Upload
chrisroby
View
269
Download
0
Embed Size (px)
Citation preview
8/10/2019 5.1 Auction Theory Slides
1/35
Today
Auction Theory Basics
1. Bidding and Equilibria in IPV model
2. Revenue Equivalence
3. General Model: Affiliated Values
4. Linkage Principle: Revenue RankingAd Auctions: Sponsored Search
8/10/2019 5.1 Auction Theory Slides
2/35
Auctions: The Independent Private Values Model
Basic Auction Environment
Bidders i= 1,...,n
One object to be sold
Bidder i observes a signal Si F(), with typical realizationsi[s, s]. Assume F is continuous.
Bidders signals S1,...,Sn are independent.
Bidder is value vi(si) =si.
A set of auction rules will give rise to a game between the bidders.
8/10/2019 5.1 Auction Theory Slides
3/35
The IPV Model
Two important features of the model:
Bidder is information (signal) is independent of bidder js infor-mation (signal).
Bidder is value is independent of bidder js information (i.e. pri-vate values).
8/10/2019 5.1 Auction Theory Slides
4/35
Vickrey (Second-Price) Auction
Auction Rules:
Bidders are asked to submit sealed bids b1,...,bn.
Bidder who submits the highest bid wins the object.
Winner pays the amount of the second highest bid.
Proposition 1In a second price auction, it is a (weakly) domi-nant strategy to bid ones value, bi(si) =si.
Proof. Biddingbimeansi will win if and only if the price is below bi.Bid bi > si sometimes win at price above value.Bid bi < si sometimes lose at price below value.
8/10/2019 5.1 Auction Theory Slides
5/35
Expected Vickrey Auction Revenue
Sellers revenue equals second highest value.
Let Si:n denote theith highest ofn draws from distributionF.
Sellers expected revenue is
E
S2:n
.
8/10/2019 5.1 Auction Theory Slides
6/35
Open Ascending Auction
Auction Rules:
Prices rise continuously from zero.
Bidders have the option to drop out at any point.
Auction ends when only one bidder remains.
Winner pays the ending price.
With private values, just like a Vickrey auction!
8/10/2019 5.1 Auction Theory Slides
7/35
Sealed Bid (First-Price) Auction
Auction Rules:
Bidders submit sealed bids b1,...,bn.
Bidders who submits the highest bid wins the object.
Winner pays his own bid.
Bidders will want to shade bids below their values.
Well look at two approaches to solving for symmetric equilibrium bid-ding strategies.
8/10/2019 5.1 Auction Theory Slides
8/35
FOC Approach to First Price Auction
Suppose bidders j =i bid bj =b(sj), b() increasing.
Bidderis expected payoff:
U(bi, si) = (si bi) Pr [bj =b(Sj) bi, j =i]
Bidderichooses bito solve:
maxbi
(si bi) Fn1
b1(bi)
.
First order condition:
0 = (si bi) (n1)Fn2
b1(bi)
f
b1(bi) 1
b(b1(bi))Fn1
b1(bi)
8/10/2019 5.1 Auction Theory Slides
9/35
FOC Approach to First Price Auction
At symmetric equilibrium, bi=b(si), first order condition is:
b(s) = (s b(s)) (n 1)f(s)
F(s).
Solve using the boundary condition b(s) =s:
b(s) =s
sis
Fn1(s)ds
Fn1(s) .
8/10/2019 5.1 Auction Theory Slides
10/35
Envelope Approach to First Price Auction
Suppose there is a symmetric increasing equilibrium: bi=b(si).
Then is equilibrium payoff given signal si:
U(si) = (si b(si)) Fn1(si). (1)
As b(si) is optimal fori:
U(si) = maxbi
(si bi) Fn1(b1(bi)).
By the envelope theorem:d
dsU(s)
s=si
=Fn1(b1(b(si)) =Fn1(si)
8/10/2019 5.1 Auction Theory Slides
11/35
Envelope Approach to First Price Auction
Rewriting the envelope formula:
U(si) =U(s) +
sis
Fn1(s)ds. (2)
Asb(s) is increasing, bidder with signal snever winsU(s) = 0.
Equating the two representations ofU(si):
b(s) =s
sis
Fn1(s)ds
Fn1
(s)
.
8/10/2019 5.1 Auction Theory Slides
12/35
Equilibrium in the First Price Auction
Weve found necessary conditions for symmetric equilibrium.
Verifying that b(s)isan equilibrium isnt too hard.
Also can show that equilibrium is unique.
8/10/2019 5.1 Auction Theory Slides
13/35
Expected First Price Auction Revenue
Revenue is high bid b(s1:n); expected revenue is E
b(S1:n)
.
b(s) =s
ss
Fn1(s)ds
Fn1
(s)
= 1
Fn1
(s)
s
s
sdFn1(s) = E S1:n1|S1:n1 s .
Expected revenue is:
E
b
S1:n
= E
S1:n1|S1:n1 S1:n
= E
S2:n
.
First and second price auction yield same expected revenue.
8/10/2019 5.1 Auction Theory Slides
14/35
Revenue Equivalence Theorem
Theorem (Revenue Equivalence)Supposenbidders have valuess1,...,snidentically and independently distributed with cdfF(). Thenany equilibrium of any auction game in which(i) the bidder with thehighest value wins the object, and(ii) a bidder with value sgets zero
profits, generates the same revenue in expectation.
8/10/2019 5.1 Auction Theory Slides
15/35
Proof of the RET
Consider any auction where bidders submit bids b1,...,bn, and wherethe auction rule specifies for all i,
xi : B1 ... Bn[0, 1]
ti : B1 ... Bn R,
xi() gives probabilityiwill get the object.
ti() gives is payment as a function of (b1,...,bn).
8/10/2019 5.1 Auction Theory Slides
16/35
Proof of the RET
Bidderis expected payoff is:
Ui(si, bi) =siEbi[xi(bi, bi)] Ebi[ti(bi, bi)] .
Letbi() , bi() denote an equilibrium of the auction game.
Bidderisequilibriumpayoff is:
Ui(si) =Ui(si, b(si)) =siFn1 (si) Esi[ti(bi(si), bi(si)] ,
where because highest value wins:
Esi[xi(b(si), b(si))] =Fn1(si).
8/10/2019 5.1 Auction Theory Slides
17/35
8/10/2019 5.1 Auction Theory Slides
18/35
Proof of the RET, continued
Combine the expressions for Ui(si) to get:
Esi[ti(bi, bi)] =siFn1 (si)
sis
Fn1(s)ds=
sis
sdFn1(s),
If auction satisfies(i) and(ii), is expected payment given si is:
E
S1:n1 |S1:n1 < si
= E
S2:n |S1:n =si
.
Thus expected revenue is constant:
E [Revenue] =nEsi[is expected payment| si] = E
S2:n
.
8/10/2019 5.1 Auction Theory Slides
19/35
Using the Revenue Equivalence Theorem
Many applications of the revenue equivalence theorem.
If the auction equilibrium satisfies(i) and(ii), we know:
Expected Revenue
Expected Bidder Profits (payoff equivalence)
Expected Total Surplus
We can use this to solve for auction equilibria.
8/10/2019 5.1 Auction Theory Slides
20/35
The All-Pay Auction
Bidders 1, . . ,n
Values s1,...,sn, i.i.d. with cdfF
Bidders submit bids b1,...,bn
Bidder who submits the highest bid gets the object.Every bidder must pay his or her bid.
8/10/2019 5.1 Auction Theory Slides
21/35
Solving the All-Pay Auction
Suppose all pay auction has a symmetric equilibrium with an in-creasing strategy bA(s). In this equilibrium, is expected payoff givenvaluesi will be:
U(si) =siFn1
(si) bA
(si) = si
s Fn1
(s)ds
So
bA(s) =sFn1(s)
ss
Fn1(s)ds
8/10/2019 5.1 Auction Theory Slides
22/35
Other Revenue Equivalent Auctions
1. The English (oral ascending) auction. All bidders start in the auc-tion with a price of zero. Price rises continuously; bidders may dropout at any point in time. Once they drop out, they cannot re-enter.Auction ends when only one bidder is left; this bidder pays the price
at which the second-to-last bidder dropped out.2. The Dutch (descending price) auction. The price starts at a very
high level and drops continuously. At any point in time, a biddercan stop the auction, and pay the current price. Then the auctionends.
8/10/2019 5.1 Auction Theory Slides
23/35
Common Value Auctions
Generalize the model to allow for the possibility that:
1. Learning bidder js information could cause bidder i to re-assesshis estimate of how much he values the object,
2. The information of i and jis not independent (when js estimateis high, is is also likely to be high).
Examples:
Selling natural resources such as oil or timber.
Selling financial assets such as treasury bills.
Selling a company.
8/10/2019 5.1 Auction Theory Slides
24/35
A General Model
Bidders i= 1,...,n
Signals S1,...,Snwith joint density f()
Value to bidder iis v(si, si).
8/10/2019 5.1 Auction Theory Slides
25/35
Affiliated Signals
Signals are(i) exchangeable, and(ii) affiliated.
Signals areexchangeable: s is permutation ofsf(s) =f(s).
Signals areaffiliated: For any s1 > s1, s
2 > s
2, we have
f(s1|s2)
f(s1|s2)
f(s1|s2)
f(s1|s2)
.
i.e. sj|si has monotone likelihood ratio property.
This implies, but is not equivalent to
F(s1|s2) F(s1|s
2)
for all s1. (FOSD)
8/10/2019 5.1 Auction Theory Slides
26/35
Examples of the General Model
1. The IPV model.
2. Pure common value with conditionally independent signals.
Example: Si = V +i, where 1, . . ,n are independent, andv(si, si) = E[V|s1,...,sn].
3. Interdependent values, independent signals model.
Example: vi(si, si) =si+
j=i sj, with 1, with S1,...,Snindependent.
Note: With interdependent values, i must realize that js bid is in-formative aboutis own value. This leads to the winners curse.
8/10/2019 5.1 Auction Theory Slides
27/35
Second Price Auction
Look for a symmetric increasing equilibrium bid strategy b(s).
Let si denote the highest signal of bidders j =i.
Ifbib(si), iwins and pays b(si).
Bidder is problem:
maxbi
ss
ESi
v(si, Si)| si, S
i =si b(si)
1{b(si)bi}f(s
i|si)dsi
or
maxbi
b1(bi)s
ESi
v(si, Si)| si, S
i =si b(si)
dF(si|si)
8/10/2019 5.1 Auction Theory Slides
28/35
Second Price Auction
The first order condition for this problem is:
0 = 1
b(b1(bi)) ESi
v(si, Si)| si, S
i =b1(bi)
b(b1(bi))
f(b1(bi)|si)
or simplifying:
bi=b(si) = ESi
v(si, Si)| si, max
j=isj =si
8/10/2019 5.1 Auction Theory Slides
29/35
Failure of Revenue Equivalence Theorem
The RET does not hold in this more general environment.
Basic idea: bidders profits are due to information rents. Whensignals are correlated, information rents can be reduced if moreinformation is used in setting the price.
Example: english auction generates more revenue than a secondprice auction.
8/10/2019 5.1 Auction Theory Slides
30/35
Linkage Principle Intuition
Bidder profits (rents) are derived from their private information
Statistical linkages align the price more closely to the buyers will-ingness to pay.
Rents associated with private information fall
Revenue rises
8/10/2019 5.1 Auction Theory Slides
31/35
Linkage Principle
Standard auctionsA, Bwith symmetric, increasing eqabA( ), bB( )
LetWA(s, s) denote the expected price paid by bidder 1 if he is thewinning bidder when he receives a signal s but bids as if his signalwere s.
Let WA2 (s, s) denote the partial w.r.t. the second argument. Thatis, W2 measures how much the expected price increases with in-creases in the bidders valuation, holding bidding strategy constant.
Theorem 1 Suppose that for alls,
W
A
2 (s, s) W
B
2 (s, s).Then the expected revenue inA is at least as large as the expected
revenue inB.
8/10/2019 5.1 Auction Theory Slides
32/35
Revenue Ranking: First vs Second Price Auction
For the first price (I) auction WI(s, s) =b1(s), so
WI2 (s, s) = 0.
For the second price (II) auction,
WII
2 (s, s)> 0,
because conditional on winning by bidding as if his signal is s...
having a higher signal smeans that others also have a higher signal...
so the expected price paid (the second highest bid) will be higher.
From the theorem, it follows that the second price auction yieldsgreater revenue than the first price auction.
8/10/2019 5.1 Auction Theory Slides
33/35
Linking Price to Seller Information
If the seller has private informationS0about the object that is affil-iated with the signals of the buyers, the seller can increase revenue bycommitting to a policy to reveal this information.
Theorem 2Verifiably revealing information variableS0raises theexpected price both in the first auction and second price auctions.Among all policies for full or partial revelation of information, thepolicy of full revelation maximizes the expected price.
8/10/2019 5.1 Auction Theory Slides
34/35
Linking Price to Seller Information
Interestingly, the policy of full information can be expected to emergeeven when the seller cannot commit to an information policy.
Theorem 3 If a seller must decide, after observing S0, whether
to report it, and if his report is verifiable, then in any equilibrium,he always reportsS0 regardless of the value.
Why would a seller report ifs0is low? Because of affiliation, biddersignals will be low with or without this information. Revealing willnonetheless link this information to price.
(If the seller observed s0 *and* bidder signals, then he may prefernot to reveal.)
8/10/2019 5.1 Auction Theory Slides
35/35
Royalties
When the value can be observed by all parties ex-post (even imper-fectly), expected revenue is higher when part of the price is a royaltybased on the observed value.
Again, a linkage of available information to final price.