REVIEW 2 major objections 3 minor 24 references
Informational Content of Auction Prices
T0 review · 2 major / 3 minor · reviewed 2026-08-08 · deepseek-v4-flash
Pith's one-line read The paper proves that a discriminatory auction's publicly reported top price is Lehmann more informative about a common value than a uniform-price auction's clearing price once the object-to-bidder ratio clears a threshold set by the…
desk verdict Genuinely new horizontal order-statistic comparison under Lehmann dominance; the auction interpretation rests on an unproved equilibrium existence claim that should be fixed before publication. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The load-bearing object is the cumulative score function $S_v(x) = -\partial_v \log F_v(x)$, the derivative with respect to the state $v$ of the negative log of the conditional CDF; MLRP makes $S_v$ non-increasing in $x$, and its endpoint values $S_v(\alpha_v)$ and $S_v(\omega_v)$ summarize how informative low and high signals are. Lehmann dominance is verified through the identity that $F$ dominates $G$ iff, whenever $F_v(x) = G_v(y)$, one has $-\partial_v \log F_v(x) \geq -\partial_v \log G_v(y)$. The proof then uses the binomial expression for the (k+1)st order statistic's CDF, $F_v^{(k+1)}(x) = B(k; n, 1 - F_v(x))$, together with the identity $\partial_q B(k;n,1-q) = (1/q)(n-k)b(k;n,1-q)$, to reduce the comparison to $n S_v(x) \geq (n-k) S_v(y)$ times a binomial ratio, which is bounded once the endpoint ratio condition holds. The auction-to-order-statistic link is carried by strictly increasing equilibrium bid functions, which turn observed prices into the corresponding signal order statistics.
What would settle it
A direct check would fix $n=3$, $k=1$ and a strict-MLRP family such as the scale family $F_v(x) = (x/v)^\alpha$ on $[0,v]$, compute the two order-statistic CDFs, and test whether every pair $(x,y)$ with $F_v^{(1)}(x) = F_v^{(2)}(y)$ satisfies $-\partial_v \log F_v^{(1)}(x) \geq -\partial_v \log F_v^{(2)}(y)$ whenever $k/n$ is above the paper's threshold; a violation would refute the theorem, and a separate numerical equilibrium computation would verify whether public prices actually invert to these order statistics.
Extended reading notes
Core claim
The paper's central claim is Theorem 6.1: for any signal family $F_v$ satisfying the monotone likelihood ratio property, $X_{(1:n)}$ Lehmann dominates $X_{(k+1:n)}$ if for every $v$, $k/n \geq 1 - S_v(\omega_v)/S_v(\alpha_v)$, with $S_v(x) = -\partial_v \log F_v(x)$. Because monotone equilibria let the discriminatory price be inverted to the highest signal and the uniform price to the (k+1)st signal, this order-statistic ranking is a ranking of the two auction formats. The proof starts from the equivalence that Lehmann dominance holds exactly when $-\partial_v \log F_v^{(1)}(x) \geq -\partial_v \log F_v^{(k+1)}(y)$ at equal quantiles, substitutes the binomial formula for the (k+1)st order statistic, and uses monotonicity of $S_v$ to turn a pointwise inequality into the ratio threshold. The paper also establishes qualitative necessity: dominance can hold only if high signals are unboundedly informative (with the right endpoint of support increasing in large samples), the left endpoint of support is constant, and $k/n$ is bounded away from zero as $n$ grows.
Load-bearing premise
The auction comparison collapses if either format fails to have a symmetric, strictly increasing equilibrium whose bid function the outside observer knows, so that the public price exactly reveals the highest or the (k+1)st highest signal; the paper states that strict MLRP guarantees this but gives no proof or direct multi-unit citation.
Editorial extensions
If this is right
- In scale families, the condition becomes $k/n \geq 1 - \rho(1)/\rho(0)$, so, for example, the truncated normal (threshold about 0.29) and truncated Gumbel (0.22) yield dominance once the object share is high enough, while power distributions yield dominance for every $k$ and $n$.
- In location families, the parallel condition $k/n \geq 1 - r(0)/r(-\infty)$ holds, extending the ranking to additive-error models.
- Because $X_{(1)}$ dominates $X_{(k+1)}$, the discriminatory auction's published top price dominates the uniform auction's clearing price even if the discriminatory auction announces additional statistics such as average or median winning bids.
- Under the same sufficient condition, the unanimity rule is welfare superior to any $k$-voting rule (including majority rule) in the jury model, and this holds for finite juries, not only in the large-$n$ limit.
- The necessary conditions show that if $k/n$ vanishes as $n$ grows, the highest order statistic cannot dominate the (k+1)st, so the uniform-price format may be the more informative one in sufficiently thin markets.
Reading between the lines
- The paper's proof does not depend on which bidder pays what, only on monotone invertibility of prices; a natural extension is to other sealed-bid formats or to markets where a regulator chooses which summary statistic, such as top, median, or average, to disclose, with a similar score threshold governing the comparison.
- Because the necessary conditions show dominance fails when $k/n \to 0$, the paper implies that in very thin markets the uniform-price format may carry more information; the authors mention this only as a contrast to information-aggregation results.
- The jury application suggests a testable implication: if conclusive innocence signals are present and the acquittal threshold requires at least $k+1$ votes, unanimity should outperform supermajority rules in simulated signal environments; the paper does not run such simulations.
- The threshold $1 - S_v(\omega_v)/S_v(\alpha_v)$ can in principle be estimated from data by fitting parametric signal distributions to bid sets, giving a practical decision rule for choosing or designing auction formats; this estimation route is not developed in the paper.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper compares the informational content, in the Lehmann sense, of the highest order statistic X_(1) and the (k+1)-th highest order statistic X_(k+1) of n i.i.d. signals drawn from a common-value experiment {F_v}. It motivates this comparison by a multi-unit auction model in which the discriminatory auction's publicly observed highest winning bid reveals X_(1), while the uniform-price auction's clearing price reveals X_(k+1). The main theorem (Theorem 6.1) gives a sufficient condition, k/n >= 1 - S_v(omega_v)/S_v(alpha_v), where S_v = -partial_v log F_v, under which X_(1) Lehmann dominates X_(k+1); the condition specializes to k/n >= 1 - rho(1)/rho(0) in scale families and to k/n >= 1 - r(0)/r(-infinity) in location families. Section 6.2 and the appendices establish qualitatively necessary conditions: unbounded likelihood ratios, no conclusive left-endpoint signals, and liminf k_n/n > 0. Section 7 applies the ranking to jury voting.
Significance. If the equilibrium-reduction step is supplied, the paper offers a clean, prior-free ranking of two auction formats in terms of the Lehmann informativeness of their public prices. The paper goes beyond local indices by using a global comparison criterion, covers fixed n rather than only asymptotics, and complements the vertical comparisons of Di Tillio, Ottaviani, and Sorensen with horizontal comparisons across order statistics. The explicit examples and the jury-voting application are useful, and the three qualitative necessity results in Section 6.2 are a genuine strength. The main derivations in Appendices A through C are coherent, and I found no algebraic error in the central inequality of Theorem 6.1 and its special cases. However, the advertised auction interpretation rests on an unproved equilibrium-existence and monotonicity assertion, and one key monotonicity fact used in the proof of Theorem 6.1 is asserted without proof. Once those points are filled, the order-statistic results themselves appear sound.
major comments (2)
- [Section 2 (equilibrium reduction)] The assertion that strict MLRP guarantees symmetric, strictly increasing equilibria in both the discriminatory and the uniform-price auctions is stated without proof and without a direct citation; footnote 5 merely says that weaker conditions suffice. This assertion is load-bearing: Section 2.1 inverts the observed price through the equilibrium bid function to recover X_(1) or X_(k+1), and the entire auction interpretation of Theorem 6.1 depends on that inversion. If no strictly increasing equilibrium exists, or if a flat low-price equilibrium of the uniform-price auction is selected, the public price need not be a monotone function of the relevant order statistic, and the comparison of X_(1) and X_(k+1) would not apply to the auctions. The authors should provide a proof or a precise citation for the multi-unit existence and monotonicity result, and they should state explicitly which equilibrium is being selected.
- [Appendix B.1 and Section 6.1] The proof of Theorem 6.1 uses the claim that S_v(x) is non-increasing in x, introduced in Section 6.1 and repeated in Appendix B.1 as 'easy to verify,' and Proposition 6.3 uses the strict version S'_v(y) < 0. This monotonicity is central because it is what allows inequality (13) to be bounded by the endpoint values S_v(omega_v) and S_v(alpha_v). The fact is true under MLRP via reverse hazard rate dominance, but as written it is unproved. Please add a lemma with a proof (or a standard reference) for the non-increasing property and a clear statement of the strict version needed in the proof of Proposition 6.3.
minor comments (3)
- [Section 6.2.3] The informal proof of Proposition 6.3 uses the notation L({l,m}) without defining it in the body; the formal proof in Appendix C.3 is correct but dense, and a one-sentence definition would improve readability.
- [Section 7] The statement that 'the unanimity rule is welfare superior to the k voting rule' can confuse readers because the paper also says that the unanimity rule has k=0; the condition in Proposition 7.1 is imposed on the alternative rule's k, not on the unanimity rule's k. Please clarify this in the text.
- [References] The reference 'Hollander, F. (2000)' should be 'den Hollander, F. (2000)' for the large-deviations monograph.
Circularity Check
No circularity: the Lehmann comparison of X_(1) and X_(k+1) is derived from the definitions of order statistics and the Lehmann criterion, with no fitted inputs and no load-bearing self-citation.
full rationale
The paper's central derivation is self-contained. The comparison of the highest and (k+1)st highest order statistics is reduced, via equation (1), to comparing derivatives of -log F_v for the two order-statistic distributions, and those distributions are computed directly from F_v in equation (2). Theorem 6.1 then follows from bounding the resulting inequality using the monotonicity of the cumulative score function S_v, without assuming the conclusion. The special cases of scale and location families are obtained by substituting S_v(x) = (1/v) rho(x/v) and S_v(x) = r(x-v), respectively, which are exact identities rather than fitted relations. No parameter is fitted to data and then renamed as a prediction; the sufficient conditions are stated in terms of the primitives F_v, k, and n. The auction interpretation relies on the asserted existence of symmetric, strictly increasing equilibria, so that the public price inverts to X_(1) or X_(k+1). That assertion is an unproved premise or an external result cited to Milgrom and Weber (1982), not a circular step: the paper does not define the equilibrium so as to force the comparison. The jury-voting application uses the mapping from Di Tillio et al. (2026), an external work by different authors, as a bridge between voting outcomes and order statistics; again this is independent support, not a self-citation. There are no steps in which a proven result is equivalent by construction to an input, and no instance of a fitted quantity being called a prediction. The main unproved points are the existence of monotone equilibria and some technical regularity conditions, but these are correctness or completeness concerns, not circularity. Accordingly, the appropriate circularity score is 0.
Assumptions & free parameters
assumptions (5)
- domain assumption Strict MLRP on signal distributions: f(x|v)/f(x|w) is strictly increasing in x for v > w.
- domain assumption Symmetric strictly increasing equilibria exist in multi-unit discriminatory and uniform-price auctions.
- standard math Lehmann (1988) characterization of informativeness for MLRP experiments.
- standard math The cumulative score S_v(x) is non-increasing in x.
- standard math Large deviation principle for binomial tails, as used in Lemmas C.2 and C.3.
Cite this review
Pith. "Pith review of Informational Content of Auction Prices." pith.science (2026). https://pith.science/paper/5QEK5RM7
@misc{pith2026260804332,
author = {Pith},
title = {Pith review of: Informational Content of Auction Prices},
year = {2026},
howpublished = {\url{https://pith.science/paper/5QEK5RM7}},
note = {Machine review of arXiv:2608.04332}
}
read the original abstract
We study auctions of k identical objects to n bidders, each of whom wants at most one. The objects have a common but unknown value and the bidders receive private signals about this value. The discriminatory price auction and the uniform-price auction are compared in terms of how informative the resulting auction prices (not bids) are in conveying the true value to an outside observer/investor. Since both auctions have symmetric, monotone equilibria, the problem reduces to comparing the informativeness of the highest order statistic of a sample to the (k+1)st highest. We find sufficient conditions under which the highest order statistic is superior---in the sense of Lehmann---in this regard. The sufficient conditions involve the informativeness of high versus low signals and the ratio k/n of objects to bidders. These conditions are also qualitatively necessary.
Reference graph
Works this paper leans on
-
[1]
Strategic sample selection , author=. Econometrica , volume=. 2021 , publisher=
work page 2021
-
[2]
Information Comparison of Order Statistics, with Applications to Auctions and Voting
Information Comparison of Order Statistics, with Applications to Auctions and Voting , author=. arXiv preprint arXiv:2607.10435 , year=
- [3]
-
[4]
Review of Economic Studies , volume=
Strategic foundations of efficient rational expectations , author=. Review of Economic Studies , volume=. 2024 , publisher=
work page 2024
-
[5]
Rational expectations, information acquisition, and competitive bidding , author=. Econometrica , pages=. 1981 , publisher=
work page 1981
-
[6]
The loser's curse and information aggregation in common value auctions , author=. Econometrica , pages=. 1997 , publisher=
work page 1997
-
[7]
Selected Works of EL Lehmann , pages=
Comparing location experiments , author=. Selected Works of EL Lehmann , pages=. 2011 , publisher=
work page 2011
-
[8]
2007 , publisher=
Stochastic orders , author=. 2007 , publisher=
2007
Show all 24 references
-
[9]
Proceedings of the Second Berkeley Symposium on Mathematical Statistics and Probability , volume=
Comparison of experiments , author=. Proceedings of the Second Berkeley Symposium on Mathematical Statistics and Probability , volume=
-
[10]
The Annals of Statistics , pages=
Comparing Location Experiments , author=. The Annals of Statistics , pages=. 1988 , publisher=
1988
-
[11]
Econometrica , volume=
Information acquisition in auctions , author=. Econometrica , volume=. 2000 , publisher=
2000
-
[12]
2007 , publisher=
Life distributions , author=. 2007 , publisher=
2007
-
[13]
Econometrica , volume=
Information aggregation in common value auctions , author=. Econometrica , volume=. 2002 , publisher=
2002
-
[14]
Econometrica , pages=
A theory of auctions and competitive bidding , author=. Econometrica , pages=. 1982 , publisher=
1982
-
[15]
2002 , publisher=
Statistical Inference , author=. 2002 , publisher=
2002
-
[16]
arXiv preprint arXiv:2211.01688 , year=
Nearly tight universal bounds for the binomial tail probabilities , author=. arXiv preprint arXiv:2211.01688 , year=
-
[17]
2000 , publisher=
Large deviations , author=. 2000 , publisher=
2000
-
[18]
2008 , publisher=
A first course in order statistics , author=. 2008 , publisher=
2008
-
[19]
Review of Economic Studies , volume=
A bidding model of perfect competition , author=. Review of Economic Studies , volume=. 1977 , publisher=
1977
-
[20]
Games and Economic Behavior , volume=
A Bayesian model of voting in juries , author=. Games and Economic Behavior , volume=. 2001 , publisher=
2001
-
[21]
The American economic review , pages=
The swing voter's curse , author=. The American economic review , pages=. 1996 , publisher=
1996
-
[22]
American Political Science Review , volume=
Convicting the innocent: The inferiority of unanimous jury verdicts under strategic voting , author=. American Political Science Review , volume=. 1998 , publisher=
1998
-
[23]
Annals of the Institute of Statistical Mathematics , volume=
Tukey's linear sensitivity and order statistics , author=. Annals of the Institute of Statistical Mathematics , volume=. 1994 , publisher=
1994
-
[24]
Econometrica , volume=
Comparative statics, informativeness, and the interval dominance order , author=. Econometrica , volume=. 2009 , publisher=
2009
Reviewed August 8, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.