriately chosen minimum or "re- only compare sealed bid auctions.
serve" price. Surprisingly, this reserve price In the high and second bid auctions, only
is independent of the number of buyers and the winner makes a payment to the seller.
is always strictly greater than the seller's However, there is an infinity of auction rules
personal value of the object. In Section II involving payment by more than one bidder.
several alternative auction rules are de- For example, all buyers might be charged a
scribed in detail and their implications for fixed entry fee. Alternatively, losers might be
the seller are compared. Finally, in Section required to pay some fraction of their bids.
III, the two commonly used auctions are A third possibility, discussed in Section II, is
once again compared under the assumption that the seller might attempt to encourage
that the buyers are risk averse rather than higher bids by offering to return some of the
risk neutral. It is shown in this setting that money paid by the winner to each of the
the English auction is dominated by the losers, the size of the rebate depending on a
sealed high bid auction, and that the optimal loser's bid.
reserve price in the latter is a declining func- Each of these alternatives is an example of
tion of the degree of buyer risk aversion. an auction with the following properties.
First, a buyer can make any bid above some
of AlternativeAuctionRules
I. Comparison minimum "reserve" price announced by the
seller. Second, the buyer making the highest
Before characterizing a broad family of bid is awarded the object. Third, the auction
alternative auction rules, a few remarks about rules are anonymous: each buyer is treated
the English or ascending bid auction will be
helpful.
In auctions of antiques, estate objects, and 3This type of auction is sometimes referred to as a
works of art, the good is awarded to the Vickrey auction.
buyer who makes the final and highest bid. 4The sealed high bid auction also has its open auc-
tion equivalent. In this Dutch auction, the sale price is
The buyer placing the highest valuation on initially set at a high level and is then lowered until a bid
the good therefore pays approximately the is made.
VOL. 71 NO. 3 RILEY A ND SA MUELSON: OPTIMA L A UCTIONS 383

alike. Fourth, there is a common equilibrium pected buyer gain. Then, from (2), we are
bidding strategy in which each buyer makes able to derive the expected payment of a
a bid bi, which is a strictly increasing func- typical buyer. Given the symmetry of the
tion of his reservation value vi, i.e., auction rule, expected seller revenue is just n
times this expected payment.
(1) bi=b(vi) i=1,...,n With buyers behaving noncooperatively, a
common strategy, bi= b(vi), is an equi-
Throughout this section we shall consider the librium strategy if, when adopted by all
family ( of auction rules for which these four buyers but one, the latter's best response is
assumptions are satisfied.' We begin with the to adopt it also. Without loss of generality,
main result. we may suppose that the buyer considering
an alternative bidding strategy is buyer 1.
PROPOSITION 1: Suppose the IID assump- With all other buyers bidding according to
tion holds and all buyers are risk neutral. The b(v), buyer 1, if he bids at all, will wish to
common equilibrium bidding strategy for any bid in the range of this function. Hence, we
member of the family &fof auction rules yields can write any bid as b, =b(x) and view
an expected revenue to the seller of buyer 1 as choosing x. It follows that b(v) is
an equilibrium bidding strategy if buyer 1
n v(vF (v)+F(v)- 1)F(v) 'dv can do no better than choose x =v, and so
bids b(v,).
To examine the optimal choice of buyer 1,
where v* is the reservation value below which we begin by assuming his bid is b(x), and
it is unprofitableto submit a bid. then ask what restrictions are implied by the
requirement that his optimal bid is b(v,).
Proposition 1 is important because it tells Any auction rule must specify the amount he
us that the expected revenue of the seller must pay, p, given his own bid b, = b(x) and
from quite different auctions can be com- those by the other n - 1 buyers, i.e.,
pared simply by determining the lowest
reservation value, v*, for which it is p=p(b,, b2 I... I bn)
worthwhile bidding. We begin the proof by
examining the behavior of a single buyer. =P(b(x), b(V2),..., b(vn))
The expected return to making a bid can be
expressed as follows. We may therefore write the expected pay-
ment by buyer 1, given a bid of b, =b(x) as
(2)
I expected' probability] (3)
buyer 4 reservation of P(x) ?
v2 ... ,Vn
p(b(x), b(V2), ..., b(vn))
gain J value winning J
Also, the bid of b(x) is the winning bid if
expected and only if all other buyers have made lower
paymentJ bids. By assumption the equilibrium bid
function is strictly increasing in v, therefore
Below we obtain simple expressions for buyer 1 wins if all other valuations are less
both the probability of winning and the ex- than x. Since the probability buyer j has a
reservation value less than x is F(x), buyer 1
5After deriving Proposition 1, we became aware of a wins with probability Fn- '(x). Combining
paper by Roger Myerson which uses a much more this last result with (2) and (3), the expected
technically demanding approach to examine expected
seller revenue in an even broader class of auctions.
gain to buyer 1, if he chooses to enter the
Generalizing our approach, Eric Maskin and Riley auction, can be expressed as
(1980a) have shown that Myerson's results imply there
there are circumstances in which expected seller revenue
can be increased by prohibiting bids over certain ranges.
384 THE AMERICAN ECONOMIC REVIEW JUNE 1981

For b(v) to be the equilibrium bidding The final step is to consider the auction
strategy,buyer l's optimalchoice must be to from the seller's viewpoint. As far as the
select x=v1 and bid b(v,). Then buyer l's seller is concerned, v, and hence the ex-
maximized expected gain is H(v,,v,) and pected payment,P(vv), is a randomvariable.
the following first-ordercondition must be The seller'sexpectedrevenuefrom buyer 1 is
satisfied.6 thereforethe expectationof P(v,). Since the
seller knows that v1 has distributionF(v1)
(5) aa (x,VI) his expectedrevenueis

Iv d F (x)-P(x)=O atx=v1 pl =f|P(v,)F'(vl)dv

This must hold for all reservationvalues Substituting for P(vl) from (8b) and in-
exceedingv., the reservationvalue for which tegratingby parts,the expectedrevenuefrom
a buyeris indifferentbetweensubmittingthe buyer 1 can be rewrittenas follows:
bid, b(v*), and not entering the auction.
That is, (5) holds for all v1 >v, where v, (9) p1V
v* vF( v)
satisfies

(6) H-I(v*,v*)=v*Fn-'(v*)-P(v*)=O +F(v) -1] F n-I(v)dv

Settingx =v, in (5), it follows that the equi- Given the equal treatmentof all n buyers,
librium expected payment by buyer 1 must expected seller revenue is just n times the
satisfy the differentialequation expectedrevenuefrom buyer 1 and the pro-
position is proved.
d F-(VI) One of the strikingfeaturesof our deriva-
(7) P(V,)=V, dv VI;.:V* tion is that nowhere is there explicit refer-
I
ence to the equilibriumbiddingstrategyb(v).
Integratingand makinguse of the boundary However,under any particularauction rule
condition,(6), buyer l's expectedpaymentis this is [Link] key to such der-
therefore ivation is (8), the expression for expected
payment, P(v), of a buyer with reservation
(8a) P(v1)=v*Fn- (V*) value v.
For example,in the high bid auction, sup-
+VIxdFn-(X) vX v1
pose the seller announcesa reserveprice bo.
Any buyerwith a reservationvaluev> bohas
an incentive to enter, i.e., v* = bo. Since a
Integratingthe second term by parts, this buyer pays if and only if he is the high
can be rewrittenmore convenientlyas bidder,his expectedpaymentis

(8b) (10) P(v)=Prob {b(v) is high bid) b(v)

P(v)=v,Ff-II(V,)
VI But b(v) is the high bid if and only if all
{ln-l(x)dx v, V* other buyers have lower reservationvalues.
Then Prob {b(v) is high bid)=Fn-'(v) and,
from (10), b(v)=P(v)/F n-(V). Substitut-
6From(4) and (5) we also have ing for P(v) from(8b), we thereforehave the
followingadditionalresult.
ai-(x, v,)=(vl
ax -x)(n- I)Fn-I( x)F'(x)
ThereforeII(x,v1) is increasingin x for x<vl and PROPOSITION 2: Suppose the IID assump-
decreasingfor x>v, and the first-orderconditionyields tion holds, all buyers are risk neutral and the
the globalmaximumfor buyer 1. seller announces a reserveprice bo. In the high
VOL. 71 NO. 3 RILEYA ND SA MUELSON: OPTIMA L A UCTIONS 385

bid auction, the equilibriumbiddingstrategy of We thereforehave the following furtherre-

a typical buyer with reservation value v >bo is sult.

b(v)=v-| F '(x) dxlF (v) PROPOSITION 3: If the IID assumption

holds and buyers are risk neutral, the members
of the family C of auction rules, which maxi-
Proposition2 indicatesthe degreeto which mize the expected gain of the seller are those
a buyerwill "shade"his bid, b(v), below his for which the reservationvalue v., below which
reservationvaluev in the high bid [Link] it is not worthwhilebidding, satisfies
is a straightforwardmatter to confirm that
b(v) is strictlyincreasingin v. Thereforethe v* =v0 + 1-F(v*)/F'(v*)
highbid auctionis a memberof the familyof
auctions described by Proposition 1. Cer- independentof the number of buyers.
tainly the secondbid auctionis in this family
since anyoneenteringwill bid his reservation An immediateimplicationof Proposition
value. In both auctions,a buyer will enter if 3 is that the high and second bid auctions
and only if his reservationvalue exceeds bo. with reservepricebo -v* are both optimalin
Then in both cases v.= bo and, from Prop- the family of auctions i. Note also that v*
osition 1, expectedsellerrevenueis the same. exceeds v0: the seller announces a reserve
We now demonstratethat these two com- price strictlygreaterthan his personalvalua-
mon auctionrules are optimal for the seller, tion.
given the appropriatechoice of a reserve To gain an understandingof this strong
price. For any auction rule there is some result, it is helpful to considera second bid
impliedminimumreservationvalue v. below auctionin which thereare two buyersand to
which buyers will choose not to bid. Then examine the implications of introducing a
there is a probabilityof Fn(v.) that all n reserve price slightly higher than v0; i.e.,
buyers will decide not to submit a bid. In v* = v0+ 8 where 8 is small. Since each
this case the seller'sgain is his own personal buyer's dominant strategyis to bid his res-
valuation vo. Then, from (9), the total ex- ervationvalue, the expectedgain to the seller
pected returnto the selleris is affected (i) if both valuationslie between
v0 and v., and (ii) if one valuation lies
(11) vOFn(v*)+nf (vF'(v) betweenv0 and v* and the other exceedsv*.
v* In the first case the seller retains the item,
+ Ffv)- I)Fn- (v)dv whichhe valuesat v0, ratherthan sellingit at
some price between v0 and v*. His loss is
It followsthat any two auctionsin the family thereforeof order 8. Since this outcome oc-
C, for which v* is the same, yield the same curs with probability (F(v*)-F( vo ))2 ;
expected gain to the seller.7Moreover,dif- (F'(Vo))282, the expectedloss is of order8'.
ferentiatingwith respect to v* the expected In the second case the seller receivesa pay-
gain of the seller is maximizedfor some v* ment of v. ratherthan some price between
satisfyingthe condition8 v0 and v. hence has a gain of order8. Since
this outcomeoccurswith probability2(F(v*)
(12) n[voF'(v*)-v*F'(v*) the
-F(vo))(1-F(v*)t2F'(vo)(1-F(vo)8,
- F(v* +11I]Fn-It .
V*)=
expected gain is of order 82. Therefore,for
sufficiently small 8, the gain to raising the
7Moreover,any two auctions for which v* is the
reserveprice above the seller'spersonalval-
sameyield the sameexpectedgain to buyeri conditional uation outweighsthe cost.
on v,. Thisresultfollowsdirectlyfromequations(4) and Whilewe have focusedon the reserveprice
(8b2. becauseof its commonusage,thereare many
Expression(12) will, in general,have multipleroots. different ways in which to discourage the
If this is the case,it is necessaryto evaluatethe expected
return,(11), at each root to determinethe global maxi- appropriatesubsetof buyersfrom participat-
mum. ing in the bidding. Suppose, for example,
386 THE AMERICAN ECONOMIC REVIEW JUNE 1981

that the seller announces a fixed entry fee c. nant strategy. Therefore the seller cannot
For all buyers with valuations less than some influence bids by concealing his reserve price.
number vc, it will be optimal to remain out It follows that the optimal silent reserve price
of the auction. Consider a buyer with the is the same as the optimal announced reserve
borderline reservation value vc. In the second price and that expected seller revenue is
bid auction he enters, and, since the entry fee identical.
is now sunk, bids his true value vc. He wins if The argument is more complex in the case
and only if there are no other bidders, in of the high bid auction, but once again it can
which case there is no additional payment. be shown that there is no advantage in using
Since this occurs with probability F(v )Cy a silent reserve price. The proof, which
his expected profit is involves a straightforward extension of Prop-
osition 1, is provided in our earlier paper.
(13) VcF(vj)n-c II. AlternativeAuctions

But for vc to be the borderline reservation To illustrate the results of Section I, we

value, the expected profit must be zero. The now compare some specific auctions under
seller then chooses an entry fee c* satisfying the simplifying assumptions that there are
only two buyers, and that reservation values
(14) c*=v*F(v*) are uniformly distributed on the unit interval
(F(v) = v, for v E [O,1]). We assume also that
A similar argument holds for the high bid the object for sale has no value to the seller,
auction. If a buyer has the borderline res- vo =0. First we indicate the gains to employ-
ervation value vc, he wins if and only if there ing an optimal reserve price in the high bid
are no bidders. The optimal bid in such auction. We then present an unusual pair of
circumstances is zero, therefore the expected auction designs which happen to belong to
profit is again given by (13) and the optimal the class of optimal auctions.9 In contrast, a
entry fee by (14). third example shows that a seemingly natural
Our general results are also helpful in (and commonly employed) auction proce-
analyzing the expected payoff to multiple dure is suboptimal.
rounds of bidding. Suppose, for concrete- Under our simplifying assumptions, it fol-
ness, that the seller is using a high bid auc- lows from Proposition 3 that it is optimal for
tion. If he can convince buyers that there will the seller to design the auction so that only
only be a single round with optimal reserve those with reservation values exceeding v*=
price v*, there will be a chance that no bids 1/2 find it worthwhile bidding. Then, in the
will be submitted. Since v* exceeds the seller's high bid auction it is optimal for the seller to
reservation value v0, the seller, after the fact, announce a minimum or reserve price bo -
has an incentive to lower his reserve price 1/2. Appealing to Proposition 2, this, in
and to call for a second round of bids. turn, implies that the equilibrium bid of buyer
However, if buyers are not fooled, they will i, with reservation value vi > 1/2, is b(vi) =
adopt first-round strategies in the expec- vi/2+ 1/8vi. By contrast, if the seller always
tation of a possible second round. In particu- sells the good (by setting the reserve price
lar, all those with reservation values V.>boo, bo = 0) buyer i's bid becomes b(vi) = vi /2.
the reserve price in the second round, will Either by direct computation or by appealing
plan to enter the two round auction. Then, to Proposition 1, it can be confirmed that
from Proposition 3 the seller's expected gain expected seller revenue is 5/12 with bo = 1/2
is lower since the entry value is no longer and 1/3 with bo =0. Thus the optimal re-
optimal.
A final point concerns the decision of the
seller whether or not to announce a reserve 9The interested reader is referred to our earlier paper,
where it is shown that the auctions of examples I and 2
price. In the second bid auction, the strategy belong to the class of optimal auctions for arbitrary
of bidding one's reservation value is a domi- F(v) and n.
VOL. 71 NO. 3 RILEY A ND SA MUELSON: OPTIMA L A UCTIONS 387

serve price strategy results in a 25 percent Example 2: Santa Claus Auction. Suppose
increase in expected revenue. there are just two buyers, and the seller
We now consider three quite different auc- announces the following auction rules.
tion rules. (i) A buyer who submits a bid b?v.
receives from the seller an amount S(b)
Example 1: Sad Loser Auction. Suppose there Jfb. F(v)dv.
are just two buyers and the seller announces (ii) The high bidder obtains the good for
the following auction rules. his bid price so that his net payment is
(i) Each buyer paying an entry fee c is b -S(b).
eligible to submit as his bid any positive real
number.10 One can confirm that the equilibrium strategy
(ii) The high bidder receives the good but of each buyer is to bid his reservation value.
retains his bid. Suppose that the second buyer bids b2 v2.
(iii) The lower bidder (if there is one) Then if buyer 1 bids bI, his expected profit is
loses his bid. given by

It is tempting to conjecture that there is no Pr{b1 is high bid}(v, -bl)+S(bl)

equilibrium bidding strategy for this set of
rules. However, not only is this incorrect, but -F(b,)(v, -b,)+S(b,)
the equilibrium bidding strategy is readily
derived. Under rules (i)-(iii), the expected It is straightforward to check that this
gain of the typical buyer is expression is maximized at b Iv,.
In this auction the seller's expected net
11(x, vi) =vi F(x) -b(x)(I -F(x)) revenue is the expected value of the higher of
the two bids less the seller's expected pay-
For b(v) to be the equilibrium bidding ments. With v*= 1/2, S(b)>b2/2- 1/8.
strategy, this gain is maximized by setting Moreover, each buyer bids his reservation
x= v so that the buyer's expected payment is value; therefore the seller's expected gross
b(vi)(I -F(vi)). Then if F(v) =v and c= 1/4, receipts and payments are easily computed.
it can be confirmed from equation (14) that Once again it can be confirmed that expected
v. 1/2, and from equation (8b) that net revenue is 5/12, exactly the sum the
seller can expect from the high bid auction.
Since the implication of Proposition 1 is
b(v) - (v I/4) for viv 21 that many seemingly different auction tech-
2(l1-v1)
niques lead to the same ultimate results, it is
Thus, as in the optimal high bid auction, any important to illustrate the range of excep-
buyer with reservation value less than 1/2 tions.
remains out of the auction. The bids of those
with higher reservation values are strictly Example 3: Matching Auction. Suppose there
increasing in v and increase without bound are just two buyers and the seller employs
as v approaches 1! Nevertheless, it is easy to the following auction rules.
confirm that expected seller revenue under (i) There is a single round of bidding.
this scheme matches that of the high bid and Buyer 1 is given the opportunity to quote a
second bid auction, cum optimal reserve price b1Iv*.
price. (ii) If buyer 1 makes a bid, buyer 2 can
Under the high bid and second bid auc- match it, if he chooses, obtaining the good
tions, only the recipient of the good gains. In for this price. If buyer 1 makes no bid, buyer
contrast, the following auction distributes a 2 can obtain the good at price v* if he
positive return to all participants. chooses. "i

'?This rules out bids such as "infinity" or "one more " For an analysis of the
matching auction when m
than my opponent." rounds of bidding are permitted, see our earlier paper.
388 THE AMERICAN ECONOMIC RE VIEW JUNE 1981

Though this auction procedure is quite com- assumption of buyer symmetry, it is shown
mon (for example, in house sales, a renter in the Appendix that the high bid auction
occupant is frequently given the right to dominates the second bid auction under
match the offer of any potential buyer), it is buyer risk aversion.
inefficient from the point of view of the
seller. In fact, in some circumstances it per- PROPOSITION 4: Suppose assumption IID
mits a buyer who values the item less highly holds and all buyers share a common utility
than his opponent to obtain the good. Thus, function displaying risk aversion. Then (i) In
it may produce an allocation of the good the second bid auction, bidders continue to bid
that is inefficient ex post. their reservation values, that is, bi = vi. (ii) In
Suppose that F(v) = v and v*.= 1/2. The the high bid auction, as bidders become more
strategy of buyer 2 is straightforward. He risk averse, they make uniformly higher bids.
matches b, if and only if V2 :b. If buyer 1 (iii) Consequently, the seller enjoys a greater
does not open the bidding, buyer 2 bids 1/2 expected profit under the high bid auction than
for the good if v2 ? 1/2. Anticipating the under the second bid auction.
behavior of buyer 2, buyer 1 bids b1 ?1/2 to
maximize It is evident that the introduction of risk
aversion does not affect the strategy domi-
( v I-bI ) Prob{buyer 2 chooses not to match) nance of bidding one's true reservation value
in a second bid auction, hence part (i). Part
Buyer 2 will not match if his reservation (iii) follows directly from part (ii) which is
value is less than bI, that is, he will not proved in the Appendix.
match with probability b1. Then buyer 1 The intuition behind these results is that
chooses b I: 1/2 to maximize his expected with risk aversion the marginal increment in
gain (v, -b,)b,. Since this expression is de- wealth associated with a successful, slightly
creasing in bI for all bI > 1/2, buyer l's lower bid is weighted less heavily than the
optimal strategy is to bid possible loss (vi -bi) if, as a result of lower-
ing the bid, the buyer is no longer the high
bidder. This leads risk-averse bidders always
{1/2 v, :- 1/2 to shade their bids less than risk-neutral
bidders.
Under risk aversion, the general equiva-
Consequently, whenever 1/2< v2 <VI, the lence result obtained in Proposition 1 no
object is awarded to buyer 2 who values it
longer holds. For instance, an auction em-
less highly than buyer 1. The expected reve-
ploying a seller reserve price will not, in
nue of the seller for this example is 3/8, a
general, be equivalent to one that specifies a
reduction of 10 percent relative to the high
buyer entry fee - even when the same
and second bid auctions.
reservation value v., below which it is not
worth bidding, is implied. Still it is natural to
III. BuyerRiskAversion explore the effect that buyer risk aversion
has on the optimal seller reserve price in the
When potential buyers are risk averse, the high bid auction. The following result is
fundamental equivalence result outlined in derived in the Appendix.
Section I is no longer valid.'2 Retaining the
PROPOSITION 5: Suppose assumption IID
12
ther authors have also considered the effects of
holds and all buyers share a common cardinal
risk aversion on bidding. Butters derives Propositions 4 utility function. Then, in the high bid auction,
and 5 for the special case in which buyers exhibit the optimal seller reserve price is a declining
constant relative risk aversion. Charles Holt examines function of the degree of risk aversion.
the effects of risk aversion in the closely related problem
of bidding on incentive contracts. Steven Matthews
compares high bid and second bid auctions when seller The proposition is intuitively plausible in
and buyers are risk averse. view of the fact that as buyers become risk
VOL. 71 NO. 3 RILEY A ND SA MUELSON: OPTIMA L A UCTIONS 389

aversein the extreme,the amount by which which the good is sold to the buyer with the
they will shade their reservationvalues ap- highest reservationvalue v, if the good is
proaches zero, b(vi) -* vi. Naturally, the seller sold at all. Under only moderaterestrictions
can do no better than to announcehis per- on the form of the distributionF(v), it can
sonal valuationas his reserveprice, bo= vo. be shown that it is never optimal to utilize a
To quote a higher price cannot "push up" rule in which the winner might be someone
buyer offers and risks the loss of beneficial other than the buyer with the highest v.
sales. Of course when bo = vo and bi = vi, the However, when these restrictions are not
high bid auctionis also efficientex post. satisfied, a stochasticauction is optimal. In
such an auction a lottery is employed to
IV. ConcludingRemarks allocate the good when buyer reservation
valuesfall in specifiedranges.'4
While a generalresult concerningthe de- Dropping the assumptionof buyer sym-
sign of optimal auctions under uncertainty metryalso causescomplicationsin the analy-
has been presented,it is importantto point sis. The derivationof the class of optimal
out the limitationsand special assumptions auctions relied explicitlyon the existenceof
of the [Link] have assumedthat: a common equilibrium bidding strategy.
(a) A single indivisiblegood is to be sold Without this, these propositions no longer
to the highestbidder. hold. The asymmetricmodel,thoughfar more
(b) The greatera bidder'sreservationvalue complex, is nevertheless amenable to the
the morehe will bid for the good. basic approach developed herein. Suppose
(c) Buyerroles are symmetrical(i.e., buyer the reservationpricesof the buyersare drawn
values are drawn from a common distribu- from the independent distributions,Fl, F2,
tion) and each buyeris risk neutral. .. Fn. Some partial results from this setting
(d) Buyervalues are independent. suggest a basic [Link] optimal auc-
Additional difficulties are raised when tion extends the asymmetry of the buyer
multiplegoods are auctionedor when a di- roles to the allocationrule itself. The assign-
visible good must be [Link] buyer ment of the good and the appropriatebuyer
valuationsare additiveand income indepen- paymentwill depend not only on the list of
dent, auctioningthe goods in sequencewill offers,but also on the identitiesof the buyers
be inefficient (ex post and ex ante). When who submit the bids. In short, an optimal
multiple goods are auctioned, each buyer auctionunderasymmetricconditionsviolates
shouldlogicallysubmita bid for each subset the principleof buyer anonymity.
of goods. Roughly speaking,the seller will As pointed out earlier,the assumptionof
allocate goods to maximize revenue under risk neutralityis crucialto our generalequiv-
one of a numberof auction [Link] the alence [Link] risk neutrality,the seller
case of a divisible good, each buyer will can do no better than to employ the second
submit a "demandschedule"indicating the bid auctionwith an optimalreserveprice. In
price he is willing to pay for any given this auction, buyers will have no difficulty
quantity of the good. The seller must for- formulatingan [Link]
mulate an auction rule which specifies the need they know the form of the distribution
allocationof the good and appropriatepay- function F(v). Against any distributionof
ment of buyers. In either instance the de- opponents' bids, each buyer's dominant
termination of optimal auctions for these strategyis to bid his reservationvalue. The
more general environmentslies beyond the clear advantageof the second bid auctionis
boundsof the presentanalysis.'3 that it economizeson the informationeach
Given assumption(b) it follows that the buyer requires to bid optimally. Further-
family of auctions consideredare those in more, Proposition3 indicatesthat the seller

'3Harrisand Raviv (1981) and Maskin and Riley 14For a presentationof the more generalframework
(1980b)analyzeoptimalauctionsfor differentclassesof from which the optimal stochasticauction can be de-
demandcurves. rivedsee Myersonor Maskinand Riley (1980a).
390 THE AMERICAN ECONOMIC REVIEW JUNE 1981

can formulate an optimal reserve price policy In most real world settings, we would ex-
without knowledge of the number of buyers pect that a good's economic value to a poten-
who might enter the auction.'" tial buyer consists of two parts- a value
Finally, one must consider the appropria- element which is common to all market par-
teness of the model's most basic assumption, ticipants and one which is buyer specific.
value independence. The analysis has as- Shell's recent $3.6 billion purchase of
sumed that each buyer is informed of his Belridge Oil- the most expensive in U.S.
own reservation price and, more important, history- is a dramatic example. Belridge was
that this price conveys no information about sold by closed sealed bid auction in which
any other buyer's value. A different auction twenty-odd prospective buyers participated.
model has been applied to bidding for off- Differences in bids presumably reflected (i)
shore oil leases.'6 Here, a tract being auc- differences in beliefs about the value of
tioned is assumed to have a common value Belridge's oil holdings (differences that might
for all parties. The tract value is unknown, have been dissipated through pooling of in-
though buyers may possess (differing) sam- formation) and (ii) differences in the extent
ple information allowing inferences about to which Belridge's operations comple-
this value. In this setting, each buyer must mented bidders' other activities. It is easy to
determine a strategy for acquiring informa- imagine, though not to solve, a hybrid model
tion concerning the value of the tract and for specifying both dependent and independent
submitting a bid based on a correct estimate components of buyer reservation values. A
of this value. These features have a direct formal analysis of optimal auction design in
influence on the determination of an optimal this more general environment remains to be
auction and raise additional policy issues. undertaken.
(Should the seller maintain a stake in an
awarded tract for the purpose of risk shar- APPENDIX
ing? Should the seller undertake measures to
facilitate information acquisition or to allow PROPOSITION 4 (ii): Suppose assumption
information pooling?)'7 IID holds and all buyers share a common
utility function displaying risk aversion. Then
15The largest auction houses (for example, Sotheby
Park Bernet, Inc. and Christie's) employ the English
in the high bid auction, as bidders become
auction (combining its open bid and sealed bid forms) more risk averse, they make uniformlyhigher
to sell rare and valuable items (art, antiques, and jewelry). bids.
A buyer can bid personally for an item on the day of the
auction or can submit a prior written offer, designating PROOF:
a representative from the auction house to bid on his
behalf. This same procedure establishes a silent seller
Let b(v) be the common equilibrium
reserve price, since a house representative is instructed strategy of n risk averse buyers, each of
to buy back the good if the sale price is insufficient. It is whom has the same von Neumann-
a common observation that the competitive features of Morgenstern utility function u(x). We as-
the open ascending auction serve to elevate buyer offers
sume that u(x) is a strictly increasing, con-
(above their prior values). This implies that the open
ascending auction enjoys a practical advantage over the cave function of x and normalize so that
sealed bid version. The "mixed" auction allows written u(O)=0. With all other buyers using the equi-
bids in order to promote the greatest possible participa- librium bidding strategy and buyer j bidding
tion while maintaining the "uplifting" features of the b(x), j's expected utility is
open ascending auction.
16See, for example, Robert Wilson (1975) and Mat-
thew Oren and Albert Williams. In this model buyers
begin with common prior beliefs about the value of a (A1) F (x)u(vj-b(x))
resource but have different posterior beliefs as a result of
independent sampling. For discussion of auctions in
which buyers have different prior beliefs, see Wilson For b(x) to be the equilibrium strategy, (Al)
(1967). must have its maximum at x=vJ. Differenti-
17For a discussion of the incentives for the seller to
make information public, see Paul Milgrom and Robert ating with respect to x and setting the deriva-
Weber. tive equal to zero at x = vj, we have the
VOL. 71 NO. 3 RILEYAND SAMUELSON: OPTIMAL A UCTIONS 391

necessarycondition PROPOSITION 5: Suppose assumption IID

holds and all buyers share a common von
(n - )Fn-2(vj)F'(vj)u(vj -b(vj)) Neumann-Morgenstern utility function. Then
-I ~~db in the high bid auction, the optimal seller
-Fn n (vj)u'(vj-b( vJ))d =0 reserve price is a declining function of the
degree of risk aversion.
Rearrangingyields the followingdifferential
equationfor b(v) PROOF:
The method of proof is to compare the
F(v) u'(v-b) effect of a change in the reserveprice v. on
the equilibrium bid function b = b( v, v.) for
different degrees of risk [Link]
With reserveprice bo=v. we also have the sellerrevenue,R(v*), is the expectedvalueof
boundarycondition the highestrankedbid, that is,
(A3) b(v*)=v*
R(v*) b(v, v*)dFn'(v)
We wish to comparethe solution for two
different utility functions, ul(*) and u2(*)
where the latter exhibits a higher degree of Then the net advantageto the sellerif utility
risk aversion,that is, is u2( ) ratherthan ul(.) can be expressedas
(A4) - u' ( ) U'( -U l'(X)/U'(X)20:- R2(v*)-RI(v*)

By inspection of (A2), if we can establish

= Jv[b2(V v*)-bl(v, v*)] dFn- I(V)
that
(A5) p(x)=U2(X)I/u2(X)
Differentiatingwith respectto v* we have
-ul(x)/u'(x)>0 for x>O
(A8) R'2(v*)-Rl(v*)
then b'(v)>b'(v) and hence b2(v)>b (v)
for all v>v.. To demonstrate(A5) we note
first that, since u(O)= 0 and u(x) is strictly
increasing,
lv[av* av*] ()

It sufficesto show that the bracketedexpres-

(A6) =(0) > ((O)= forallx>O sion in (A8) is negative, for then R'2(v*)is
UX) u'(0)
negativewhen R1(v*) is zero.
Inequality(A5) holds if we can establishthat From (A2), the equilibriumbid function
for all x such that 4(x)= 0, (x) is strictly b(v,v*) is the solution to
[Link](A5) we have
(
(A9) a b(v, v*)= (n -1) F'(v) u'(v-b)
u(v-b)
(A7) 0 (x)= u2 )(U)( l) ul
with the boundarycondition,
From (A4)- (A6), x > 0 and ?(x) =0 implies
that ?'(x) > 0. Moreover,differentiating(A7) (A I0) b(v*, v*)= v*
and settingx = 0 we also have
O"(0)>0'(0) =0 Assumingu(*) is twice differentiable,we can
differentiate(A9) with respectto the reserve
Thus p(x) is strictlyincreasingat x = 0. pricev* and so obtain the followingdifferen-
392 THE AMERICAN ECONOMIC REVIEW JUNE 1981

tial equationfor ab/av. Pricing Schemes with Demand Uncer-

tainty," Amer. Econ. Rev., June 1981, 71,
(All) 3 l(nb 1 F(v) 347-65.
C. A. Holt, Jr., "Competitive Bidding for Con-
tracts under Alternative Auction Proce-
dures," J. Polit. Econ., June 1980, 88, 433-
445.
E. S. MaskinandJ. G. Riley,(1980a) "Auction-
From (A4) and (A5) the bracketin (Al 1) ing an Indivisible Object," working paper,
is larger for the utility function u2(x) ex- Kennedy School Government, Harvard
hibitinggreaterrisk [Link] if we can Univ. 1980.
establishthat ab2/av*=ab1/av*>O at v= and , (1980b) "Price Dis-
v*, it will follow from (Al 1) that crimination and Bundling, Monopoly Sell-
ing Strategies when Information is Incom-
a lab2 a labi plete," mimeo., MIT, 1980.
S. Matthews, "Risk Aversion and the Ef-
aTVv*J
>TV av* ficiency of First and Second Price Auc-
tions," mimeo., Univ. Illinois, 1979.
for v>v* and hencethat ab2/av* >ab,/av* P. R. Milgromand R. J. Weber, "A Theory of
for v>v*. Auctions and Competitive Bidding," work-
From (A10) we have, ing paper, Grad. School Management,
Northwestern Univ. 1980.
(A12) ab (V,
v*)IV=V + ab (V,v*)I-V =1 R. B. Myerson, "Optimal Auction Design,"
Math. Operations Res., 1981, forthcoming.
M. E. Oren and A. C. Williams,"On Competi-
Since b(v*,v*)=v* and u(O)=O, it follows tive Bidding," Operations Research, Nov.-
from (A2) that for any concave utility func- Dec. 1975, 23, 1072-79.
tion and any v* >0, the firsttermin (A12) is A. Ortega-Reichert,"Models for Competitive
zero. Then the second termin (Al 2) is equal Bidding Under Uncertainty," unpublished
to unity for both ul(x) and u2(x). doctoral dissertation, Stanford Univ. 1968.
J. G. Riley and W. F. Samuelson, "Optimal
REFERENCES Auctions," working paper, Univ. Cali-
fornia-Los Angeles 1979.
G. R. Butters, "Equilibrium Price Distribu- W. F. Samuelson, "Models of Competitive
tions and the Economics of Information," Bidding," unpublished doctoral disserta-
unpublished doctoral dissertation, Univ. tion, Harvard Univ. 1978.
Chicago 1975. R. M. Stark and M. H. Rothkopf,"Competitive
R. Engelbrecht-Wiggans,"Auctions and Bi- Bidding: A comprehensive Bibliography,"
dding Models: A Survey," Management Operations Res., Mar.-Apr. 1979, 27, 364-
Sci., Feb. 1980, 26, 119-42. 91.
J. C. Harsanyi,"Games with Incomplete In- W. Vickrey, "Counterspeculation, Auctions
formation Played by Bayesian Players," and Competitive Sealed Tenders," J. Fi-
Parts I; II; III, Management Sci., Nov. nance, Mar. 1961, 16, 8-37.
1967; Jan. 1968; Mar. 1968, 14, 159-82; R. B. Wilson, "Competitive Bidding with
321-34; 486-502. Asymmetrical Information," Management
M. Harris and A. Raviv, "Allocation Mecha- Sci., July 1967, 13, A816-20.
nisms and the Design of Auctions," work- , "On the Incentive for Information
ing paper, Grad. School Ind. Admin., Acquisition in Competitive Bidding with
Carnegie-Mellon Univ. 1979. Asymmetrical Information," report, dept.
and , "A Theory of Monopoly econ., Stanford Univ. 1975.