A. Existence of Pure-strategy Bayesian Nash Equilibrium

Size: px
Start display at page:

Download "A. Existence of Pure-strategy Bayesian Nash Equilibrium"

Transcription

1 Web Suppement for The Roe of Quaity in On-ine Service Markets Eena Krasnokutskaya, Kyungchu Song and Xun Tang August, 4 This web suppement is organized as foows. Section A estabishes the existence of purestrategy Bayesian Nash Equiibrium in the mode considered in the paper. Section B provides further detais regarding the identification of mode eements that are not incuded in the main text. Section C discusses extension to modes where seers participation decisions are endogenous. Section D provides detais in the nonparametric cassification procedure. Section E eaborates on the technica detais in the GMM estimation of the mode (incuding the forms of moments used in the estimation). Section F provides additiona empirica resuts that are not incuded in the main text of the paper. A. Existence of Pure-strategy Bayesian Nash Equiibrium In this section, we show PSBNE exists in our mode. To simpify the notation and exposition, we abstract away from any auction or seer characteristics that are reported in the data. Our arguments can be extended to accommodate these observed heterogeneities by conditioning the game on them. Proposition A. Suppose that the foowing conditions hod. (i) The bids are commony chosen from a cosed interva of a rea ine. (ii) The conditiona CDF of ɛ given (α, U, Q) = (ᾱ, ū, q) are uniformy continuous for a (ᾱ, ū, q) in the support of (α, U, Q). (iii) The project costs C i are private, independent and identicay distributed, and have a density function bounded uniformy away from zero. (iv) The entry costs E i are private, independent, and identicay distributed, and have a density function bounded uniformy away from zero. (v) The entry costs and the private costs are independent from each other. Then there exists a PSBNE in the game. Proof of Proposition A. We first consider a subgame in the bidding stage where a subset A N of bidders have aready decided to enter. We first show that in each subgame with A, there exists a PSBNE. For this, we invoke Coroary. of Athey () by which it suffices to show that the payoff faced by each bidder is continuous in the bid profie, and the expected payoff satisfies singe crossing of incrementa returns. Given the other payers strategies b i = (b j ) j A\{i}, the interim expected payoff for bidder i is given by U i (b i, c i ; b i ) u i (b i, b i (c i ); c i )df (c i c i ),

2 where u i (b; c i ) is the payoff for bidder i conditiona on the vector of bids b and private cost c i, and independence among the private costs are used. In our mode, the payoff u i (b; c i ) is given as foows: u i (b; c i ) (b i c i ) Pr{i wins b}, where, with F ( ε ᾱ, ū, q) denoting the joint CDF of ɛ given (α, U, Q) = (ᾱ, ū, q), Pr{i wins b} = h(b; ᾱ, ū, q)df (ᾱ, ū, q), and h(b; ᾱ, ū, q) H i (b;ᾱ,ū, q) df ( ε ᾱ, ū, q) and H i (b; ᾱ, ū, q) = { ε R A : ᾱ q j,i + ε j,i b j b i and ū ᾱq i ε i b i, j N p α( q k q i ) + ε k,i b k b i, k N t }. Since F ( ᾱ, ū, q) is uniformy continuous for a (ᾱ, ū, q) in the support of (α, U, Q), we find that h( ; ᾱ, ū, q) is continuous. By the Bounded Convergence Theorem, Pr{i wins b} is continuous in b. As for singe crossing of incrementa returns, we consider the foowing. For each bidder i, for b i b i and c i c i u i (b i, b i, c i ) u i (b i, b i, c i ) = (b i c i ) Pr{i wins b i, b i } (b i c i ) Pr{i wins b i, b i } = (b i b i) Pr{i wins b i, b i } + (b i c i ) (Pr{i wins b i, b i } Pr{i wins b i, b i }). The ast term is bounded from beow by because (b i c i) (Pr{i wins b i, b i } Pr{i wins b i, b i }) Pr{i wins b i, b i } Pr{i wins b i, b i }. Therefore, the payoff u i (b i, b i, c i ) satisfies non-decreasing differences in (b i, c i ). This impies that its expected payoff U i (, ; b i ) satisfies singe crossing of incrementa returns. Hence by Coroary. of Athey (), each subgame with given A has a PSBNE. We now consider a PSBNE in the entry stage. For this, we use Theorem of Athey (). The action space is {, } with action denoted by d. We set d = to denote participation and d = to denote non-participation. The private type for each bidder i is entry cost E i. The interim expected payoff function for each bidder i with entry cost e i is given by ( d i ) (E [ U i (b i (C i ), C i ; (b j ) j A(d(e))\{i} ) E i = e i ] ei ) = ( d i ) (E [ U i (b i (C i ), C i ; (b j ) j A(d(e))\{i} ) ] e i ) where A(d) = {i N : d i = }, the bid strategy profie b(c) are given by the subgame PSBNE whose existence is previousy estabished, and d(e) = (d i (e i )) i N. (Reca that the private costs C i are reveaed ony after the bidder decides to enter and are independent of E i.) Certainy Assumption A and singe crossing of incrementa returns in Theorem of Athey () are satisfied. Thus, the existence proof of PSBNE foows from the theorem. Q.E.D.

3 3 B. Further Detais about Identification Resuts B. Quaity differences and buyers tastes We now provide more detais in the argument about the identification of the difference in quaity eves, the distribution of buyers weights for quaity α, and the match components ɛ. To do so, it is convenient to first iustrate the main idea using a simpe case where the set of participants consist of two permanent bidders i, j and a singe transitory bidder k. Under the maintained independence between (U, α, ɛ) and C i s (and hence B i s) and the assumption that participation decisions are non-strategic (in the sense that they do not depend on the reaization of (U, α, ɛ) and private costs), the probabiity that i wins conditiona on the reaized vector of bids (b i, b j, b k ) is Pr (α q j,i + ɛ j,i b i b j and Y i,k ɛ i b i ), () where random variabe Y i,k is defined as the maximum of U αq i and α(q k q i ) b k + ɛ k. First, suppose q i is identica to q j so that q i,j = (which is an event that can be conditioned on given that the quaity cassifications are identified before). Then a deconvoution argument based on the Kotarski Theorem impies the margina distributions of ɛ i, ɛ j and Y i,k are identified up to scae, provided the joint support of (B i B j, B i ) is arge reative to that of ( ɛ j,i, Y i,k ɛ i ) conditiona on the price from the transitory bidder b k. Next, consider a simiar winning probabiity except that q i q j. The joint distribution of (α q j,i + ɛ j,i, Y i,k ɛ i ) is recovered from () as ong as the support of (B i B j, B i ) is sufficienty arge given b k. Thus the independence between (α, Y i,k ) and (ɛ i, ɛ j ) impies the joint distribution of (α q j,i, Y i,k ) is recovered using knowedge of the distribution of ɛ i obtained in the previous step. A scae normaization such as E[α] = then impies the difference in quaity eves and the distribution of buyers taste for quaity are identified from the margina distribution of α q j,i. In addition, the distribution of Y i,k given α is aso recovered. This means, with a ocation normaization that q j =, the distribution of max{u, αq k B k + ɛ k } given α and b k is identified as that of Y i,k + αq i given α and b k. The support conditions invoved in the identification of the distribution of buyers s tastes is aso discussed in further detais in Section B4 beow. We now formaize the argument for identifying buyers tastes presented. For any pair of bidders i, j, et B i,j denote the vector of bids submitted by the set of the other entrants A\{i, j}, where A A p A t denote the set of entrants. Let Y i,j (B i,j ; a) denote the maximum of U αq i and α(q k q i ) b k + ɛ k for k A\{i, j}. The foowing assumption is needed. (A3) There exists a set a = a p a t, a pair i, j a p and a vector of bids b i,j {b k } k a\{i,j} such that (i) the characteristic function of the joint distribution of (α q j,i + ɛ j ɛ i, Y i,j (b i,j ; a) ɛ i ) does not vanish; and (ii) the joint support of (B j B i, B i ) incudes the joint support of (α q j,i + ɛ j ɛ i, Y i,j (b i,j ; a) ɛ i ) conditiona on b i,j. Proposition B. Suppose (A), (A) hod. (i) If (A3) hods for some i, j with q i = q j and some (a, b i,j ), then the margina distribution of ɛ i and the distribution of Y i,j given (a, b i,j ) are jointy identified up to a ocation normaization (e.g. E(ɛ i ) = ). (ii) If in addition (A3) aso hods for some i, j with q i q j and some (a, b i,j), then the quaity difference q i,j and the distribution of α are jointy identified up to a scae normaization (e.g. E(α) = ). Besides, the joint distribution of (α q j,i, Y i,j ) given (a, b i,j) is aso identified. Proof of Proposition B. Part (i). Consider i, j and a t, a p, b i,j satisfying Assumption (A3) and q i = q j. Under Assumptions (A) and (A), ɛ i, ɛ j, Y i,j are mutuay independent given any

4 4 (b i,j, a). Hence, for any b i, b j R, the probabiity that i wins conditiona on B i = b i, B j = b j and a, b i,j equas ϕ(b i, b j ; q i, q j ) Pr {α q j,i + ɛ j,i b j b i and Y i,j (b i,j ; a) ɛ i b i } where q j,i q j q i and ikewise for ɛ j,i. Note the equaity uses independence of private costs (and hence bids) from U, α, and ɛ. With B j and B i being independent of (U, α, ɛ), evauating ϕ(.,.; q i, q j ) at different reaizations of B i, B j ony amounts to evauating the same joint distribution of (ɛ j ɛ i, Y i,j (b i,j ; a) ɛ i ) (with b i,j, a fixed) at different points on the support. Thus, under Assumption A3-(ii), i.e. the arge support condition, this joint distribution is identified. Mutua independence between ɛ i, ɛ j and Y i,j (b i,j ; a) given (b i,j, a) impies their margina distributions are identified up to some ocation normaizations (e.g. E(ɛ) = ) by an appication of the Kotarski s Theorem, or Theorem.. in Rao (99). Part (ii). Without oss of generaity, suppose conditions for part (ii) hod for q i > q j. Repicating arguments in part (i) with i, j, a t, b i,j satisfying Assumption A3 and q i > q j shows that the joint distribution of (α q j,i + ɛ j,i, Y i,j (b i,j ; a ) ɛ i ) given (b i,j, a ) is identified. This impies the margina distribution of α q j,i + ɛ j,i and the distribution of Y i,j (b i,j ; a ) ɛ i given (b t, a ) are aso recovered. With the margina distribution of match components aready identified from part (i), so is the distribution of ɛ j,i. With α q j,i being independent from ɛ j,i, this means the distribution of α q j,i can be identified as ong as the characteristic-function of ɛ j,i is non-vanishing. It then foows that q j,i and the distribution of α are identified up to some scae normaization (such as E[α] = ). Finay, note the joint distribution (α q j,i, Y i,j (b i,j ; a )) given (b i,j ; a ) can be recovered from knowedge of the distributions of ( ɛ j,i, ɛ i) and the distribution of (α q j,i + ɛ j,i, Y i,j (b i,j ; a ) ɛ i ) given (b i,j ; a ), as ong as the characteristic function for ( ɛ j,i, ɛ i) is non-vanishing. This is because (A) and (A) impy ( ɛ j,i, ɛ i) is independent from (α q j,i, Y i,j (b i,j ; a )) with (b i,j ; a ) fixed. The proof of identification of the distribution of outside option and the distribution of quaities of transitory bidders conditiona on bids is as presented in the text. Q.E.D. B. Buyers taste for observed characteristics We now expain how to recover the distribution of β (buyers tastes for observed characteristics) in a context when a participants in the auction are permanent seers. Let their quaity eves be identified. Then Pr (i wins A = a, (b i ) i a ) ɛ j,i + α q j,i + x j,i β b j b i j a\{i} = Pr and U αq i ɛ i b i where x j,i x j x i. We need to introduce the foowing condition for identification of the distribution of buyers taste for observed heterogeneity. (B) There exist some i and a subset of permanent seers a with i a such that (i) the ( a )- by-j matrix ( x j,i ) j a\{i}, with J being the number of coordinates in x i, has fu rank; (ii) the support of (( ɛ ji + α q ji + x ji β) j a\{i}, U αq i ɛ i ) is a subset of the support of ((B j B i ) j a\{i}, B i ); and (iii) (( ɛ j,i + α q j,i ) j a\{i}, U αq i ) have non-vanishing characteristic functions.

5 5 Both (i) and (iii) in (B) are mid and standard assumptions in nonparametric identification. The support conditions in (ii) can be structuray justified aong simiar ines discussed in Section B4 beow. Proposition B Suppose the distribution of (α, ɛ, U ) and the quaity eves are identified. If the condition (B) hods, then the distribution of β is identified. We now sketch the proof of this proposition. Provided the joint support of (B j B i ) j a\i and B i is arge enough to cover the joint support of (( ɛ ji + α q ji + x ji β) j a\i, U αq i ɛ i ), we can recover the joint distribution of the atter. With F ɛi and F α identified as above, and under our assumption that (ɛ i ) i a are i.i.d. and jointy independent from α, the distribution of ɛ j,i + α q j,i is known. With β assumed to be independent from α and (ɛ i ) i N, we can identify the distribution of ( x j,i β) j a\i by taking the ratio of the characteristic functions of ( ɛ j,i + α q j,i + x j,i β) j a\i and ( ɛ j,i + α q j,i ) j a\i. As ong as the matrix ( x j,i ) j a\i is furank, then the joint density of β is identified as the product of the joint density of ( x j,i β) j a\i and the absoute vaue of the determinant of the square matrix ( x j,i ) j a\i by changing variabes. B3. Distribution of private costs We discuss how the distribution of project s costs can be identified. We consider a simpe case when bidders entry costs are independent of the project s costs. The genera resut obtains by combining steps presented beow with the identification strategy proposed by?. The identification of the distribution of project s costs. in a simpe case of signas independence foows an argument simiar to Guerre, Perrigne, and Vuong (). To see this, note that the quaity eves for permanent seers, the distribution of quaity eves of transitory seers, and buyer tastes can be considered known since they are identified in preceding sub-sections. The inverse bidding strategy can be recovered as foows. The first-order condition for bidder i choosing price b in equiibrium is: (b c i ) b Pr {ϖ(a) b} b=b = Pr {ϖ(a) b } () where ϖ(a) denotes the maximum of max j A\{i} (αq j αq i + x j,i β + ɛ j,i B j ) and U αq i. Thus, it suffices to show that the distribution on the right-hand side can be identified. It woud impy that the derivative on the eft-hand side woud aso be identified, in which case the inverse bidding strategy (and consequenty the distribution of private cost C i ) woud aso be recovered for every subpopuation of seers with (x, q). Note the right-hand side of () is: {a:a N} P {ϖ(a) b i A = a} Pr {A = a}. For those entrants in A\{i} who are transitory seers, their quaities are mutinomia random variabes whose distributions are recoverabe, with the distribution of quaities assumed to be the same across the two subpopuations of permanent seers and transitory seers (i.e. π p,k = π t,k ). Finay note that, given any fixed a, the distribution of ϖ(a) can be constructed from the joint distribution of α, ɛ, β, U identified earier and the distribution of submitted bids given a independence assumptions in (A) and (A).

6 6 B4. Discussion of Support Conditions Parts of our identification requires the supports of prices quoted by permanent seers be arge in the sense of condition (ii) in (A3). In this section we provide a heuristic argument showing that our mode is capabe of generating the variation in prices necessary for this condition to hod. We do so in the context of a simpe mode which abstracts away from the differences in unobserved quaities, stochastic participation, and the presence of transitory bidders but aows for the aocation rue to incude a stochastic match component. Adding these features compicates the agebra but does not require any additiona insight. In this simpe mode, the support condition for identifying the distribution of match components is reduced to: There exist i, j with q i = q j such that the support of (B j B i, B i ) incudes that of (ɛ j ɛ i, U ɛ i ). Note that, in a type-symmetric equiibrium which we consider here, the support of bids from i, j are identica, and denoted by [b i, b i ]. The respective supports of (B j B i, B i ) and (ɛ j ɛ i, U ɛ i ) are depicted in Figure??. It is cear from these figures that a set of sufficient conditions for the support restrictions above is: there exist i, j with q i = q j s.t. b > ε u and b < ε u, which can be satisfied if (a) b > ε u and (b) b b > (u u ) + (ε ε). Condition (a) hods provided b c i > ε u, where c i is the supreme of the support of private costs for i and j. Condition (b) essentiay requires the support of bids to be arge reative to that of outside utiity and match components. Intuitivey, (b) aso hods when the support of seers private costs is sufficienty arge. We now provide a simpe argument for this intuition. The idea is to show that the bidding strategy is continuous in the ength of the support of match component ε ε and the support of outside option. Under type-symmetric PSBNE the bidders strategies sove the maximization probem: ( ) σ i (c) arg max(b c) Pr (U ɛ i b) Pr b ε ε max { B j + ɛ j } ɛ i b j A\{i} The second probabiity in (3) represents i s beief, which is formed from i s knowedge of S, the distribution of private costs {F k } k K, and the distribution of quaities in the popuation of seers. Suppose ɛ i are i.i.d. uniform over [ε, ε]. Appying the Law of Tota Probabiity, we can write the objective function for seer i with costs c as ε [( ε ) ] F Bj (b ε i,j ) FU ( b + ε i ) (b c) dε j dε i. (4) ε ε ε ε Changing variabes between ε r and τ r εr ε ε ε ( (b c) for r = i, j, we can write (4) as F U (ε b + δτ i ) [ F Bj (b δ τ i,j ) ] dτ j ) dτ i (5) where δ ε ε is the ength of support of match components. Note (5) is continuous in both δ and the ength of support for U. It then foows from an appication of the Theorem of Maximum that the support of bids is continuous in the size of the support of match component and outside options. In this Figure, ε and ε denote the infinum and supremum of the support of ɛ i (and ɛ j ) and u, u denote the infinum and supremum for the support of U. (3)

7 7 Provided private costs vary sufficienty, the support of bids in a standard auction mode with no match components (i.e. ɛ is degenerate at ) and no outside option is an interva with non-degenerate interior. It then foows from the impication of the Theorem of Maximum that condition (b) hods whenever the support of ɛ and U is sma enough. By the same token, we can provide simiar structura justifications for the support conditions for recovering quaity eves using i, j with q i q j, as ong as variation in private costs and quaity differences are sufficienty arge reative to that of buyers tastes in (α, ɛ, U ). In the current exampe, the presence of transitory seers woud not entai any quaitative different arguments for the structura justification of the support conditions. C. Mode with Endogenous Entry This section provides further detais of the mode where seers decisions to participate in auctions are endogenous. We discuss how to extend the arguments for the identification from the text in this case. Existence of PSBNE in a mode with endogenous entry is shown in the preceding section. Throughout this section, we assume an anaog of (A) and (A) in the text hod for a set of potentia bidders. (A ) The private signa E i, and the private cost C i, are independent from each other, and the random vectors (E i,, C i, ) are independent across a i N and across auctions. For each i with q i = q k, these costs are independent draws from the continuous distributions Fk E and with a density positive over supports [e k, e k ] and [c k, c k ] respectivey. F C k (A ) The three random vectors (α, U, ), ɛ and (E i,, C i, ) i N are mutuay independent; match components ɛ i, are i.i.d. across i s; ɛ i, and (α, V, ) are continuousy distributed with a density positive over [ε, ε] and over [, α] [u, u ] respectivey. We focus on type-symmetric equiibria in which any pair of participants i, j who are ex ante identica (i.e. either i, j N p and q i = q j or i, j N t ) adopt the same strategies. The strategy of seer i from the quaity group q k, σ r,k, consists of a participation strategy σ r,k E (as defined above) and a bidding strategy, σ r,k B : [c k, c k ] Supp(I N, ) R +. Conditiona on participation, a seer i s expected profit from bidding b is given by Π r,k (b, c i, I N, ; σ i E, σ i B ) (b c i) Pr(i wins b, (r, k), I N, ; σ i E, σ i B ), where (σ i E, σ i B ) is the profie of strategies of the other participants. A PSBNE is a profie of strategies {(σ r,k E, σr,k B )} r {p,t}, k {,...,K} 3 such that and if and ony if σ r,k E (e i,, I N, ) =. σ r,k B (c i,, I N, ) = arg max b Π r,k (b, c i,, I N, ; σ i E, σ i B ) E[Π r,k (σ r,k B (C i,, I N, ), C i, I N, ) E i, = e i, ] e i, Our identification resuts coud be extended to the case when E i, and C i, are correated for each seer i in an auction, but the random vectors (E i,, C i, ) are independent across i N and across auctions. 3 We assume that σ r,k B = if σr,k E =.

8 8 As discussed above, PSBNE with monotone strategies exists. Given orthogonaity between E i, and C i, and independence of (E i,, C i, ) across bidders, the equiibrium participation strategies are monotone and are characterized by a threshod rue. That is, for any I N,, there exists e p,k and e t,k such that for a permanent potentia bidder i with quaity eve qk, σ r,k E (e i,, I N, ) = iff e i, e p,k ) whereas for each transitory seers with quaity eve q k, σ r,k E (e i,, I N, ) = iff e i, e t,k. It is easy to estabish that under Assumptions (A ), (A ) and (A3) and for any given I N, and in any pure-strategy Bayesian Nash equiibrium, the buyer s tastes {α, ɛ, U } are independent from {D i, } i N (and therefore A ); participation decisions {D i, } i N are mutuay independent across a i N ; and bids are mutuay independent within each reaized set of active bidders. The pairwise comparison approach for identification in Proposition?? remains unchanged in the presence of endogenous participation. Since participation decisions are independent across potentia seers, for given i and j the probabiities for various sets of competitors to reaize conditiona on i A and j A are sti identica to those conditiona on j A and i A. Thus, the proof of Proposition which reies on this property remains vaid. 4 Identification of quaity differences and the distribution of buyers tastes aso foow from the same arguments as before. In particuar, since entry decisions and bid distributions are orthogona to (α, ɛ, U ) the variation in bids by permanent seers is sti a vaid source of exogenous variation for identification. Further, these resuts aso rey on the independence of entry and bidding strategies across active bidders (and in particuar on independence of permanent bidders strategies from those of transitory bidders) which continue to hod under endogenous entry as mentioned above. Finay, note that with C i, being independent from E i, the standard arguments used for backing out the distribution of private costs remain vaid in the presence of endogenous entry. D. Detais of Cassification Procedure This section provides the detais of the nonparametric cassification methodoogy. A compete treatment of this methodoogy and forma resuts are found in Krasnokutskaya, Song, and Tang (4). D. Estimation of Cassifications with Known Number of Groups We consider the case where X i takes vaues from a finite set (x,, x Λ ). This impies that the set of the permanent seers can be partitioned into S p groups, S,, S Λ, such that for each λ =,, Λ, and any i, j S λ, x i = x j. The anaysis in this section is performed conditiona on x. For brevity, we omit conditioning on x in the exposition beow, so that we simpy write S instead of S λ. Define K to be the number of distinct quaity eves among the permanent seers. For ease of exposition, we first present the case with the number of quaity eves K equa to so that q i { q h, q } for a pair of unknown numbers q h and q. We expain how the agorithm generaizes to the case with K > ater. Let S h S be the coection of high quaity seers within group λ and S S be the coection of ow quaity seers within group λ. We estimate 4 Notice that in the mode with endogenous entry the pairwise index aso depends on the set or rather on reevant information about the set of potentia bidders, I N. In practice, the index is computed by pooing a the observations that are characterized by the same I N which may entai using auctions with distinct (in terms of identities) sets of potentia bidders.

9 9 an ordered partition (S h, S ) of S in three steps. First, for each i S, we estimate two ordered partitions: one partition consists of the group of the seers with higher or equa quaity than that of i (denoted by S (i)) and the rest (denoted by S\S (i)), and the other partition consists of the group of seers with ower or equa quaity than that of i (denoted by S (i)) and the rest. Second, among the two ordered partitions, we choose the one that is mosty ikey to coincide with (S h, S ). Third, we choose i such that the estimated partition associated with this i is most strongy supported by the data. Our method reies on the estimates of winning probabiities in Proposition. Let W p i, {, } be an indicator taking if the i-th seer that is permanent wins at auction and otherwise. Define ˆδ ij (b) ˆr ij (b) ˆr ji (b), where ˆr ij (b) L = W p i, K h(b i, b){j / A } L = K h(b i, b){j / A }, where K h (v) = K(v/h)/h for a univariate kerne function K. Then we construct test statistics: ˆτ + ij = max{ˆδ ij (b), }db, ˆτ ij = max{ ˆδ ij (b), }db, and ˆτ ij = ˆδ ij (b) db. We confine the integra domains to the intersection of bids submitted by i and j, and this restriction is omitted from the notation. We use a bootstrap method to estimate the finite sampe distributions of the test statistics. We first construct ˆr ij,s(b) and ˆδ ij,s(b) using the s-th bootstrap sampe, s =,, B, and consider the re-centered bootstrap test statistics: ˆτ + ij,s = max{ˆδ ij,s(b) ˆδ ij (b), }db ˆτ ij,s = max{ ˆδ ij,s(b) + ˆδ ij (b), }db, and ˆτ ij,s = ˆδ ij,s(b) ˆδ ij (b) db. Note that Lee, Song, and Whang (3) estabished asymptotic properties of these bootstrap test statistics in a more genera set-up. Using the bootstrap test statistics, we define the bootstrap p-vaues as foows: p z(i, j) = B B s= z {ˆτ ij,s > ˆτ ij} z with z {+,, }. We proceed in three steps as outined above. Step : Define Ŝ (i) = {j S\{i} : p +(i, j) p (i, j)} and Ŝ (i) = {j S\{i} : p +(i, j) > p (i, j)}. Step : We determine now whether seer i has the same quaity as those of S (i) or S (i),

10 i.e., ow quaity or high quaity. If both Ŝ(i) and Ŝ(i) are non-empty, 5 we cassify seer i as foows: Take Ŝ(i) = Ŝ(i) and Ŝh(i) = Ŝ(i) {i} if min og p (i, j) < min og p (i, j) j Ŝ(i) j Ŝ(i) Take Ŝ(i) = Ŝ(i) {i} and Ŝh(i) = Ŝ(i) if min og p (i, j) min og p (i, j). j Ŝ(i) j Ŝ(i) That is, we put seer i into the high-quaity group, if the evidence against the hypothesis that seer i has the same quaity as the seers in group Ŝ(i) is stronger than the evidence against the hypothesis that seer i has the same quaity as the seers in group Ŝ(i). Step 3: For each i S, we compute the foowing index: s (i) = { Ŝ(i) Ŝh(i) j Ŝ(i) og p +(i, j) if i Ŝh(i) j Ŝh(i) og p (i, j) if i Ŝ(i), where Ŝ(i) denotes the cardinaity of the set Ŝ(i). The quantity s (i) indicates the weakness of the ikeihood that i is cassified into her right quaity group. Then choose i that minimizes s (i) over i S, and et Ŝh = Ŝh(i ) and Ŝ = Ŝ(i ). We take Ĉ = (Ŝh(i ), Ŝ(i )) as an estimated cassification of payers into two quaity groups. 6 The generaization of the procedure to the case of K > with K known can proceed as foows. First, we spit S into Ŝh and Ŝ using the agorithm for K =. Then we find a minimum vaue (denoted by ˆp h ) of og p (i, j) among the pairs (i, j) such that i j, and i, j Ŝh, and a minimum vaue (denoted by ˆp ) of og p (i, j) among the pairs (i, j) such that i j, and i, j Ŝ. If ˆp h < ˆp, we spit Ŝh into Ŝhh and Ŝh using the same agorithm for K =, and otherwise, we spit Ŝ into Ŝh and Ŝ using the same agorithm for K =. We repeat the procedure. For exampe, suppose that we have cassifications Ŝ,, Ŝk obtained. For each r =,, k, we compute the minimum vaue (say, ˆp r ) of og p (i, j) among the pairs (i, j) such that i j and i, j Ŝr, and then seect its minimum (say, ˆp r ) over r =,, k. We spit Ŝr into Ŝr h and Ŝ r using the agorithm for K = to obtain a cassification of S into k groups. We continue unti the groups become as many as K. 5 If Ŝ(i) is empty, we pick some eve α =.5: Take Ŝ(i) = and Ŝh(i) = Ŝ(i) {i} if min j Ŝ(i) p (i, j) α. Take Ŝ(i) = {i} and Ŝh(i) = Ŝ(i) if min j Ŝ(i) p (i, j) < α. We proceed in a simiar way if Ŝ(i) is empty. 6 There may be aternative ways to obtain estimators of the quaity partition. One way is to fix i and appy hypothesis testing to the nu hypothesis that q i q j. This essentiay bois down to comparing the p- vaue p z(i, j) with a certain eve of the test. One unattractive feature of this aternative procedure is that the resut can be different depending on whether the nu hypothesis is taken to be q i q j or q i q j. This is because the conventiona hypothesis testing procedure treats the nu hypothesis and the aternative hypothesis asymmetricay. Hence in this paper, instead of comparing a p-vaue with a fixed eve of the test, we compare a p-vaue with an aternative p-vaue, to capture the symmetry of the comparison.

11 D. Seection of the Number of Groups The methodoogy outined above assumes that we know the exact number of groups. To accommodate the situation with rea ife data without knowedge of the number of the groups, we offer a method of consistent seection of the number of groups. We suggest that the number of groups shoud be seected to minimize the criterion function that baances a measure of goodness-offit that captures a misspecification bias versus a penaty term that penaizes overfitting. The goodness-of-fit measure is based on the variance test approach. Given an estimated cassification S = K k=ŝk with K groups, et ˆV k (K) = min og p (i, j), i,j Ŝk for each k =,, K. Suppose that K is the true number of groups. Let g(l) be such that g(l)/ L as L. Then, define ˆQ(K) K ˆV k (K) + Kg(L). K We seect K as foows: k= ˆK = argmin K N ˆQ(K). The idea of this seection is based on the foowing intuition. First, K K k= ˆV k (K) = O P (), as L, if K K, because there is no misspecification bias in this case. The imit O P () measures the asymptotic behavior of the goodness-of-fit of the mode when the cassification is weaky finer than the true cassification. Since Kg(L) as L, the minimization of ˆQ(K) over K eans toward a ower choice of K that is coser to K. On the other hand, if the cassification is stricty coarser than the true cassification, i.e., K < K, the quantity K K k= ˆV k (K) diverges at a rate faster than g(l), as L, due to misspecification. In this case, the minimization of ˆQ(K) over K excudes K such that K < K for arge sampes. Thus we obtain the consistency of ˆK under reguarity conditions. Detais in a more genera set-up are found in Krasnokutskaya, Song, and Tang (4). E. Detais of GMM Estimation E. Moment conditions Using the resut in Proposition, we derive the moment conditions as foows. For any vector vaued map h p (x,q), : R A R d h, and for each x X, q Qx, we can write the moment condition as: [ ] E h p (x,q), (B, I )W p (x,q), I [ ] = K A, E k= h p (x,q), (B, I )e p k,(x,q), (B, I ; θ) g k (B t, I,, I t, ; θ) I KA, d= g d(b t, I,, I t, ; θ).

12 where g k (B t, I,, I t, ; θ) is a parametric approximation of g k(b t, I,, I t, ), W p (x,q), = e p (x,q),k, (B ; θ) = A p (x,q), j A p (x,q), A p (x,q), j A p (x,q), W p j, and e (x,q),k, (B ; θ, j). This can be further re-written as: [ ) ] E h p (x,q), (B, I ) (W p(x,q), ep(x,q), (θ, g) I = with (6) K e p (x,q),(θ, g) = e p (k,x),q, (B ; θ) k= g k (B t, I,, I t, ; θ). KA, d= g d(b t, I,, I t, ; θ) As we discussed in the previous subsection, it may be usefu, especiay in the parametric setting, to separatey specify and estimate f(bi, t Qt A,i, = q A,i,k, I, ) and P (i A t x, Qt A,i, = q A,i,k, I, ) functions. In such a case, restrictions associated with (a) transitory seers bid distribution, (b) the transitory seers probabiity of participation, as we as (c) restrictions summarizing optima participation behavior may be additionay imposed. More detais are given ater. To construct a sampe version of the moment conditions, we first obtain a consistent estimator ˆω k, of ω k, via the sampe anaog principe, and construct ) ˆρ (x,q), (θ) = h p (x,q), (Bp, I ) (W p(x,q), êp(x,q), (θ), where ê p (x,q), (θ) is equa to ep (x,q), (θ) except that ω k, is repaced by ˆω k, and ˆω k, is the sampe anaogue of ω k,. We define ˆρ (θ) to be a coumn vector with ˆρ (x,q), (θ) stacked up with (x, q) running in X Q. Hence the dimension of ˆρ (θ) is d h,p X Q, where X Q is the cardinaity of the set X Q. Then define a genera method of moment estimator as foows: ˆθ GMM = argmin θ Θ ˆQ GMM (θ), where ˆQ GMM (θ) = ˆΣ = L ( L ) L ˆρ (θ) = L ˆρ ( θ)ˆρ ( θ), = ˆΣ ( L ) L ˆρ (θ), and = and θ is the first step estimator of θ which is a minimizer of ˆQ GMM (θ) ony with ˆΣ repaced by the identity matrix. Under reguarity conditions, the estimator is known to be asymptoticay norma with a positive definite covariance matrix. Note that the estimation error due to ˆω k, does not affect the asymptotic variance matrix because the components P {Q t A,i, = q A,i,k I } of ω k, take vaues ony from a finite set and hence has a convergence rate that is arbitrariy fast as L.

13 3 Next, we expain some additiona restrictions used in estimation. (a) The restriction associated with transitory seers bid distribution: f(b t i I ) = k f(b t i Q t A,i, = q A,i,k, I, ) Pr(Q t A, = q A,k, I ) = (7) K A k= ωa,k, t {f(b j Āt j Q t A,j, = q A,j,k, I, ) Pr(j A t Qt A,j, = q A,j,k, I, )} Pr(j A t j Āt Qt A,j, = q. A,j,d, I, ) K d= ωt A,d, Moment conditions associated with this restriction woud reate the empirica moments of the f(b t i I ) distribution to the theoretica moments computed using (7). (b) The restriction associated with the transitory seers probabiity of participation: Pr( A t x, = m x, I, ) = K A, k= j Āt K A, k= Pr( A t x, = m x, and Q t A, = q A,k I, ) = (8) { Pr(j A t Q t A,j, = q A,j,k, I, ) Pr(Q t N,j, = q j x t j,) } q N A Q N A Ω i N t Āt { Pr(i N t A t Q t N,i, = q i, I, ) Pr(Q t N,i, = q i x t i,) }. The derivation for this expression is provided in the proof of Proposition. Moment conditions associated with this restriction woud reate the transitory seers empirica probabiity of participation and expected x characteristics of entrants conditiona on I, to their theoretica counterparts using (8). (c) The restriction reated to the expected profit condition. This restriction summarizes optima participation behavior. It is summarized by the threshod strategy where potentia bidders with entry cost draws beow the ex-ante expected profit participate in the auctions and those with higher draws stay out. This impies that in equiibrium Pr(j A t x, Q t j, = q k,x, I, ) = F E (E[π t (j, Q t A,j, = q k,x, I, )]), where F E (.) is the distribution function of entry costs E and q k,x Q x. E. Semiparametric Estimation The estimation method in the previous section empoys parametrization of nonparametric functions g k. The functions g k invove the density of the transitory seers bids and participation probabiities. Since the bids and participations are equiibrium objects, one might prefer to use a more fexibe specification for the nonparametric functions g k. In this section, we expain how this extension can be done in practice. Reca our definition e p x,q, (θ, g) in (6) and define ˆρ x,q,s (θ, g) = h p x,q,(b p, I ) ( W p x,q, êp x,q, (θ, g)),

14 4 making its dependence on the nonparametric function g expicit. Let ˆm x,q, (θ, g) = L s= ˆρ x,q,s(θ, g) {I s = I } L s= {I s = I } and define ˆm (θ, g) to be a coumn vector with ˆm x,q, (θ, g) stacked up with (x, q) running in X Q. Hence the dimension of ˆm (θ, g) is d h,p X Q, where X Q is the cardinaity of the set X Q. In the semiparametric estimation, we regard the nonparametric function g as an infinite dimensiona nuisance parameter, and empoy the sieve minimum distance estimation method of Ai and Chen (3). As argued previousy, the functions g are nonparametricay identified under our set-up. Now et G L be the space of finite dimensiona sieves whose dimension increases as the number of auctions L grows. Detais about the appropriate sieves are found in the appendix. We construct the estimator as foows: where (ˆθ CMD, ĝ) = argmin (θ,g) Θ GL ˆQ CMD (θ, g), ˆQ CMD (θ, g) = L L = ˆm (θ, g)ˆσ ˆm (θ, g), and ˆΣ is a consistent estimator of a non-singuar matrix Σ. Ai and Chen (3) showed that under reguarity conditions, we have n (ˆθCMD θ ) d N(, V ), where V is a non-singuar covariance matrix. We turn to the form of the covariance matrix formua which foows Ai and Chen (3). For each i =,, N, we et for α = (θ, g) A L = Θ G L, and r () (α) = [r () x,q, (α) ] q Q x,x X r () x,q, (α) = E [ h p x,q,(b, I ){W p x,q, ep x,q, (α)} I ] r () x,q, (α) = E [ h t x,q,(b, I ){Wx, t et x,q, (α)} I ] and r() (α) = [r () x,q, (α) ] q Q x,x X. In other words, r() (α) is a coumn vector with vectors r () x,q, (α), q Q x, x X, stacked up and r () (α) = [r () x,q, (α) ] q Q x,x X is a coumn vector with vectors r () x,q, (α), q Q x, x X, stacked up. We define m (α) = For each j =,, d θ, we define for w G, m (θ, g ) g [ r () r () ] (α). (α) [w g ] = m (θ, τw + ( τ)g ) τ=. τ The eft-hand side term represents the pathwise derivative of m (θ, g ) aong the direction w g.

15 5 Let w j be the soution to the minimization probem: where ( m (θ, g ) inf m ) ( (θ, g ) m (θ, g ) [w j g ] Σ m ) (θ, g ) [w j g ], w j G θ j g θ j g Then we take [ ] Σ = E [r () (α), r () (α)][r () (α), r () (α) ] I. D w, = m (θ, g ) m (θ, g ) [ ] w θ g g,, wd θ g. The asymptotic covariance matrix formua becomes V = E [ D w,σ D w,]. The estimation of the asymptotic covariance matrix formua can be done in a straightforward way. For exampe, we may foow Section 5 of Ai and Chen (3), except that instead of using the series estimation to obtain the sampe anaogue of the conditiona expectation E[ I ], we use the usua sampe anaogue of the conditiona expectation with discrete conditiona variabes because I is a discrete random vector. Detais are omitted. Now et us consider the choice of the sieve space G L. Let K > be an integer such that for a L, max =,,L K K. Now et us discuss the construction of the sieve space G L. For each k =,, K, and reaized vaue of (I,, I t, ), the function g k(, I,, I t, ) defined in (??) is of particuar form. First we take G L = G,L G K,L, where G k,l s are constructed as foows. For i =,, N, L, and k =,, K, et G k,l and F k,l be sieve spaces, where for each g k,l G k,l, g k,l : I [, ] and for each f k,l F k,l, f k,l : R I [, ) and for each I, I, fk,l (b, I, )db =. Then we construct a sieve space G k,l as the coection of maps g k,l (b, I, ), where g k,l (b, I, ) = x X Āt x, j= f k,j,l (b, I, )v x k,j,l(i, ), where vk,j,l x Vx k,j,l, and f k,j,l F k,j,l. For Vk,j,L x, we choose a sieve space Mx k,j,l of rea-vaued functions and define { } exp(m( )) Vk,j,L x = + exp(m( )) : m Mx k,j,l. As for M x k,j,l, we can take poynomia series. As for F k,l = K j= F k,j,l, we use a Hermite poynomia sieve, where F k,j,l = { fk,kl,k (x; σ k, ε,k, r,k, a k ) : ε >, σ >, r, a s,k R, s =,, K L,k fk,kl,k (x; σ k, ε,k, r,k, a k )dx = and f k,kl,k (x; σ k, ε,k, r,k, a k ) is defined to be ε,k + πσk K L,k s= a s,k ( x r,k σ k ) s exp ( (x r,k) σ k ). },

16 6 Observe that f k,kl,k (x; σ k, ε,k, r,k, a k )dx = ε,k + K L,k K L,k s= t= a s,k a t,k E ( Z s+t), where Z is a standard norma random variabe. The quantity E (Z s+t ) can be expicity computed from the moment generating function. F. Additiona Empirica Resuts Tabe : Estimated Quaity Structure for a Given Number of Groups Number of Groups K = K = K = 3 K = 4 K = 5 K = V Q(K) This tabe shows the estimated quaity group structures for the various numbers of quaity groups for Eastern European suppiers with the medium eves of average reputation score. Rows -6 record the number of suppiers estimated to beong to a respective group. Rows 7 and 8 record the vaue of the p vaue component of the criterion function and the vaue of the criterion function. The resuts are based on the penaty function g(l) = og(og(l)). Resuts indicate that the number of groups most supported by the data is equa to three.

17 7 Tabe : Participation Decision and Bid Distribution Score Q I(T) II(T) I(P) II(P) Mean Constant.55 (.9) -.5 (.9).585 (.5) -.73 (.9) No Ratings -.83 (.5) -.33 (.8) < Ratings (.6).5 (.) 3 < Ratings (.7).5 (.9) Number of Ratings.7 (.63).3 (.7) Average Score -.5 (.5) -.4 (.) Average Score.5 (.). (.3) North America, Low, -.8 (.3).3 (.) -.37 (.46).3 (.7) North America, Low, -.73 (.8) -.4 (.) -.7 (.44) -.53 (.5) North America, Medium, -.38 (.) -.39 (.3) -.6 (.46) -.5 (.6) North America, High, -.73 (.).34 (.3) -.4 (.4).65 (.7) North America, High, -.8 (.) -.65 (.3) -.66 (.39) -.93 (.3) Eastern Europe, Low, -.6 (.7).3 (.) -.83 (.43).5 (.7) Eastern Europe, Low, -.6 (.) -.94 (.3) -.76 (.38) -.8 (.) Eastern Europe, Medium, -.99 (.35).34 (.5) -.46 (.44).3 (.) Eastern Europe, Medium, -.9 (.4) -.45 (.) -.6 (.48) -.98 (.) Eastern Europe, Medium, (.3) -.57 (.7) -.67 (.5) -.3 (.9) Eastern Europe, High, -.9 (.4) -.3 (.3) -.48 (.38) -.57 (.34) Eastern Europe, High, -.78 (.43) -.9 (.3) -.49 (.5) -.99 (.) Eastern Europe, High, (.) -. (.) -.7 (.38) -.6 (.9) South-East Asia, Low, -.4 (.34).5 (.).86 (.43).88 (.3) South-East Asia, Low, -.46 (.34) -.3 (.) -.6 (.4) -.4 (.) South-East Asia, Low, (.44) -.75 (.) (.43) -.5 (.33) South-East Asia, Medium, -.8 (.63) -.75 (.) -.96 (.44) -.7 (.7) South-East Asia, Medium, -.83 (.8) -.7 (.3) -.3 (.38).33 (.) South-East Asia, Medium, (.35) -.74 (.3) ( (.7) South-East Asia, High, -. (.37) -.95 (.) -.78 (.38) -.5 (.) South-East Asia, High, -.95 (.4) -.99 (.4) -.65 (.53) -.8 (.34) Std Error.7 (.9).38 (.) This tabe reports the effects of the covariates and the group premiums on seers bid distribution and participation decisions for the specification presented in the paper. Coumns I(T), II(T), and I(P), II(P) report estimated coefficients for the bid distribution and probabiity of participation of the transitory and permanent seers respectivey. Average Score and Average Score denote interactions of the current average score variabe with the indicators for Ratings 3 and 3 Ratings 6. The stars,, indicate that a coefficient is significant at the 95% significance eve.

18 8 Tabe 3: Participation Decision and Bid Distributions: Competitive Effects I(T) II(T) I(P) II(P) North America, ow,. (.7). (.3).3 (.). (.5) North America, ow,. (.).5 (.) -.5 (.3). (.4) North America, medium, -.3 (.) -.89 (.) -. (.) -.4 (.4) North America, medium, -.5 (.8) -.3 (.5) -.4 (.) -.7 (.3) North America, high,.5 (.3). (.78) -.7 (.) -. (.3) North America, high, -.6 (.) -.85 (.) -.8 (.3) -.6 (.4) Eastern Europe, ow, -.5 (.). (.7) -. (.). (.5) Eastern Europe, ow, -.7 (.) -.5 (.) -.5 (.6) -.9 (.6) Eastern Europe, medium, -.5 (.3) -.7 (.4) -.3 (.) -.7 (.) Eastern Europe, medium,.34 (.3).6 (.). (.4). (.3) Eastern Europe, medium, 3 -. (.) -.6 (.5). (.) -.4 (.4) Eastern Europe, high, -. (.) -. (.7) -. (.3) -.3 (.5) Eastern Europe, high,.8 (.5).7 (.4). (.5).3 (.) Eastern Europe, high, (.4) -. (.9). (.3) -. (.3) South-East Asia, ow,. (.) -.7 (.3) -. (.) -. (.) South-East Asia, ow, -.3 (.4) -.4 (.8) -.8 (.3) -.9 (.) South-East Asia, ow, -.3 (.) -.6 (.) -. (.). (.3) South-East Asia, medium, -. (.) -.3 (.) -.4 (.3) -. (.5) South-East Asia, medium, -.3 (.) -. (.) -. (.) -.4 (.) South-East Asia, medium, 3.3 (.33).5 (.4) -. (.).7 (.5) South-East Asia, high, -.4 (.3) -.7 (.4) -. (.). (.) South-East Asia, high, -.4 (.) -.39 (.8) -.4 (.) -.7 (.3) This tabe reports the coefficients summarizing the impact of the various potentia competitors on seers bid distribution and participation decisions. Coumns I(T), II(T), and I(P), II(P) report estimated coefficients for the bid distribution and the probabiity of participation of transitory and permanent seers respectivey. The resuts are based on the data set consisting, 3 projects. The quaity eve for South and East Asia, ow score, Q =, is normaized to be equa to zero. The stars,, indicate that a coefficient is significant at the 95% significance eve.

19 9 Figure : Bid Functions Low Score Medium Score High Score North America Low Quaity High Quaity Diagona 3.5 ow quaity high quaity diagona 3.5 ow quaity high quaity diagona Bids.5 bids.5 bids costs costs Eastern Europe medium quaity high quaity diagona 3.5 Low Quaity Medium Quaity High Quaity Diagona 3.5 Low Quaity Medium Quaity High Quaity Diagona Bids.5 Bids.5 Bids South and East Asia Low Quaity Medium Quaity High Quaity Diagona 3.5 Low Quaity Medium Quaity High Quaity Diagona 3.5 Low Quaity High Quaity Diagona Bids.5 Bids.5 Bids The figure shows the equiibrium bidding strategies of permanent seers recovered from the first order conditions of bidders optimization program. The convexity at the upper end of the costs support arises due to presence of stochastic component in buyers tastes.

20 Figure : Density of Project s Cost Distribution Low Score Medium Score High Score North America 5 Low Quaity High Quaity Low Quaity High Quaity Density 6 Density Density Eastern Europe 8 6 Medium Quaity High Quaity Low Quaity Medium Quaity High Quaity Low Quaity Medium Quaity High Quaity Density Density 6 Density South and East Asia 4 5 Low Quaity Medium Quaity High Quaity 6 Low Quaity Medium Quaity High Quaity Medium Quaity High Quaity Density 6 Density 3 Density The figure shows the permanent seers distribution of costs recovered by combining estimated bidding strategies and bid distributions.

21 Using Standard as Opposed to Muti-Attribute Auctions In this section we evauate the wefare gains associated with using muti-attribute auction aocation mechanism rather than standard auction mechanism in an on-ine market. We abstract away from characteristics other than quaity and focus on Eastern European permanent seers with medium reputation scores as potentia auction participants. We set the number and the composition of the set of potentia bidders to match these statistics for a typica auction. The resuts are presented in the tabe beow. Coumn () and coumn () summarize the resuts of a muti-attribute and a standard auctions respectivey where a quaity eves are aowed to participate. In Coumns (3), in addition to using a standard auction mechanism, we aow the intermediary to imposes a pre-certification requirement that ensures that ony medium- and high-quaity bidders are aowed to participate in these auctions. We find that using a standard instead of a muti-attribute auction mechanism resuts in a substantia oss of vaue to buyers. Whie the prices paid by buyers decrease by 5.8% on average, this reduction is aso associated with 9.5% reduction in the average purchased quaity which in turn entais 6.5% reduction in buyer s vaue and 9.% reduction in the overa surpus. The reasons for these effects are cear. Our findings indicate that the provision of high-quaity service is substantiay more expensive than the provision of medium-quaity service. This makes highquaity seers reativey uncompetitive on price. Thus, they win a sma fraction of standard auctions and buyers end up mosty procuring the services of medium-quaity seers even if they woud be wiing to pay a premium for higher quaity eve. The standard auction mechanism aso does not aow to take into account seer s idiosyncratic suitabiity for a given project (ɛ component). Both effects contribute to the oss of vaue. The tota surpus is further reduced by ower profits earned in this market. Pre-certification requirement improves the average purchased quaity. However, this effect is off-set by the increase in price associated with the reduction in the number of participating seers. Overa, buyers in this market gain substantiay from using muti-attribute rather than standard auction as an aocation mechanism. These resuts importanty depend on the features of our environment. For exampe, in settings where owquaity seers are more competitive the effects are ikey to be even stronger.

22 Tabe 4: Standard Auction with Certification Muti-Attribute Standard Auction Auction Excude Low A Leves A Leves Quaity Average Price Average Net Expected Vaue Net Expected Vaue, τ α,9%..6. Net Expected Vaue, τ α,7% Net Expected Vaue, τ α,5% Net Expected Vaue, τ α,3% Net Expected Vaue, τ α,% Expected Tota Surpus Average Number of Entrants Market Shares High Quaity 3% 3% 7% Medium Quaity 5% 6% 66% Low Quaity 4% % Not Aocated % 5% 7% Note: In this tabe the expected vaue and the tota surpus are computed reative to the outside option. To faciitate the comparison across auction formats we assume that in a standard auction a buyer does not aocate the project if his net vaue from the owest bid is beow his outside option. References Ai, C., and X. Chen (3): Efficient Estimation of Modes with Conditiona Moment Restrictions Containing Unknown Functions, Econometrica, 7, Athey, S. (): Singe Crossing Properties and the Existence of Pure Strategy Equiibria in Games of Incompete Information, Econometrica, 69(5), Guerre, E., I. Perrigne, and Q. Vuong (): Optima Nonparametric Estimation of First-Price Auctions, Econometrica, 68(3), Krasnokutskaya, E., K. Song, and X. Tang (4): Uncovering Unobserved Agent Heterogeneity through Pairwise Comparisons, Working paper, University of Pennsyvania. Lee, S., K. Song, and Y.-J. Whang (3): Testing Functiona Inequaities, Journa of Econometrics, 7, 4 3.

Web Supplement for The Role of Quality in Internet Service Markets

Web Supplement for The Role of Quality in Internet Service Markets Web Supplement for The Role of Quality in Internet Service Markets Elena Krasnokutskaya Kyungchul Song Johns Hopkins University University of British Columbia Xun Tang Rice University January 16, 2018

More information

Automobile Prices in Market Equilibrium. Berry, Pakes and Levinsohn

Automobile Prices in Market Equilibrium. Berry, Pakes and Levinsohn Automobie Prices in Market Equiibrium Berry, Pakes and Levinsohn Empirica Anaysis of demand and suppy in a differentiated products market: equiibrium in the U.S. automobie market. Oigopoistic Differentiated

More information

MARKOV CHAINS AND MARKOV DECISION THEORY. Contents

MARKOV CHAINS AND MARKOV DECISION THEORY. Contents MARKOV CHAINS AND MARKOV DECISION THEORY ARINDRIMA DATTA Abstract. In this paper, we begin with a forma introduction to probabiity and expain the concept of random variabes and stochastic processes. After

More information

CS229 Lecture notes. Andrew Ng

CS229 Lecture notes. Andrew Ng CS229 Lecture notes Andrew Ng Part IX The EM agorithm In the previous set of notes, we taked about the EM agorithm as appied to fitting a mixture of Gaussians. In this set of notes, we give a broader view

More information

Two-Stage Least Squares as Minimum Distance

Two-Stage Least Squares as Minimum Distance Two-Stage Least Squares as Minimum Distance Frank Windmeijer Discussion Paper 17 / 683 7 June 2017 Department of Economics University of Bristo Priory Road Compex Bristo BS8 1TU United Kingdom Two-Stage

More information

A. Distribution of the test statistic

A. Distribution of the test statistic A. Distribution of the test statistic In the sequentia test, we first compute the test statistic from a mini-batch of size m. If a decision cannot be made with this statistic, we keep increasing the mini-batch

More information

A Brief Introduction to Markov Chains and Hidden Markov Models

A Brief Introduction to Markov Chains and Hidden Markov Models A Brief Introduction to Markov Chains and Hidden Markov Modes Aen B MacKenzie Notes for December 1, 3, &8, 2015 Discrete-Time Markov Chains You may reca that when we first introduced random processes,

More information

Alberto Maydeu Olivares Instituto de Empresa Marketing Dept. C/Maria de Molina Madrid Spain

Alberto Maydeu Olivares Instituto de Empresa Marketing Dept. C/Maria de Molina Madrid Spain CORRECTIONS TO CLASSICAL PROCEDURES FOR ESTIMATING THURSTONE S CASE V MODEL FOR RANKING DATA Aberto Maydeu Oivares Instituto de Empresa Marketing Dept. C/Maria de Moina -5 28006 Madrid Spain Aberto.Maydeu@ie.edu

More information

Do Schools Matter for High Math Achievement? Evidence from the American Mathematics Competitions Glenn Ellison and Ashley Swanson Online Appendix

Do Schools Matter for High Math Achievement? Evidence from the American Mathematics Competitions Glenn Ellison and Ashley Swanson Online Appendix VOL. NO. DO SCHOOLS MATTER FOR HIGH MATH ACHIEVEMENT? 43 Do Schoos Matter for High Math Achievement? Evidence from the American Mathematics Competitions Genn Eison and Ashey Swanson Onine Appendix Appendix

More information

Akaike Information Criterion for ANOVA Model with a Simple Order Restriction

Akaike Information Criterion for ANOVA Model with a Simple Order Restriction Akaike Information Criterion for ANOVA Mode with a Simpe Order Restriction Yu Inatsu * Department of Mathematics, Graduate Schoo of Science, Hiroshima University ABSTRACT In this paper, we consider Akaike

More information

SUPPLEMENTARY MATERIAL TO INNOVATED SCALABLE EFFICIENT ESTIMATION IN ULTRA-LARGE GAUSSIAN GRAPHICAL MODELS

SUPPLEMENTARY MATERIAL TO INNOVATED SCALABLE EFFICIENT ESTIMATION IN ULTRA-LARGE GAUSSIAN GRAPHICAL MODELS ISEE 1 SUPPLEMENTARY MATERIAL TO INNOVATED SCALABLE EFFICIENT ESTIMATION IN ULTRA-LARGE GAUSSIAN GRAPHICAL MODELS By Yingying Fan and Jinchi Lv University of Southern Caifornia This Suppementary Materia

More information

Supplemental Online Appendix for Trading Across Borders in On-line Auctions

Supplemental Online Appendix for Trading Across Borders in On-line Auctions Supplemental Online Appendix for Trading Across Borders in On-line Auctions Elena Krasnokutskaya Christian Terwiesch Johns Hopkins University Wharton School of Business Lucia Tiererova Johns Hopkins University

More information

Problem set 6 The Perron Frobenius theorem.

Problem set 6 The Perron Frobenius theorem. Probem set 6 The Perron Frobenius theorem. Math 22a4 Oct 2 204, Due Oct.28 In a future probem set I want to discuss some criteria which aow us to concude that that the ground state of a sef-adjoint operator

More information

Bayesian Learning. You hear a which which could equally be Thanks or Tanks, which would you go with?

Bayesian Learning. You hear a which which could equally be Thanks or Tanks, which would you go with? Bayesian Learning A powerfu and growing approach in machine earning We use it in our own decision making a the time You hear a which which coud equay be Thanks or Tanks, which woud you go with? Combine

More information

Expectation-Maximization for Estimating Parameters for a Mixture of Poissons

Expectation-Maximization for Estimating Parameters for a Mixture of Poissons Expectation-Maximization for Estimating Parameters for a Mixture of Poissons Brandon Maone Department of Computer Science University of Hesini February 18, 2014 Abstract This document derives, in excrutiating

More information

(1 ) = 1 for some 2 (0; 1); (1 + ) = 0 for some > 0:

(1 ) = 1 for some 2 (0; 1); (1 + ) = 0 for some > 0: Answers, na. Economics 4 Fa, 2009. Christiano.. The typica househod can engage in two types of activities producing current output and studying at home. Athough time spent on studying at home sacrices

More information

Separation of Variables and a Spherical Shell with Surface Charge

Separation of Variables and a Spherical Shell with Surface Charge Separation of Variabes and a Spherica She with Surface Charge In cass we worked out the eectrostatic potentia due to a spherica she of radius R with a surface charge density σθ = σ cos θ. This cacuation

More information

MATH 172: MOTIVATION FOR FOURIER SERIES: SEPARATION OF VARIABLES

MATH 172: MOTIVATION FOR FOURIER SERIES: SEPARATION OF VARIABLES MATH 172: MOTIVATION FOR FOURIER SERIES: SEPARATION OF VARIABLES Separation of variabes is a method to sove certain PDEs which have a warped product structure. First, on R n, a inear PDE of order m is

More information

Manipulation in Financial Markets and the Implications for Debt Financing

Manipulation in Financial Markets and the Implications for Debt Financing Manipuation in Financia Markets and the Impications for Debt Financing Leonid Spesivtsev This paper studies the situation when the firm is in financia distress and faces bankruptcy or debt restructuring.

More information

Stochastic Variational Inference with Gradient Linearization

Stochastic Variational Inference with Gradient Linearization Stochastic Variationa Inference with Gradient Linearization Suppementa Materia Tobias Pötz * Anne S Wannenwetsch Stefan Roth Department of Computer Science, TU Darmstadt Preface In this suppementa materia,

More information

FRST Multivariate Statistics. Multivariate Discriminant Analysis (MDA)

FRST Multivariate Statistics. Multivariate Discriminant Analysis (MDA) 1 FRST 531 -- Mutivariate Statistics Mutivariate Discriminant Anaysis (MDA) Purpose: 1. To predict which group (Y) an observation beongs to based on the characteristics of p predictor (X) variabes, using

More information

4 1-D Boundary Value Problems Heat Equation

4 1-D Boundary Value Problems Heat Equation 4 -D Boundary Vaue Probems Heat Equation The main purpose of this chapter is to study boundary vaue probems for the heat equation on a finite rod a x b. u t (x, t = ku xx (x, t, a < x < b, t > u(x, = ϕ(x

More information

(This is a sample cover image for this issue. The actual cover is not yet available at this time.)

(This is a sample cover image for this issue. The actual cover is not yet available at this time.) (This is a sampe cover image for this issue The actua cover is not yet avaiabe at this time) This artice appeared in a journa pubished by Esevier The attached copy is furnished to the author for interna

More information

A proposed nonparametric mixture density estimation using B-spline functions

A proposed nonparametric mixture density estimation using B-spline functions A proposed nonparametric mixture density estimation using B-spine functions Atizez Hadrich a,b, Mourad Zribi a, Afif Masmoudi b a Laboratoire d Informatique Signa et Image de a Côte d Opae (LISIC-EA 4491),

More information

Some Measures for Asymmetry of Distributions

Some Measures for Asymmetry of Distributions Some Measures for Asymmetry of Distributions Georgi N. Boshnakov First version: 31 January 2006 Research Report No. 5, 2006, Probabiity and Statistics Group Schoo of Mathematics, The University of Manchester

More information

BALANCING REGULAR MATRIX PENCILS

BALANCING REGULAR MATRIX PENCILS BALANCING REGULAR MATRIX PENCILS DAMIEN LEMONNIER AND PAUL VAN DOOREN Abstract. In this paper we present a new diagona baancing technique for reguar matrix pencis λb A, which aims at reducing the sensitivity

More information

XSAT of linear CNF formulas

XSAT of linear CNF formulas XSAT of inear CN formuas Bernd R. Schuh Dr. Bernd Schuh, D-50968 Kön, Germany; bernd.schuh@netcoogne.de eywords: compexity, XSAT, exact inear formua, -reguarity, -uniformity, NPcompeteness Abstract. Open

More information

Appendix for Stochastic Gradient Monomial Gamma Sampler

Appendix for Stochastic Gradient Monomial Gamma Sampler 3 4 5 6 7 8 9 3 4 5 6 7 8 9 3 4 5 6 7 8 9 3 3 3 33 34 35 36 37 38 39 4 4 4 43 44 45 46 47 48 49 5 5 5 53 54 Appendix for Stochastic Gradient Monomia Gamma Samper A The Main Theorem We provide the foowing

More information

Competitive Diffusion in Social Networks: Quality or Seeding?

Competitive Diffusion in Social Networks: Quality or Seeding? Competitive Diffusion in Socia Networks: Quaity or Seeding? Arastoo Fazei Amir Ajorou Ai Jadbabaie arxiv:1503.01220v1 [cs.gt] 4 Mar 2015 Abstract In this paper, we study a strategic mode of marketing and

More information

Explicit overall risk minimization transductive bound

Explicit overall risk minimization transductive bound 1 Expicit overa risk minimization transductive bound Sergio Decherchi, Paoo Gastado, Sandro Ridea, Rodofo Zunino Dept. of Biophysica and Eectronic Engineering (DIBE), Genoa University Via Opera Pia 11a,

More information

Week 6 Lectures, Math 6451, Tanveer

Week 6 Lectures, Math 6451, Tanveer Fourier Series Week 6 Lectures, Math 645, Tanveer In the context of separation of variabe to find soutions of PDEs, we encountered or and in other cases f(x = f(x = a 0 + f(x = a 0 + b n sin nπx { a n

More information

Appendix for Stochastic Gradient Monomial Gamma Sampler

Appendix for Stochastic Gradient Monomial Gamma Sampler Appendix for Stochastic Gradient Monomia Gamma Samper A The Main Theorem We provide the foowing theorem to characterize the stationary distribution of the stochastic process with SDEs in (3) Theorem 3

More information

Pricing Multiple Products with the Multinomial Logit and Nested Logit Models: Concavity and Implications

Pricing Multiple Products with the Multinomial Logit and Nested Logit Models: Concavity and Implications Pricing Mutipe Products with the Mutinomia Logit and Nested Logit Modes: Concavity and Impications Hongmin Li Woonghee Tim Huh WP Carey Schoo of Business Arizona State University Tempe Arizona 85287 USA

More information

$, (2.1) n="# #. (2.2)

$, (2.1) n=# #. (2.2) Chapter. Eectrostatic II Notes: Most of the materia presented in this chapter is taken from Jackson, Chap.,, and 4, and Di Bartoo, Chap... Mathematica Considerations.. The Fourier series and the Fourier

More information

Mat 1501 lecture notes, penultimate installment

Mat 1501 lecture notes, penultimate installment Mat 1501 ecture notes, penutimate instament 1. bounded variation: functions of a singe variabe optiona) I beieve that we wi not actuay use the materia in this section the point is mainy to motivate the

More information

PHYS 110B - HW #1 Fall 2005, Solutions by David Pace Equations referenced as Eq. # are from Griffiths Problem statements are paraphrased

PHYS 110B - HW #1 Fall 2005, Solutions by David Pace Equations referenced as Eq. # are from Griffiths Problem statements are paraphrased PHYS 110B - HW #1 Fa 2005, Soutions by David Pace Equations referenced as Eq. # are from Griffiths Probem statements are paraphrased [1.] Probem 6.8 from Griffiths A ong cyinder has radius R and a magnetization

More information

arxiv: v1 [math.co] 17 Dec 2018

arxiv: v1 [math.co] 17 Dec 2018 On the Extrema Maximum Agreement Subtree Probem arxiv:1812.06951v1 [math.o] 17 Dec 2018 Aexey Markin Department of omputer Science, Iowa State University, USA amarkin@iastate.edu Abstract Given two phyogenetic

More information

JENSEN S OPERATOR INEQUALITY FOR FUNCTIONS OF SEVERAL VARIABLES

JENSEN S OPERATOR INEQUALITY FOR FUNCTIONS OF SEVERAL VARIABLES PROCEEDINGS OF THE AMERICAN MATHEMATICAL SOCIETY Voume 128, Number 7, Pages 2075 2084 S 0002-99390005371-5 Artice eectronicay pubished on February 16, 2000 JENSEN S OPERATOR INEQUALITY FOR FUNCTIONS OF

More information

A Comparison Study of the Test for Right Censored and Grouped Data

A Comparison Study of the Test for Right Censored and Grouped Data Communications for Statistica Appications and Methods 2015, Vo. 22, No. 4, 313 320 DOI: http://dx.doi.org/10.5351/csam.2015.22.4.313 Print ISSN 2287-7843 / Onine ISSN 2383-4757 A Comparison Study of the

More information

Statistical Learning Theory: A Primer

Statistical Learning Theory: A Primer Internationa Journa of Computer Vision 38(), 9 3, 2000 c 2000 uwer Academic Pubishers. Manufactured in The Netherands. Statistica Learning Theory: A Primer THEODOROS EVGENIOU, MASSIMILIANO PONTIL AND TOMASO

More information

Formulas for Angular-Momentum Barrier Factors Version II

Formulas for Angular-Momentum Barrier Factors Version II BNL PREPRINT BNL-QGS-06-101 brfactor1.tex Formuas for Anguar-Momentum Barrier Factors Version II S. U. Chung Physics Department, Brookhaven Nationa Laboratory, Upton, NY 11973 March 19, 2015 abstract A

More information

Transport Cost and Optimal Number of Public Facilities

Transport Cost and Optimal Number of Public Facilities Transport Cost an Optima Number of Pubic Faciities Kazuo Yamaguchi Grauate Schoo of Economics, University of Tokyo June 14, 2006 Abstract We consier the number an ocation probem of pubic faciities without

More information

u(x) s.t. px w x 0 Denote the solution to this problem by ˆx(p, x). In order to obtain ˆx we may simply solve the standard problem max x 0

u(x) s.t. px w x 0 Denote the solution to this problem by ˆx(p, x). In order to obtain ˆx we may simply solve the standard problem max x 0 Bocconi University PhD in Economics - Microeconomics I Prof M Messner Probem Set 4 - Soution Probem : If an individua has an endowment instead of a monetary income his weath depends on price eves In particuar,

More information

4 Separation of Variables

4 Separation of Variables 4 Separation of Variabes In this chapter we describe a cassica technique for constructing forma soutions to inear boundary vaue probems. The soution of three cassica (paraboic, hyperboic and eiptic) PDE

More information

Partial permutation decoding for MacDonald codes

Partial permutation decoding for MacDonald codes Partia permutation decoding for MacDonad codes J.D. Key Department of Mathematics and Appied Mathematics University of the Western Cape 7535 Bevie, South Africa P. Seneviratne Department of Mathematics

More information

6 Wave Equation on an Interval: Separation of Variables

6 Wave Equation on an Interval: Separation of Variables 6 Wave Equation on an Interva: Separation of Variabes 6.1 Dirichet Boundary Conditions Ref: Strauss, Chapter 4 We now use the separation of variabes technique to study the wave equation on a finite interva.

More information

<C 2 2. λ 2 l. λ 1 l 1 < C 1

<C 2 2. λ 2 l. λ 1 l 1 < C 1 Teecommunication Network Contro and Management (EE E694) Prof. A. A. Lazar Notes for the ecture of 7/Feb/95 by Huayan Wang (this document was ast LaT E X-ed on May 9,995) Queueing Primer for Muticass Optima

More information

Stochastic Complement Analysis of Multi-Server Threshold Queues. with Hysteresis. Abstract

Stochastic Complement Analysis of Multi-Server Threshold Queues. with Hysteresis. Abstract Stochastic Compement Anaysis of Muti-Server Threshod Queues with Hysteresis John C.S. Lui The Dept. of Computer Science & Engineering The Chinese University of Hong Kong Leana Goubchik Dept. of Computer

More information

SydU STAT3014 (2015) Second semester Dr. J. Chan 18

SydU STAT3014 (2015) Second semester Dr. J. Chan 18 STAT3014/3914 Appied Stat.-Samping C-Stratified rand. sampe Stratified Random Samping.1 Introduction Description The popuation of size N is divided into mutuay excusive and exhaustive subpopuations caed

More information

C. Fourier Sine Series Overview

C. Fourier Sine Series Overview 12 PHILIP D. LOEWEN C. Fourier Sine Series Overview Let some constant > be given. The symboic form of the FSS Eigenvaue probem combines an ordinary differentia equation (ODE) on the interva (, ) with a

More information

Componentwise Determination of the Interval Hull Solution for Linear Interval Parameter Systems

Componentwise Determination of the Interval Hull Solution for Linear Interval Parameter Systems Componentwise Determination of the Interva Hu Soution for Linear Interva Parameter Systems L. V. Koev Dept. of Theoretica Eectrotechnics, Facuty of Automatics, Technica University of Sofia, 1000 Sofia,

More information

ASummaryofGaussianProcesses Coryn A.L. Bailer-Jones

ASummaryofGaussianProcesses Coryn A.L. Bailer-Jones ASummaryofGaussianProcesses Coryn A.L. Baier-Jones Cavendish Laboratory University of Cambridge caj@mrao.cam.ac.uk Introduction A genera prediction probem can be posed as foows. We consider that the variabe

More information

Asymptotic Properties of a Generalized Cross Entropy Optimization Algorithm

Asymptotic Properties of a Generalized Cross Entropy Optimization Algorithm 1 Asymptotic Properties of a Generaized Cross Entropy Optimization Agorithm Zijun Wu, Michae Koonko, Institute for Appied Stochastics and Operations Research, Caustha Technica University Abstract The discrete

More information

Common Value Auctions with Costly Entry

Common Value Auctions with Costly Entry Common Vaue Auctions wit Costy Entry Paui Murto Juuso Väimäki Marc 25, 2019 Abstract We anayze an affiiated common vaues auction wit costy participation wit an unknown number of competing bidders. We ca

More information

8 Digifl'.11 Cth:uits and devices

8 Digifl'.11 Cth:uits and devices 8 Digif'. Cth:uits and devices 8. Introduction In anaog eectronics, votage is a continuous variabe. This is usefu because most physica quantities we encounter are continuous: sound eves, ight intensity,

More information

Consistent linguistic fuzzy preference relation with multi-granular uncertain linguistic information for solving decision making problems

Consistent linguistic fuzzy preference relation with multi-granular uncertain linguistic information for solving decision making problems Consistent inguistic fuzzy preference reation with muti-granuar uncertain inguistic information for soving decision making probems Siti mnah Binti Mohd Ridzuan, and Daud Mohamad Citation: IP Conference

More information

Uniprocessor Feasibility of Sporadic Tasks with Constrained Deadlines is Strongly conp-complete

Uniprocessor Feasibility of Sporadic Tasks with Constrained Deadlines is Strongly conp-complete Uniprocessor Feasibiity of Sporadic Tasks with Constrained Deadines is Strongy conp-compete Pontus Ekberg and Wang Yi Uppsaa University, Sweden Emai: {pontus.ekberg yi}@it.uu.se Abstract Deciding the feasibiity

More information

Probabilistic Graphical Models

Probabilistic Graphical Models Schoo of Computer Science Probabiistic Graphica Modes Gaussian graphica modes and Ising modes: modeing networks Eric Xing Lecture 0, February 0, 07 Reading: See cass website Eric Xing @ CMU, 005-07 Network

More information

STA 216 Project: Spline Approach to Discrete Survival Analysis

STA 216 Project: Spline Approach to Discrete Survival Analysis : Spine Approach to Discrete Surviva Anaysis November 4, 005 1 Introduction Athough continuous surviva anaysis differs much from the discrete surviva anaysis, there is certain ink between the two modeing

More information

A NOTE ON QUASI-STATIONARY DISTRIBUTIONS OF BIRTH-DEATH PROCESSES AND THE SIS LOGISTIC EPIDEMIC

A NOTE ON QUASI-STATIONARY DISTRIBUTIONS OF BIRTH-DEATH PROCESSES AND THE SIS LOGISTIC EPIDEMIC (January 8, 2003) A NOTE ON QUASI-STATIONARY DISTRIBUTIONS OF BIRTH-DEATH PROCESSES AND THE SIS LOGISTIC EPIDEMIC DAMIAN CLANCY, University of Liverpoo PHILIP K. POLLETT, University of Queensand Abstract

More information

From Margins to Probabilities in Multiclass Learning Problems

From Margins to Probabilities in Multiclass Learning Problems From Margins to Probabiities in Muticass Learning Probems Andrea Passerini and Massimiiano Ponti 2 and Paoo Frasconi 3 Abstract. We study the probem of muticass cassification within the framework of error

More information

Reichenbachian Common Cause Systems

Reichenbachian Common Cause Systems Reichenbachian Common Cause Systems G. Hofer-Szabó Department of Phiosophy Technica University of Budapest e-mai: gszabo@hps.ete.hu Mikós Rédei Department of History and Phiosophy of Science Eötvös University,

More information

Stat 155 Game theory, Yuval Peres Fall Lectures 4,5,6

Stat 155 Game theory, Yuval Peres Fall Lectures 4,5,6 Stat 155 Game theory, Yuva Peres Fa 2004 Lectures 4,5,6 In the ast ecture, we defined N and P positions for a combinatoria game. We wi now show more formay that each starting position in a combinatoria

More information

An explicit Jordan Decomposition of Companion matrices

An explicit Jordan Decomposition of Companion matrices An expicit Jordan Decomposition of Companion matrices Fermín S V Bazán Departamento de Matemática CFM UFSC 88040-900 Forianópois SC E-mai: fermin@mtmufscbr S Gratton CERFACS 42 Av Gaspard Coriois 31057

More information

14 Separation of Variables Method

14 Separation of Variables Method 14 Separation of Variabes Method Consider, for exampe, the Dirichet probem u t = Du xx < x u(x, ) = f(x) < x < u(, t) = = u(, t) t > Let u(x, t) = T (t)φ(x); now substitute into the equation: dt

More information

Section 6: Magnetostatics

Section 6: Magnetostatics agnetic fieds in matter Section 6: agnetostatics In the previous sections we assumed that the current density J is a known function of coordinates. In the presence of matter this is not aways true. The

More information

Primal and dual active-set methods for convex quadratic programming

Primal and dual active-set methods for convex quadratic programming Math. Program., Ser. A 216) 159:469 58 DOI 1.17/s117-15-966-2 FULL LENGTH PAPER Prima and dua active-set methods for convex quadratic programming Anders Forsgren 1 Phiip E. Gi 2 Eizabeth Wong 2 Received:

More information

ESTIMATING UNOBSERVED INDIVIDUAL HETEROGENEITY THROUGH PAIRWISE COMPARISONS

ESTIMATING UNOBSERVED INDIVIDUAL HETEROGENEITY THROUGH PAIRWISE COMPARISONS ESTIMATING UNOBSERVED INDIVIDUAL HETEROGENEITY THROUGH PAIRWISE COMPARISONS ELENA KRASNOKUTSKAYA, KYUNGCHUL SONG, AND XUN TANG Abstract. This paper proposes a new method to study environments with unobserved

More information

Data Mining Technology for Failure Prognostic of Avionics

Data Mining Technology for Failure Prognostic of Avionics IEEE Transactions on Aerospace and Eectronic Systems. Voume 38, #, pp.388-403, 00. Data Mining Technoogy for Faiure Prognostic of Avionics V.A. Skormin, Binghamton University, Binghamton, NY, 1390, USA

More information

Statistics for Applications. Chapter 7: Regression 1/43

Statistics for Applications. Chapter 7: Regression 1/43 Statistics for Appications Chapter 7: Regression 1/43 Heuristics of the inear regression (1) Consider a coud of i.i.d. random points (X i,y i ),i =1,...,n : 2/43 Heuristics of the inear regression (2)

More information

Online Load Balancing on Related Machines

Online Load Balancing on Related Machines Onine Load Baancing on Reated Machines ABSTRACT Sungjin Im University of Caifornia at Merced Merced, CA, USA sim3@ucmerced.edu Debmaya Panigrahi Duke University Durham, NC, USA debmaya@cs.duke.edu We give

More information

II. PROBLEM. A. Description. For the space of audio signals

II. PROBLEM. A. Description. For the space of audio signals CS229 - Fina Report Speech Recording based Language Recognition (Natura Language) Leopod Cambier - cambier; Matan Leibovich - matane; Cindy Orozco Bohorquez - orozcocc ABSTRACT We construct a rea time

More information

Appendix of the Paper The Role of No-Arbitrage on Forecasting: Lessons from a Parametric Term Structure Model

Appendix of the Paper The Role of No-Arbitrage on Forecasting: Lessons from a Parametric Term Structure Model Appendix of the Paper The Roe of No-Arbitrage on Forecasting: Lessons from a Parametric Term Structure Mode Caio Ameida cameida@fgv.br José Vicente jose.vaentim@bcb.gov.br June 008 1 Introduction In this

More information

Physics 505 Fall 2007 Homework Assignment #5 Solutions. Textbook problems: Ch. 3: 3.13, 3.17, 3.26, 3.27

Physics 505 Fall 2007 Homework Assignment #5 Solutions. Textbook problems: Ch. 3: 3.13, 3.17, 3.26, 3.27 Physics 55 Fa 7 Homework Assignment #5 Soutions Textook proems: Ch. 3: 3.3, 3.7, 3.6, 3.7 3.3 Sove for the potentia in Proem 3., using the appropriate Green function otained in the text, and verify that

More information

Asynchronous Control for Coupled Markov Decision Systems

Asynchronous Control for Coupled Markov Decision Systems INFORMATION THEORY WORKSHOP (ITW) 22 Asynchronous Contro for Couped Marov Decision Systems Michae J. Neey University of Southern Caifornia Abstract This paper considers optima contro for a coection of

More information

Research Article On the Lower Bound for the Number of Real Roots of a Random Algebraic Equation

Research Article On the Lower Bound for the Number of Real Roots of a Random Algebraic Equation Appied Mathematics and Stochastic Anaysis Voume 007, Artice ID 74191, 8 pages doi:10.1155/007/74191 Research Artice On the Lower Bound for the Number of Rea Roots of a Random Agebraic Equation Takashi

More information

Homogeneity properties of subadditive functions

Homogeneity properties of subadditive functions Annaes Mathematicae et Informaticae 32 2005 pp. 89 20. Homogeneity properties of subadditive functions Pá Burai and Árpád Száz Institute of Mathematics, University of Debrecen e-mai: buraip@math.kte.hu

More information

Competition, Market Coverage, and Quality Choice in Interconnected Platforms

Competition, Market Coverage, and Quality Choice in Interconnected Platforms Competition, Market Coverage, and Quait Choice in Interconnected Patforms Pau Njoroge njoroge@mit.edu Asuman Ozdagar asuman@mit.edu Gabrie Weintraub gweintraub@coumbia.edu Nicoas Stier ns2224@coumbia.edu

More information

Physics 127c: Statistical Mechanics. Fermi Liquid Theory: Collective Modes. Boltzmann Equation. The quasiparticle energy including interactions

Physics 127c: Statistical Mechanics. Fermi Liquid Theory: Collective Modes. Boltzmann Equation. The quasiparticle energy including interactions Physics 27c: Statistica Mechanics Fermi Liquid Theory: Coective Modes Botzmann Equation The quasipartice energy incuding interactions ε p,σ = ε p + f(p, p ; σ, σ )δn p,σ, () p,σ with ε p ε F + v F (p p

More information

Online Appendices for The Economics of Nationalism (Xiaohuan Lan and Ben Li)

Online Appendices for The Economics of Nationalism (Xiaohuan Lan and Ben Li) Onine Appendices for The Economics of Nationaism Xiaohuan Lan and Ben Li) A. Derivation of inequaities 9) and 10) Consider Home without oss of generaity. Denote gobaized and ungobaized by g and ng, respectivey.

More information

Limits on Support Recovery with Probabilistic Models: An Information-Theoretic Framework

Limits on Support Recovery with Probabilistic Models: An Information-Theoretic Framework Limits on Support Recovery with Probabiistic Modes: An Information-Theoretic Framewor Jonathan Scarett and Voan Cevher arxiv:5.744v3 cs.it 3 Aug 6 Abstract The support recovery probem consists of determining

More information

The distribution of the number of nodes in the relative interior of the typical I-segment in homogeneous planar anisotropic STIT Tessellations

The distribution of the number of nodes in the relative interior of the typical I-segment in homogeneous planar anisotropic STIT Tessellations Comment.Math.Univ.Caroin. 51,3(21) 53 512 53 The distribution of the number of nodes in the reative interior of the typica I-segment in homogeneous panar anisotropic STIT Tesseations Christoph Thäe Abstract.

More information

On the evaluation of saving-consumption plans

On the evaluation of saving-consumption plans On the evauation of saving-consumption pans Steven Vanduffe Jan Dhaene Marc Goovaerts Juy 13, 2004 Abstract Knowedge of the distribution function of the stochasticay compounded vaue of a series of future

More information

Target Location Estimation in Wireless Sensor Networks Using Binary Data

Target Location Estimation in Wireless Sensor Networks Using Binary Data Target Location stimation in Wireess Sensor Networks Using Binary Data Ruixin Niu and Pramod K. Varshney Department of ectrica ngineering and Computer Science Link Ha Syracuse University Syracuse, NY 344

More information

Moreau-Yosida Regularization for Grouped Tree Structure Learning

Moreau-Yosida Regularization for Grouped Tree Structure Learning Moreau-Yosida Reguarization for Grouped Tree Structure Learning Jun Liu Computer Science and Engineering Arizona State University J.Liu@asu.edu Jieping Ye Computer Science and Engineering Arizona State

More information

Statistical Learning Theory: a Primer

Statistical Learning Theory: a Primer ??,??, 1 6 (??) c?? Kuwer Academic Pubishers, Boston. Manufactured in The Netherands. Statistica Learning Theory: a Primer THEODOROS EVGENIOU AND MASSIMILIANO PONTIL Center for Bioogica and Computationa

More information

IE 361 Exam 1. b) Give *&% confidence limits for the bias of this viscometer. (No need to simplify.)

IE 361 Exam 1. b) Give *&% confidence limits for the bias of this viscometer. (No need to simplify.) October 9, 00 IE 6 Exam Prof. Vardeman. The viscosity of paint is measured with a "viscometer" in units of "Krebs." First, a standard iquid of "known" viscosity *# Krebs is tested with a company viscometer

More information

More Scattering: the Partial Wave Expansion

More Scattering: the Partial Wave Expansion More Scattering: the Partia Wave Expansion Michae Fower /7/8 Pane Waves and Partia Waves We are considering the soution to Schrödinger s equation for scattering of an incoming pane wave in the z-direction

More information

c 2007 Society for Industrial and Applied Mathematics

c 2007 Society for Industrial and Applied Mathematics SIAM REVIEW Vo. 49,No. 1,pp. 111 1 c 7 Society for Industria and Appied Mathematics Domino Waves C. J. Efthimiou M. D. Johnson Abstract. Motivated by a proposa of Daykin [Probem 71-19*, SIAM Rev., 13 (1971),

More information

Melodic contour estimation with B-spline models using a MDL criterion

Melodic contour estimation with B-spline models using a MDL criterion Meodic contour estimation with B-spine modes using a MDL criterion Damien Loive, Ney Barbot, Oivier Boeffard IRISA / University of Rennes 1 - ENSSAT 6 rue de Kerampont, B.P. 80518, F-305 Lannion Cedex

More information

The Group Structure on a Smooth Tropical Cubic

The Group Structure on a Smooth Tropical Cubic The Group Structure on a Smooth Tropica Cubic Ethan Lake Apri 20, 2015 Abstract Just as in in cassica agebraic geometry, it is possibe to define a group aw on a smooth tropica cubic curve. In this note,

More information

First-Order Corrections to Gutzwiller s Trace Formula for Systems with Discrete Symmetries

First-Order Corrections to Gutzwiller s Trace Formula for Systems with Discrete Symmetries c 26 Noninear Phenomena in Compex Systems First-Order Corrections to Gutzwier s Trace Formua for Systems with Discrete Symmetries Hoger Cartarius, Jörg Main, and Günter Wunner Institut für Theoretische

More information

6.434J/16.391J Statistics for Engineers and Scientists May 4 MIT, Spring 2006 Handout #17. Solution 7

6.434J/16.391J Statistics for Engineers and Scientists May 4 MIT, Spring 2006 Handout #17. Solution 7 6.434J/16.391J Statistics for Engineers and Scientists May 4 MIT, Spring 2006 Handout #17 Soution 7 Probem 1: Generating Random Variabes Each part of this probem requires impementation in MATLAB. For the

More information

Biometrics Unit, 337 Warren Hall Cornell University, Ithaca, NY and. B. L. Raktoe

Biometrics Unit, 337 Warren Hall Cornell University, Ithaca, NY and. B. L. Raktoe NONISCMORPHIC CCMPLETE SETS OF ORTHOGONAL F-SQ.UARES, HADAMARD MATRICES, AND DECCMPOSITIONS OF A 2 4 DESIGN S. J. Schwager and w. T. Federer Biometrics Unit, 337 Warren Ha Corne University, Ithaca, NY

More information

Costly Voting with Correlated Preferences

Costly Voting with Correlated Preferences Costy Voting with Correated Preferences Jacob K. Goeree and Jens Grosser 1 CREED Facuteit voor Economie en Econometrie Universiteit van Amsterdam Roetersstraat 11, 1018 WB Amsterdam The Netherands Abstract

More information

AST 418/518 Instrumentation and Statistics

AST 418/518 Instrumentation and Statistics AST 418/518 Instrumentation and Statistics Cass Website: http://ircamera.as.arizona.edu/astr_518 Cass Texts: Practica Statistics for Astronomers, J.V. Wa, and C.R. Jenkins, Second Edition. Measuring the

More information

Cryptanalysis of PKP: A New Approach

Cryptanalysis of PKP: A New Approach Cryptanaysis of PKP: A New Approach Éiane Jaumes and Antoine Joux DCSSI 18, rue du Dr. Zamenhoff F-92131 Issy-es-Mx Cedex France eiane.jaumes@wanadoo.fr Antoine.Joux@ens.fr Abstract. Quite recenty, in

More information

17 Lecture 17: Recombination and Dark Matter Production

17 Lecture 17: Recombination and Dark Matter Production PYS 652: Astrophysics 88 17 Lecture 17: Recombination and Dark Matter Production New ideas pass through three periods: It can t be done. It probaby can be done, but it s not worth doing. I knew it was

More information

LECTURE NOTES 9 TRACELESS SYMMETRIC TENSOR APPROACH TO LEGENDRE POLYNOMIALS AND SPHERICAL HARMONICS

LECTURE NOTES 9 TRACELESS SYMMETRIC TENSOR APPROACH TO LEGENDRE POLYNOMIALS AND SPHERICAL HARMONICS MASSACHUSETTS INSTITUTE OF TECHNOLOGY Physics Department Physics 8.07: Eectromagnetism II October 7, 202 Prof. Aan Guth LECTURE NOTES 9 TRACELESS SYMMETRIC TENSOR APPROACH TO LEGENDRE POLYNOMIALS AND SPHERICAL

More information

Determining The Degree of Generalization Using An Incremental Learning Algorithm

Determining The Degree of Generalization Using An Incremental Learning Algorithm Determining The Degree of Generaization Using An Incrementa Learning Agorithm Pabo Zegers Facutad de Ingeniería, Universidad de os Andes San Caros de Apoquindo 22, Las Condes, Santiago, Chie pzegers@uandes.c

More information