Inference in Auctions with Many Bidders Using Transaction Prices

Federico A. Bugni; Yulong Wang

REVIEW 3 major objections 2 minor 12 references

Inference in Auctions with Many Bidders Using Transaction Prices

T0 review · 3 major / 2 minor · reviewed 2026-05-24 · grok-4.3

Challenge this review Download PDF Read on arXiv ↗

Pith's one-line read Transaction prices support asymptotically exact inference on winner utility, seller revenue, and valuation tails when bidder numbers grow large with fixed auctions.

desk verdict The paper pushes inference on revenue, utility and valuation tails from transaction prices alone under fixed-T large-n asymptotics, but the limits are likely degenerate and the abstract supplies no fix. read the letter →

arxiv 2311.09972 v4 submitted 2023-11-16 econ.EM

Federico A. Bugni , Yulong Wang This is my paper

classification econ.EM

keywords auctionsinferencetransactionpricesmanybiddersfirst-pricesecond-pricevaluationdistributionrevenue

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper develops inference procedures for first-price and second-price auctions in an asymptotic setting where the number of bidders increases without bound while the number of auctions is held fixed. The methods permit construction of estimators and confidence sets for the winner's expected utility, the seller's expected revenue, and the tail of the valuation distribution that become exact in the limit, and these procedures use only the observed transaction prices. A sympathetic reader would care because many empirical settings, including online auctions and art sales, feature high bidder participation yet record only the winning price rather than the full bid vector. The paper verifies the finite-sample performance through simulations and demonstrates the approach on Hong Kong vehicle license data.

What carries the argument

The asymptotic limiting distribution of the transaction price as the number of bidders tends to infinity.

What would settle it

Generating auction data with a very large number of bidders per auction and a small number of auctions, then checking if the constructed intervals for expected revenue contain the true value at the correct rate, would falsify the claim if coverage fails.

Watch

Extended reading notes

Core claim

Under the asymptotic framework in which the number of bidders per auction tends to infinity with the number of auctions fixed, the distribution of the transaction price converges in a manner that allows asymptotically exact inference on the winner's expected utility, the seller's expected revenue, and the tail of the valuation distribution using only transaction price data, and this holds for both first-price and second-price sealed-bid auctions.

Load-bearing premise

The asymptotic regime with bidder numbers tending to infinity and a fixed number of auctions yields the limiting distributions that justify the inference methods.

Editorial extensions

If this is right

The seller's expected revenue can be inferred without observing losing bids.
Confidence intervals for the winner's expected utility achieve correct coverage in large bidder samples.
The upper tail of the bidders' valuation distribution can be estimated from price data alone.
These results apply equally to first-price and second-price formats under the many-bidder asymptotics.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

Platforms observing many participants could evaluate auction design changes using only sale prices.
The fixed-auction dimension suggests that inference power comes from variation across different items rather than repeated bidding.
Similar techniques might apply in procurement settings with many suppliers competing for contracts.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Desk Editor's Note

read the letter

The new piece is the asymptotic regime itself: T fixed while n grows, combined with an estimator that uses only the observed transaction prices to target winner utility, seller revenue, and the upper tail of valuations in both first-price and second-price settings. That combination is not standard in the auction literature, and the Hong Kong single-letter plate application shows a setting where full bid data are hard to get. The simulations are presented as evidence that the methods work in moderate samples. Those are the concrete contributions. The soft spot is exactly the one in the stress-test note. Under ordinary IPV assumptions the transaction price converges in probability to the upper endpoint of the support as n rises, so with T fixed the T prices all pile up at essentially the same point. Any statistic built from them then has a degenerate limit unless an explicit centering or scaling sequence is supplied and can itself be recovered from the prices. The abstract gives no sign that such a sequence is identified or estimated without knowing n or the density at the top. If the full derivations do not deliver a non-degenerate, estimable limit, the claimed asymptotically exact inference does not hold. Because the assessment here rests on the abstract and the stress-test logic, the paper is worth a referee only if the authors can show the rate is recoverable from the transaction prices alone. Otherwise it is not ready for serious review. I would not bring it to a reading group or cite it until that step is verified.

Referee Report

3 major / 2 minor

Summary. The paper develops methods for asymptotically exact inference on auction objects including the winner's expected utility, seller's expected revenue, and the tail of the valuation distribution in first-price and second-price sealed-bid IPV auctions. It employs a fixed-T, large-n asymptotic regime using only observed transaction prices, supports the claims with simulations, and applies the methods to Hong Kong single-letter vehicle license auctions.

Significance. If the limiting distributions are non-degenerate and the required normalizations are identified from the transaction prices alone, the framework would permit inference in empirically relevant settings with high bidder counts but few auctions. The approach is data-minimal and targets policy-relevant quantities directly.

major comments (3)

[§2] §2 (asymptotic framework) and the main limiting theorem: the central claim of non-degenerate, asymptotically exact inference requires explicit centering and scaling sequences for the transaction prices that remain estimable when T is fixed. Under standard IPV assumptions the winning bid converges in probability to the upper endpoint of the support as n→∞; with T fixed this produces a degenerate point-mass limit for any un-normalized statistic. The manuscript must derive the rate (typically involving n and the density or tail index at the upper endpoint) and show that this rate is recoverable from the T prices without knowledge of n.
[§3] Theorem on tail and revenue estimators (likely §3): the paper must verify that the proposed estimators for the valuation tail and expected revenue remain consistent and asymptotically normal after the necessary normalization, and that the normalization constants do not require direct observation of n or the full valuation distribution. If the normalization is itself estimated from the same T prices, the joint limiting distribution and its effect on inference must be derived.
[§4] Simulation design (§4): the reported Monte Carlo experiments must include designs with large n and small fixed T that match the asymptotic regime, report the exact data-generating process (including the upper endpoint and density), and display coverage probabilities or interval lengths for the target quantities (winner utility, revenue, tail) rather than only point-estimate accuracy.

minor comments (2)

Notation for the transaction price and the number of bidders should be introduced once and used consistently; the abstract and introduction use slightly different phrasing for the same objects.
The application section would benefit from a table reporting the estimated quantities with the new standard errors alongside any existing benchmarks.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the thoughtful and detailed report. We address each major comment below and will revise the manuscript accordingly to strengthen the exposition and empirical validation.

read point-by-point responses

Referee: [§2] §2 (asymptotic framework) and the main limiting theorem: the central claim of non-degenerate, asymptotically exact inference requires explicit centering and scaling sequences for the transaction prices that remain estimable when T is fixed. Under standard IPV assumptions the winning bid converges in probability to the upper endpoint of the support as n→∞; with T fixed this produces a degenerate point-mass limit for any un-normalized statistic. The manuscript must derive the rate (typically involving n and the density or tail index at the upper endpoint) and show that this rate is recoverable from the T prices without knowledge of n.

Authors: We agree that the explicit form of the centering and scaling sequences, together with their recoverability from the T transaction prices alone, requires clearer derivation. Section 2 presents the fixed-T large-n framework and limiting distributions, but we will expand it in revision to state the precise rates (involving the tail index or density at the upper endpoint) and prove that the normalizers are consistently estimable from the observed prices without knowledge of n or the full distribution. revision: yes
Referee: [§3] Theorem on tail and revenue estimators (likely §3): the paper must verify that the proposed estimators for the valuation tail and expected revenue remain consistent and asymptotically normal after the necessary normalization, and that the normalization constants do not require direct observation of n or the full valuation distribution. If the normalization is itself estimated from the same T prices, the joint limiting distribution and its effect on inference must be derived.

Authors: We concur that the joint limiting distribution must be derived when normalizers are estimated from the same T prices. The manuscript states the main consistency and normality results for the tail and revenue estimators, but we will add the joint asymptotic analysis in the revision to confirm that the feasible estimators (with estimated normalizers) remain asymptotically normal and that the resulting inference procedures are valid. revision: yes
Referee: [§4] Simulation design (§4): the reported Monte Carlo experiments must include designs with large n and small fixed T that match the asymptotic regime, report the exact data-generating process (including the upper endpoint and density), and display coverage probabilities or interval lengths for the target quantities (winner utility, revenue, tail) rather than only point-estimate accuracy.

Authors: The referee correctly notes that the simulation design should more closely match the asymptotic regime and report inferential performance. We will revise Section 4 to include Monte Carlo designs with large n and small fixed T, fully specify the DGP (including upper endpoint and density), and report coverage probabilities and interval lengths for winner utility, revenue, and tail quantities in addition to point-estimate metrics. revision: yes

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; derivation is self-contained asymptotic analysis

full rationale

The paper presents a direct asymptotic theory for inference on auction features (winner utility, seller revenue, valuation tails) under n→∞ with T fixed, using only transaction prices. No quoted equations or steps reduce the target quantities to fitted parameters by construction, nor do they rely on self-citation chains for uniqueness or ansatz smuggling. The claimed limiting distributions are derived from the model primitives rather than renamed inputs, and the approach does not exhibit self-definitional or fitted-input-called-prediction patterns. This is the standard case of an independent statistical derivation.

Assumptions & free parameters 0 free parameters · 0 assumptions · 0 invented entities

Only the abstract is available; no equations, assumptions, or data details are provided to identify free parameters, axioms, or invented entities.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Inference in Auctions with Many Bidders Using Transaction Prices." pith.science (2026). https://pith.science/paper/2311.09972

@misc{pith2026231109972,
  author       = {Pith},
  title        = {Pith review of: Inference in Auctions with Many Bidders Using Transaction Prices},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2311.09972}},
  note         = {Machine review of arXiv:2311.09972}
}

read the original abstract

This paper studies inference in first-price and second-price sealed-bid auctions with many bidders, using an asymptotic framework where the number of bidders increases while the number of auctions remains fixed. Our approach enables asymptotically exact inference on key features, such as the winner's expected utility, the seller's expected revenue, and the tail of the valuation distribution, using only transaction price data. Our simulations demonstrate the accuracy of the methods in finite samples. We apply our methods to Hong Kong vehicle license auctions, focusing on high-priced, single-letter plates. Other relevant applications include online and art auctions.

Figures

Figures reproduced from arXiv: 2311.09972 by the authors.

Discussion (0). Sign in to comment.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Foundation/AbsoluteFloorClosure.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

asymptotic framework where the number of bidders increases while the number of auctions remains fixed... tail index ξ... G_ξ(x)
IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

Lemma 3.1... joint density of ˜Z... Γ(2(N+2))... exp(−2(N+2)ln(...))

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

12 extracted references · 12 canonical work pages

[1]

This conclusion relies on x ∈ ¯Sξ (and so Gξ(x)2 > 0), xK → x ∈ ¯Sξ as K → ∞ , the continuity of Gξ, (A.35), (A.42), and van der Vaart 45 (1998, Lemma 2.11)

As K → ∞ , we can show that the numerator and the denominator of the right-hand side of (A.43) converge to zero and Gξ(x)2 > 0, respectively. This conclusion relies on x ∈ ¯Sξ (and so Gξ(x)2 > 0), xK → x ∈ ¯Sξ as K → ∞ , the continuity of Gξ, (A.35), (A.42), and van der Vaart 45 (1998, Lemma 2.11). From this and (A.43), the desired result follows. Lemma A...

work page 1998
[2]

P (µK ∈ U(P)) → Pξ(Yµ ∈ ˜U( ˜X)),

work page
[3]

E[lg(U(P))]/aK → Eξ[κξ( ˜X) lg(˜U( ˜X))]. Proof. This proof follows from the same arguments used to prove Theorem 3.1. The main difference between the proofs is that we replace {(Pj − bK)/aK : j = 1, . . . , N} with {LKj((V(1),j − bK)/aK) : j = 1, . . . , N} with LKj as in (A.34) (instead of {(V(2),j − bK)/aK : j = 1, . . . , N} as in second-price auction...

work page
[4]

P (πK ∈ U(P)) → Pξ(Yπ ∈ ˜U( ˜X)),

work page
[5]

E[lg(U(P))]/aK → Eξ[κξ( ˜X) lg(˜U( ˜X))]. Proof. This proof follows from the same arguments as in Theorems 3.1 and A.1. Theorem A.3. Assume (2.1) holds. In the hypothesis testing problem in (A.5), the test defined by (A.6) satisfies the following properties:

work page
[6]

It is asymptotically valid and level α, i.e., limK→∞ Eξ0[φ∗ K(P)] = α

work page
[7]

It is asymptotically efficient. Proof. This proof follows from the same arguments as in Theorems 3.3 and 3.4. 47 A.4 Computational details A.4.1 Second-price auctions This section provides computational details for objects introduced in Section 3. Throughout this section, we use ˜ z= (z1, . . . , zN) ∈ Σ and N = n − 2. First, recall that f˜Z|ξ(˜ z) is as ...

work page 2015
[8]

, ξM } between ξ1 = inf{Ξ} and ξM+1 = sup{Ξ} (we use M = 50 uniformly located points between ξ1 and ξM+1)

Discretize Ξ into a fine grid Ξ M ≡ {ξ1, ξ2, . . . , ξM } between ξ1 = inf{Ξ} and ξM+1 = sup{Ξ} (we use M = 50 uniformly located points between ξ1 and ξM+1). Set s = 1, and define an arbitrary set of initial positive weights λ(s) = {λ(s) m : m = 1, . . . , M} over ΞM (we use a uniform weights, i.e., λ(1) = {1/M, . . . ,1/M})

work page

Show all 12 references

[9]

, M, simulate a large number B of n i.i.d

For each m = 1, . . . , M, simulate a large number B of n i.i.d. draws the EV distribution with parameter ξm (we use B = 10 , 000). For each m = 1 , . . . , M and b = 1 , . . . , B, the samples is denoted by Zξm(b) = {Zξm,1(b), . . . , Zξm,n(b)}. By applying (3.9) to each samp...
[10]

For each m = 1, . . . , M, use the random draws in step 2) to approximate the limiting coverage probabilities for parameter ξm in the following manner: (a) For the winner’s expected utility, we approximate Pξm(Γ(1 − ξm)/(Z(n) − Z(1)) ∈ U(˜Z)) with ˆPm ≡ 1 B BX b=1 1 Γ(1−ξm) Zξ...
[11]

, M, where α is the significance level and ε > 0 is a small step length (we use ε = 0.05)

Update the weights by setting λ(s+1) m = λ(s) m + ε((1 − ˆPm) − α) for all m = 1, . . . , M, where α is the significance level and ε > 0 is a small step length (we use ε = 0.05). Intuitively, the weight on ξm is decreased or increased if the CI has overcoverage and undercovera...
[12]

Repeat steps 3)-4) a large number of times S (we use S = 2, 000), to get λ(S) = {λ(S) m : m = 1, . . . , M}. The Lagrange multipliers Λ is obtained by interpolating λ(S) from ΞM to Ξ. If the algorithm’s tuning parameters are appropriately chosen, the Lagrange multipliers gener...

1964

Pith tools

Reviewed May 24, 2026 · model on record in the stance chip above.

[1] [1]

This conclusion relies on x ∈ ¯Sξ (and so Gξ(x)2 > 0), xK → x ∈ ¯Sξ as K → ∞ , the continuity of Gξ, (A.35), (A.42), and van der Vaart 45 (1998, Lemma 2.11)

As K → ∞ , we can show that the numerator and the denominator of the right-hand side of (A.43) converge to zero and Gξ(x)2 > 0, respectively. This conclusion relies on x ∈ ¯Sξ (and so Gξ(x)2 > 0), xK → x ∈ ¯Sξ as K → ∞ , the continuity of Gξ, (A.35), (A.42), and van der Vaart 45 (1998, Lemma 2.11). From this and (A.43), the desired result follows. Lemma A...

work page 1998

[2] [2]

P (µK ∈ U(P)) → Pξ(Yµ ∈ ˜U( ˜X)),

work page

[3] [3]

E[lg(U(P))]/aK → Eξ[κξ( ˜X) lg(˜U( ˜X))]. Proof. This proof follows from the same arguments used to prove Theorem 3.1. The main difference between the proofs is that we replace {(Pj − bK)/aK : j = 1, . . . , N} with {LKj((V(1),j − bK)/aK) : j = 1, . . . , N} with LKj as in (A.34) (instead of {(V(2),j − bK)/aK : j = 1, . . . , N} as in second-price auction...

work page

[4] [4]

P (πK ∈ U(P)) → Pξ(Yπ ∈ ˜U( ˜X)),

work page

[5] [5]

E[lg(U(P))]/aK → Eξ[κξ( ˜X) lg(˜U( ˜X))]. Proof. This proof follows from the same arguments as in Theorems 3.1 and A.1. Theorem A.3. Assume (2.1) holds. In the hypothesis testing problem in (A.5), the test defined by (A.6) satisfies the following properties:

work page

[6] [6]

It is asymptotically valid and level α, i.e., limK→∞ Eξ0[φ∗ K(P)] = α

work page

[7] [7]

It is asymptotically efficient. Proof. This proof follows from the same arguments as in Theorems 3.3 and 3.4. 47 A.4 Computational details A.4.1 Second-price auctions This section provides computational details for objects introduced in Section 3. Throughout this section, we use ˜ z= (z1, . . . , zN) ∈ Σ and N = n − 2. First, recall that f˜Z|ξ(˜ z) is as ...

work page 2015

[8] [8]

, ξM } between ξ1 = inf{Ξ} and ξM+1 = sup{Ξ} (we use M = 50 uniformly located points between ξ1 and ξM+1)

Discretize Ξ into a fine grid Ξ M ≡ {ξ1, ξ2, . . . , ξM } between ξ1 = inf{Ξ} and ξM+1 = sup{Ξ} (we use M = 50 uniformly located points between ξ1 and ξM+1). Set s = 1, and define an arbitrary set of initial positive weights λ(s) = {λ(s) m : m = 1, . . . , M} over ΞM (we use a uniform weights, i.e., λ(1) = {1/M, . . . ,1/M})

work page