SØK/ECON 535 Imperfect Competition and Strategic Interaction AUCTIONS Lecture notes 05.11.02 Introduction Auctions are commonly used as a selling or buying mechanism: art, antiques, used cars oil, metals, coffee, fish, electricity treasury bills, certificates procurement, privatisation concessions, licences ”Money out of thin air” Allocation of spectrum rights UMTS auctions Great Britain: 5 licenses, 37.1 billion Euro, 129.4 Euro/pop/license The Netherlands: 5 licenses, 2.7 billion Euro, 33.7 Euro/pop/license Issues resource, property rights, allocation auction or ‘beauty contest’? auction format incumbents, entrants, market structure, competition The Market Mechanism in detail One agent with complete market power can determine trading rules/institution auction versus negotiations Asymmetric information do not know value/cost of opponent(s) Formalised trading rules ‘anti corruption mechanism’ Classification of auction types number of units simultaneous or sequential trading sale and/or purchase price determination Game theoretic models strategic interaction well-defined ‘rules of the game’ normative versus positive theory maximisation of income, or minimisation of costs efficiency Elementary auction theory Overview a simple model auction formats The Revenue Equivalence Theorem The base model One object for sale A given number of risk-neutral bidders (n) No reservation (minimum) price more precisely: reservation price equal to smallest possible valuation 2 Valuations Let vi be the value of the object to (maximum willingness to pay of) bidder i, i = 1,2,...,n each bidder knows his or her valuation; common knowledge that vi is drawn from a distribution Fi with support [v ,v ] . We assume symmetry: Fi(v) = F(v), i = 1,2,...,n example: uniform distribution on [0,1] Eksperiment no 1 – sealed-bid first-price auction Valuations = birth day January 1 = 1, February 3 = 31+3 = 34, May 17 = 31+28+31+30+17 = 137 and so on Sealed bids (secret bidding) each bidder notes his or her bid Maximum bid wins First-price auction the winner pays a price equal to his or her bid Pay off = valuation – price, if winner; 0 otherwise Sealed-bid first-price auction – optimal strategies Risk neutral bidders choose the strategy that maximises expected pay off. Let Gi(b) be the probability that bidder i wins with bid b Expected pay off for i: π i (b,v i ) = [v i − b ]Gi (b ) F Necessary condition for optimal bid: 3 F dπ i = [v i − b ]Gi′(b ) − Gi (b ) = 0 db or b = vi − Gi (b ) Gi ′ (b ) Note: bid is shaded below valuation. Sealed-bid first-price auction - equilibrium Let bi(v) be i’s strategy, i.e. the bid as a function of valuation. Symmetry: suppose all follow the same strategy, i.e., bi(v) = b(v), all i. The probability of winning with a bid b is Gi (b ) = G(b ) = [F (v )] n −1 Optimal strategy: b(v ) = v − G (b) b′ (v ) F (v ) =v − G′ ( b ) n − 1 F ′(v ) Example: uniform distribution: b (v ) = n n −1 n . n −1 v n 2 3 5 10 25 50 100 ∞ 50% 67% 80% 90% 96% 98% 99% 100% Experiment no 1 cont – sealed-bid second-price auction Sealed-bid (secret) bidding Maximum bid wins Winner pays a price equal to the second-highest bid (i.e., highest loosing bid) Sealed-bid second-price auction – optimal strategies Dominant strategy to bid valuation (b = vi) b > vi: risk paying price higher than valuation b < vi: risk loosing even though price is less than valuation 4 Vickrey auction price paid independent of own bid bid affects probability of winning only Experiment no 1 cont – open descending auction Price reduced from a high value until only one bidder accepts. Winner pays a price equal to accepted price. This auction format is used in, for instance, the Netherlands (flowers), Israel (fish) and Canada (tobacco) Open ascending auction – optimal strategies Gevinst ved å vente på en reduksjon i salgsprisen fra b til b-db (db er ”liten”): dπ i = Gi (b )db − [v i − b ]Gi′ ( b ) db N N Optimalt å stoppe når dπ i = 0 , hvilket impliserer at optimalt bud er b = vi − Gi ( b ) Gi′ ( b ) Nederlandsk auksjon strategisk ekvivalent med lukket førsteprisauksjon ”åpen førsteprisauksjon” Experiment no 1 cont – open ascending auction English auction bidders over-bid each other alternatively, an auctioneer increases the price until only one bidder accepts Japanese variant price increased gradually bidders who have withdrawn cannot re-enter Winner pays final price 5 the highest price accepted by the penultimate bidder English auction – optimal strategies Dominat strategy to accept all bids below valuation English auction equivalent to sealed-bid second-price auction not strategically equivalent because the set of possible strategies is different (in English auction strategies can be conditioned on competitors’ behaviour) Bidding strategies – summary Private values: vi, i = 1,2,...,n, independently (identically) distributed Sealed-bid first-price auction equilibrium strategies (uniform distribution): b(v ) = n −1 v n Sealed-bid second-price auctions dominant strategy to bid valuation: b(v) = v Open descending (Dutch) auction strategically equivalent to sealed-bid first-price auction Open ascending (English) auction equivalent to sealed-bid second-price auction Selling prices Sealed-bid first-price auction F 1 expected price equal to expected highest bid: EP = ∫ b(v )dF (v ) , where F 1 = [F ] is distribution of highest valuation. n Sealed-bid second-price auction expected price equal to expected second-highest valuation: n n −1 EP A = ∫ vdF 2 (v ) , where F 2 = [F ] + [ n − 1][1 − F ][F ] is the distribution of the second-highest valuation. 6 F A N E Uniform distribution: EP = EP = EP = EP = n −1 n +1 n 2 3 5 10 25 50 100 ∞ n −1 n +1 33% 50% 67% 82% 92% 96% 98% 100% The Revenue Equivalence Theorem We are looking at mechanisms for determining which of n bidders should receive an object and which prices bidders (incl. losers) should pay. Assume each of the n potential risk-neutral bidders has a privately known valuation for the object drawn from the same, strictly increasing and continuous distribution function F on [v ,v ] . Then it can be shown that any mechanism that (i) allocates the object to the bidder with the highest valuation, and (ii) bidders with the lowest possible valuation v obtains zero payoff, results in the same expected selling price and the same expected surplus for a bidder with valuation v. Examples of auctions that results in same selling price: sealed-bid first-price, sealed-bid second-price, English, Dutch, all pay Even if these formats result in the same expected selling price they do not necessarily lead to the highest possible selling price if the object must be sold, they do provide maximum price; mechanisms that lead to higher price must involve positive probability for either (i) bidder with highest valuation does not obtain the object or (ii) object is not sold even if there are bidders with valuations higher than that of the seller. Generalisation of the model to correlated (affiliated) valuations risk aversion 7 asymmetries Optimal (revenue-maximising) auctions Issues auction format reservation (minimum) price competition (entry) information provision Auction format The Revenue Equivalence Theorem Risk Strategic complexity Co-ordinated bidder behaviour (collusion) Seller risk Given the highest value among bidders, the selling price is certain in the Dutch auction (equal to the bid of the bidder with the highest valuation), but uncertain in the English auction (equal to the second-highest bid). Consequently, the selling price in the Dutch auction (second-order) stochastically dominates that of the English auction. A risk averse bidder will therefore prefer a Dutch to an English auction format by a corresponding argument, he or she would prefer the sealed-bid firstprice format to the sealed-bid second-price format. Bidder risk In the English and sealed-bid second-price auction buyers’ attitude towards risk does not affect behaviour bidding at valuation is a dominant strategy. 8 However, in the Dutch and sealed-bid first-price auction, risk aversion leads to more aggressive bidding aggressive bidding reduces a bidder’s risk of losing the auction when the price is below his or her valuation. Uncertainty about participation (number of participants and their types) may also lead to more aggressive bidding (and hence higher selling price). Strategic complexity The English and sealed-bid second-price auctions are strategically simple there is a dominant strategy; although this may be more difficult to see in the sealed-bid second-price auction. I the Dutch and sealed-bid first-price auctions optimal strategies depends on expectations about competitors’ behaviour. Coordinated action among bidders By co-operating (colluding) bidders may increase their (total) gain, although such collusion involves side payments; the risk that some bidders ’cheat’. Open versus sealed-bid formats: in the English auction bidders may outbid (’punish’) bidders who do not cooperate; hence there is less to gain from ‘cheating’ and so easer to achieve collusion. First-price versus second-price format (sealed-bids): since, in the second-price format, a higher bid increases the chance of winning and not the price one has to pay, the incentive to ’cheat’ is greater ’Bidding rings’ bidders co-operate over a series of auctions; relevant for the choice between sequential and simultaneous formats. 9 Choice of auction format – summary English and sealed-bid second-price auctions strategically simpler Dutch and sealed-bid first-price auctions less price-risk for the seller more aggressive bidding among risk-averse bidders less danger of collusion among bidders Reservation price The seller determines a minimum price R such that the object is sold only if there are bids above this price. Bidding behaviour will typically be affected by reservation prices the equilibrium strategies in a sealed-bid first-price auction with symmetric and uniform distribution of valuations become: v if v ≤ R n −1 b (v ) = n − 1 R R v otherwise + n v n More aggressive bidding the higher the selling price, but also risk of no sale expected selling price in a sealed-bid first-price auction with symmetric and uniform distribution of valuations become: EP F ( R ) = 2n n −1 + R n 1 − R n +1 n +1 Optimal reservation price Trade-off between gain from more aggressive bidding against loss from no sale. Optimal (i.e. price-maximising) reservation price in a sealed-bid first-price auction is given by R F ′ (v ) 1 − F (v ) =1 10 Note: independent of number of bidders. Uniform distribution optimal reservation price: R* = ½ maximum expected selling price: EP F* n −1 1 1 = + n + 1 n + 1 2 n n 2 3 5 10 25 50 100 ∞ R=0 .3333 .5000 .6667 .8182 .9231 .9608 .9802 1 R = R* .4167 .5313 .6719 .8183 .9231 .9608 .9802 1 Competition Increased participation increases probability of high-valuation bidders, and may lead to more aggressive bidding. Increase in selling price when participation increases from n to n+1 (no reservation price, uniform distribution): EP n +1 − EP n = 2 [n + 1][n + 2] n 2 3 5 10 25 50 100 ∞ EP n .3333 .5000 .6667 .8182 .9231 .9608 .9802 1 .1667 .1000 0.0476 .0152 .0028 .0008 .0002 0 EP n +1 − EP n Participation Participation will be dependent upon expected gains from winning and 11 costs incurred from participation. Lower costs leads to more participation and hence higher expected price. It can be shown that it may be beneficial to subsidise participation, if this can be done selectively (i.e. only to some bidders); but not if the subsidy has to be given to everyone. Participation fee If potential bidders know their valutions before deciding whether to participate, then a participation fee will have the same effect as a reservation price. If the participation fee is a, among n potential bidders only those will pariticipate whose valuation v satisfies [v − v ] F (v ) n −1 ≥a marginal participant wins only if no one else participates; the fee reduces participation more the more potential participants there is. The optimal fee balances the (marginal) increase in income from the fee from the reduction in the selling price due to less competition 1 uniform distribution: a = 2 n * Reservation prices and participation fees – summary As a general rule, the seller benefits from setting an effective reservation price (or positive participation fee). An effective reservation price leads to more aggressive bidding, but a positive probability of no sale. The optimal reservation price balances the gain from more aggressive bidding from the loss of reduced probability of a sale. It may be beneficial to subsidise participation, but only if this can be done selectively. 12 Generalisations More general assumptions about relationship between bidders valuations correlated (or affiliated) values common values Experiment no 2 Value of object given by day in the year of lecturer’s (or someone in his family) birth day. Pay off = value - price Selad-bid first-price auction, tre variants: 1) Everybody symmetrically informed. 2) One bidder is informed about month. 3) One bidder is informed about date (i.e. becomes perfectly informed) Open (English) auction Common Value General model: each bidder receives a ’signal’ xi about the value of the object; the value of the object to bidder i is v i ( x1,...., xn ) , i = 1,2,...,n. Special cases private values: v i = xi , with the xi ’s independently distributed common value: v1 = v 2 = ... = v n = v . Examples of (nearly) common value objects oil fields patents objects with second-hand markets (shares, bonds, treasury bills etc) 13 The Winner’s Curse The winner is the bidder with the most optimistic estimate of the value! Others’ bids provide information about value. Optimal strategy (for a risk-neutral bidder) is to bid expected value, conditional upon winning. Bid lower the better the information of others; the more optimistic or aggressive the others are; and the more risk averse you are. Information In the private-values case information about the object is of no importance for bidding behaviour although information about competitors’ valuations or strategies may affect bidding. With correlated (affiliated) valuations providing more information about the object is always good: No news is bad news! reduces the Winner’s Curse an open (ascending) auction provides more information and hence reduces the Winner’s Curse a Japanese auction provides more information than other open and ascending formats. Revenue comparison Risk neutral bidders Risk averse agents Private values Affiliated values EP F = EP S = EP E = EP D EP E > EP S > EP F = EP D EP F = EP D > EP S = EP E ? 14 Almost common values Small asymmetries may have profound effects on bidding behaviour. Assume it is common knowledge that one of the bidders has a slightly higher valuation than the others: example: v1 = v 2 = ... = v n −1 = v , v n = v + ε (ε small). In an open ascending auction the others can only win at a price higher than their valuation they will only accept prices that gives a guaranteed gain in order not to be hit by the Winner’s Curse; if there are any participation costs at all, they will not participate. In a sealed-bid auction the asymmetry only marginally affects bidding behaviour therefore, for the seller a sealed-bid first-price auction is better than an English auction Example: ’Toeholds and takeover battles’ Efficiency – summary The object should be allocated to the bidder with the highest valuation. In the common values case, from an efficiency point-of-view it does not matter who wins. More generally, the four standard auction formats considered above are efficient only if the potential bidders are ex ante symmetric, and bid according to symmetric strategies. Choice of auction format - summary Price risk co-ordinated behaviour amond bidders correlation (affiliation) of valuations 15 asymmetries Efficiency Administrative costs open versus sealed-bid Simplicity Conclusion Not all auctions are the same! What may appear to be small differences in design may lead to wildly different outcomes. ‘The devil is in the details!’ 16