Nonparametric Identification of First-Price Auction with Unobserved Competition: A Density Discontinuity Framework

Emmanuel Guerre; Yao Luo

REVIEW 2 major objections 4 minor 1 cited by

Nonparametric Identification of First-Price Auction with Unobserved Competition: A Density Discontinuity Framework

T0 review · 2 major / 4 minor · reviewed 2026-08-14 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The winning bid alone can identify both the bidder-count distribution and the private-value distribution in a first-price auction.

desk verdict The density-discontinuity identification of unobserved N is new and the theorem is sound, but the full-support assumption is narrow and the version I read does not match its own abstract's empirical claims. read the letter →

arxiv 1908.05476 v3 pith:QRBDDMYC submitted 2019-08-15 econ.EM

Emmanuel Guerre , Yao Luo This is my paper · ORCID

classification econ.EM MSC 91B2662G05

keywords auctionmodelsunobservedcompetitionnonparametricidentificationdensitydiscontinuitiesendogenousparticipationheterogeneitydiscretemixturefirst-priceauctions

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

The paper tries to establish that, in a symmetric independent-private-value first-price auction, observing only the winning bid—not the number of bidders, not the losing bids—is enough to identify the full model. The reason is that each possible number of active bidders generates its own upper endpoint for the bid support, and the winning-bid density jumps at those endpoints; the locations and sizes of the jumps encode the support and the probabilities of the unobserved competition variable. A second iterative step, built from two equilibrium quantile mappings, then recovers the private-value distribution over its whole support. The same logic extends to endogenous participation through a reserve price or entry cost when participants observe competition, and the paper derives testable restrictions for whether they do. This matters because many real markets—informal procurement, housing 'bidding wars', subsidy competitions—record only the transaction price, and because recorded bid counts can misstate true competition.

What carries the argument

The load-bearing object is the unconditional winning-bid density $g(b)$, which is a finite mixture $g(b)=\sum_{n=\underline n}^{\overline n} p_n\, n\, G_n^{n-1}(b)\, g_n(b)$. Equilibrium bidding makes the conditional support upper bounds $b_n$ strictly increasing in $n$, and Corollary 2.1 shows each conditional density is positive at its upper boundary, $g_n(b_n)=1/((n-1)(\bar v-b_n))$. Hence $g$ has a jump at every $b_n$, with size $\Delta_n=n p_n/((n-1)(\bar v-b_n))$; the locations give the support of $N$, the sizes give the probabilities after $\bar v$ is solved from $\sum p_n=1$, and the iterative quantile mappings $V(\alpha)=B_n(\alpha)+\alpha B_n'(\alpha)/(n-1)$ and $B_n(\alpha)=\frac{n-1}{\alpha^{n-1}}\int_0^\alpha t^{n-2}V(t)\,dt$ extend identification of $V$ from the top quantile interval down to $\alpha=0$.

What would settle it

Simulate the benchmark model with $F$ uniform on $[0,1]$, buyers observing $N$, and $N$ taking values 2 and 4 with equal probability. The winning-bid density has jumps exactly at $b_2$ and $b_4$, so the paper's rule $\bar n = \underline n + \#\{\text{jumps}\} - 1$ returns $\bar n = 3$, and the estimated $p_3$ is positive although $P(N=3)=0$; this violates the true data-generating process and shows that Assumption N's full-support requirement is what carries the jump-counting step.

Watch

Extended reading notes

Core claim

The paper's central claim is Theorem 2.1: under Assumptions N and IPV, if buyers observe the number of active bidders $N$, then the private-value c.d.f. $F(\cdot)$ and the distribution of $N$ are identified from the winning-bid distribution alone. The argument uses the fact that the conditional bid quantile $B_n(\alpha)$ is strictly increasing in $n$, so the support upper bounds $b_n=B_n(1)$ satisfy $\underline v=b_{\underline n}<\cdots<b_{\overline n}<\bar v$, and that the conditional bid density is positive at each upper bound, $g_n(b_n)=1/((n-1)(\bar v-b_n))$. These ingredients force the unconditional winning-bid density to jump at every $b_n$, with jump size $\Delta_n=n p_n/((n-1)(\bar v-b_n))$. Jump locations identify the support of $N$, jump sizes identify the probabilities $p_n$ once the upper bound $\bar v$ is recovered from $\sum p_n=1$, and an iterative quantile argument extends identification of $V(\alpha)=F^{-1}(\alpha)$ from the top bid interval down to $\alpha=0$.

Load-bearing premise

Identification collapses if some intermediate number of bidders never occurs, because that number would produce no density jump, the support count would be wrong, and the iterative expansion down the bid distribution could not begin.

Editorial extensions

If this is right

In markets where only transaction prices are recorded, competition intensity and bidder valuations become estimable without observing the number of bidders or any losing bids.
When active buyers observe the number of competitors, endogenous participation through a reserve price or entry cost does not block identification from winning bids; when they do not observe it, identification requires an instrument or auxiliary observations such as unsold objects.
The derived inequalities on jump sizes and locations provide a direct specification test: winning-bid data violating them cannot be rationalized by the benchmark model.
Empirically, using the recorded number of bids as the competition level can overstate true competition; in the USFS timber data, most three-bid auctions behave as if only two bidders compete, and risk-aversion bounds move toward risk neutrality.

Reading between the lines

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

The same jump-counting logic could identify any finite mixture whose components have strictly ordered support endpoints and positive densities at those endpoints, so the auction setting is one instance of a more general mixture-identification principle.
If an intermediate bidder count has zero probability in the population, the jump count would misreport the support of $N$; a robust alternative would treat detected jumps as a subset of possible boundaries and use the jump-size inequalities to prune impossible configurations.
The USFS finding that a third recorded bid is often non-competitive suggests a testable screen for passive or coordinated bidding in procurement data, since the model itself does not explain why the dominated bid occurs.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

Two things to know. First, the central identification theorem is real: when N has contiguous support {n,...,n}, the winning-bid density discontinuities identify p_n, and the iterative quantile argument recovers F. I checked the algebra in equations (7), (9), (11), (12) and the induction in Lemma 2.3; it hangs together. Second, the paper does not deliver the Bayesian estimation or Shanghai procurement application advertised in its metadata; the full text is a USFS timber illustration with small samples and no standard errors. That mismatch needs fixing.

What is new: the jump-size formula (9) is a genuinely new way to identify the mixture weights from a single bid, and the mix of support-location and jump-size information is clever. The extensions to reserve price and entry cost, and the nonidentification result when buyers don't observe N, are useful and clearly argued.

Soft spots. The load-bearing assumption is Assumption N: every integer in [n,n] must have positive probability. If one intermediate N has probability zero, there is no discontinuity at its would-be boundary, so the count of jumps underestimates n-n and the recursion in Lemma 2.3 cannot start. This is not a niche technicality; it's the mechanism of the identification. The paper gives no test for gaps and no relaxation. The theorem is correct as stated, but the scope is narrower than the abstract's general claim. That should be made explicit and ideally relaxed.

The empirical application is illustrative at best. Sub-samples are 44-53 auctions, bandwidths are hand-chosen, and the estimated probabilities in Tables 1-2 have no confidence intervals. The CRRA bounds in Table 3 are built on the same fragile estimates. I wouldn't take any of the applied numbers as evidence about timber auctions.

Also, the omitted proofs of Proposition 3.6 and Lemma 3.2 matter little; those are supporting.

Who is this for? Auction theorists and empirical IO researchers working on unobserved competition. The identification idea is worth engaging with seriously. A referee should spend time on Section 2 and ask for a statement of the support assumption's bite, plus a clean version with a coherent abstract. I would send it to peer review.

Recommendation: engage with it. It deserves a serious referee, but only after the abstract/body mismatch is corrected.

Referee Report

2 major / 4 minor

Summary. The paper studies nonparametric identification of first-price auction primitives when the analyst observes only winning bids and the number of active bidders is unobserved. Under the maintained assumptions that N is exogenous, independent of private values, has support equal to a full integer interval, and is observed by the bidders, it shows that discontinuities in the winning bid density at conditional bid upper bounds identify the support and probabilities of N, and that an iterative use of the equilibrium bid-quantile and value-quantile mappings identifies the private value distribution. The framework is then extended to endogenous participation through a reserve price or an entry cost, with and without bidder knowledge of participation, yielding testable restrictions for information and entry models. An empirical illustration using USFS timber auctions concludes that many reported three-bid auctions have only two competitive bidders and that risk-aversion estimates are sensitive to this form of unobserved competition.

Significance. If the identification result holds, it is a substantial contribution: the winning bid alone, through the location and size of density discontinuities, identifies both the distribution of competition and the private value distribution without instruments or multiple bids. The proofs are coherent and carefully structured; the jump formula (9) is correctly derived from the equilibrium conditions, and the iterative identification argument in Lemma 2.3 is rigorous. The extension to endogenous participation with testable restrictions for whether bidders observe competition is valuable and connects cleanly to the discrete-mixture literature. The empirical application is suggestive but not decisive because of the very small subsamples and the heuristic nature of the discontinuity-detection algorithm.

major comments (2)

[Section 2.1, Lemma 2.2, Theorem 2.1] Assumption N requires P(N=k)>0 for every integer k between n and n. The identification of n as n plus the number of density discontinuities minus one, and the formulas for p_n in equations (11)-(12), rely on a jump appearing at every intermediate boundary b_k. If P(N=k)=0 for some intermediate k, no jump occurs at b_k, so the count understates n and the probabilities are assigned to an incorrect support. The iterative construction in Lemma 2.3 also fails, because Step 1 uses the interval [b_{n-1}, b_n] starting from the top boundary of the next-lower component. Since n and n are not directly observed, the data cannot distinguish a genuine support gap from a zero-probability support point, and the paper provides no test or relaxation of this condition. The abstract's claim that the resulting discontinuities identify the distribution of N therefore overstates robustness; the identified object is the distribution of N conditional on a full-interval support assumption.
[Section 4, Tables 1-3, Appendix] The empirical conclusions rest on very small subsamples (45, 53, and 44 auctions), a k-NN discontinuity-detection algorithm with hand-set bandwidths h0=0.2 and h1=0.5 and epsilon=0.01, and no reported standard errors or sensitivity analysis. The point estimates in Table 1 (e.g., 0.90-0.95 for unobserved competition versus 0.45-0.53 from observed bids) and the CRRA lower bound of 0.9 in Section 4.2 are therefore not robustly supported; the text itself acknowledges that the variance is likely to be very high. The paper should provide bootstrap confidence intervals and a sensitivity analysis over (h0, h1, epsilon, K, M) before drawing conclusions about non-competitive bidding or risk aversion.

minor comments (4)

[Section 2.5.1] The text says 'Expression (9) allows for the identification of p1 = p2 = 1/2', but p_1 is undefined because the support of N starts at n=2; the intended statements concern p_2 and p_3.
[Abstract and Section 4] The abstract included at the top of the submission mentions a parametric Bayesian estimation procedure and an application to Shanghai Government IT procurements, but the full text proposes no Bayesian procedure and applies the methods to USFS timber auctions. These statements should be reconciled.
[Section 4.2] The sentence 'we obtain the best lower bound for the CRRA coefficient θ 0.9' should read 'a lower bound of θ = 0.9' or 'the best lower bound is 0.9'.
[Appendix] The displayed formula for the critical value c(epsilon;h0) is difficult to parse because of the mixed square-root and logarithmic terms; please format it more clearly and define each term.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: identification from winning-bid density jumps is self-contained; self-citations are standard and non-load-bearing.

full rationale

The derivation chain is self-contained. Section 2.2 derives the two equilibrium mappings, Eqs. (2) and (4), from the bidders' first-order conditions rather than importing them as black boxes. Section 2.4 derives the jump formula (9) from the mixture expression (10) and Corollary 2.1, and Section 2.5 solves for the upper value bound v and the probabilities p_n from the normalization sum p_n = 1 in Eqs. (11)-(12). Theorem 2.1 then combines Lemmas 2.2 and 2.3, with Lemma 2.3's recursion proven in Section 6.1. No target quantity, namely F(.) or the distribution of N, is fed back into the identifying equations. The authors' own prior work (Guerre-Perrigne-Vuong 2000, 2009; Liu-Luo 2016; Guerre-Gimenes 2019) is cited for the quantile framework, but the framework is re-derived in the text, so these self-citations are not load-bearing. The empirical risk-aversion bounds in Section 4.2 use the same jump equations to constrain theta, which is a valid inequality bound rather than a fitted parameter relabeled as a prediction. Footnote 1 explicitly notes an omitted proof, and the front abstract advertises a Bayesian estimation procedure absent from the body; these are completeness issues, not circularity. Assumption N's contiguous-support condition is substantive and may limit robustness, but an assumption's strength is not circularity. No equation in the paper reduces by construction to its own inputs.

Assumptions & free parameters 5 free parameters · 6 assumptions · 0 invented entities

The identification proofs are derived from the standard IPV equilibrium plus the stated support and observability assumptions; no free parameters enter the theory. The free parameters listed are empirical tuning choices in the application that materially affect the p_n and theta estimates. No new theoretical entities are introduced; the latent N is a primitive of the model, and the density discontinuity is a property of the equilibrium mixture rather than an added construct.

free parameters (5)

h0 = 0.2
Bandwidth for the initial k-NN density discontinuity estimates in the appendix; it controls the size of the intervals around each potential jump and affects which discontinuities are detected (Section A and Section 4.1.2).
h1 = 0.5
Bandwidth for the final discontinuous density estimator; combined with h0 it determines the estimated jump sizes and hence the probabilities p_n (Section A).
epsilon = 0.01
Tuning parameter in the critical value c(epsilon;h0); it sets the threshold that decides whether a tentative jump is declared significant (Section A).
K = 2
Number of quantile cells in each covariate dimension used to estimate the lower bound v(X) before applying the Hill estimator (Section 4.1.1).
M = 21 to 65
Number of upper-tail observations in the Hill estimator (28); the estimate of n varies with M, and the reported n=2 is only the most common value over this range (Section 4.1.1, Figure 4).

assumptions (6)

standard math Symmetric IPV first-price auction best responses are strictly increasing, continuously differentiable, and satisfy the differential equation V(alpha)=B_n(alpha)+alpha B_n'(alpha)/(n-1).
Used to derive the quantile mappings (2)-(4) and the boundary density formulas in Corollary 2.1; this is a standard result from Maskin and Riley (1984).
domain assumption Assumption IPV: F has compact support [v,v] with density f continuous and strictly positive on [v,v].
Ensures g_n(v)>0 and g_n(b_n)>0, which create the density jumps used for identification; invoked in Lemma 2.1 and throughout Section 2.
domain assumption Assumption N: N has support {n,...,n} with p_n>0 for all n=n,...,n.
Needed to count discontinuities to identify n and to make the iterative extrapolation in Lemma 2.3 well-defined; a gap in the support would break the recursion.
domain assumption Buyers observe the number of active bidders N before bidding in the benchmark model.
Observability of N is the source of the support variation and density discontinuities; the paper's central theorem states identification under this assumption, and Propositions 3.2-3.3 show identification fails without it.
domain assumption In endogenous participation models (Sections 3.1 and 3.2), the number of active bidders is binomial with parameter n and one minus the screening probability, following a threshold rule for reserve price or entry cost.
Used to derive the binomial mixture structure in Propositions 3.1, 3.2, 3.5, and 3.6; relies on the equilibrium characterization of Gentry and Li (2014).
domain assumption The instrument z affects only the reserve price R(z) or entry cost c(z), while the private value distribution and n are independent of z (Assumptions R and E).
This exclusion restriction is required for the instrumental-variable identification results in Propositions 3.4 and 3.7.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Nonparametric Identification of First-Price Auction with Unobserved Competition: A Density Discontinuity Framework." pith.science (2026). https://pith.science/paper/QRBDDMYC

@misc{pith2026190805476,
  author       = {Pith},
  title        = {Pith review of: Nonparametric Identification of First-Price Auction with Unobserved Competition: A Density Discontinuity Framework},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QRBDDMYC}},
  note         = {Machine review of arXiv:1908.05476}
}

abstract

We consider nonparametric identification of independent private value first-price auction models, in which the analyst only observes winning bids. Our benchmark model assumes an exogenous number of bidders $N$. We show that, if the bidders observe $N$, the resulting discontinuities in the winning bid density can be used to identify the distribution of $N$. The private value distribution can be nonparametrically identified in a second step. This extends, under testable identification conditions, to the case where $N$ is a number of potential buyers, who bid with some unknown probability. Identification also holds in presence of additive unobserved heterogeneity drawn from some parametric distributions. A parametric Bayesian estimation procedure is proposed. An application to Shanghai Government IT procurements finds that the imposed three bidders participation rule is not effective. This generates loss in the range of as large as $10\%$ of the appraisal budget for small IT contracts.

Figures

Figures reproduced from arXiv: 1908.05476 by the authors.

**Figure 2.** Winning bid distribution (V (α) = √ α and P(N = 2) = P(N = 3) = 1/2) first-price auctions that does not hold in ascending or second-price ones.1 Lemma 2.1-(ii) focuses on the winning bid p.d.f. discontinuities and its jumps. Lemma 2.1 Suppose Assumptions N and IPV hold. Then, all of the following hold. (i). For all α in (0, 1], Bn (α) < · · · < Bn (α) < V (α) with Bn (0) = V (0) for all n. In particular, for bn = Bn… view at source ↗

**Figure 3.** Iterative identification of Bn(α) and V (α) from G(·), as in [PITH_FULL_IMAGE:figures/full_fig_p019_3.png] view at source ↗

Figures from the paper (3 more)

**Figure 4.** Figure 4: Hill estimators ne: two-bid and three-bid auctions constant. To address this issue, we first estimate the lower bound v(X) and use it to normalize the winning bids. Applying to the normalized sample of winning bid, the Hill estimator is consistent. Note that auction th…

**Figure 5.** Figure 5: Conditional density estimation for ’Low’, ’Medium’ and ’High’ subsamples [PITH_FULL_IMAGE:figures/full_fig_p048_5.png]

**Figure 6.** Figure 6: Hill estimators ne: three-bid auctions of having two bidders contributing to the winning bid is lower but still higher than 0.5. However, the variance of these estimations are likely to be very high due to their nonparametric nature and small sample sizes. Therefore, t…

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

Does the ratio of Laplace transforms of powers of a function identify the function?
math.PR 2019-09 conditional novelty 7.0 of 10

The ratio of Laplace transforms of powers of a function is injective on monotone right-analytic functions, and this yields an identification result for auction models with unobserved heterogeneity.

Reference graph

Works this paper leans on

2 extracted references · 2 canonical work pages · cited by 1 Pith paper

[1]

Abrantes-Metz, R. & P. Bajari (2009). Screens for conspiracies and their mul- tiple applications.Antitrust 24, 66–71. An, Y., Y. Hu & M. Shum (2010).Estimatingﬁrst-priceauctionswithan unknown number of bidders: a misclassiﬁcation approach.Journal of Econometrics157, 328–

work page 2009
[341]

Regression Discontinuity Designs

Athey, S. & P.A. Haile (2007). Nonparametric approaches to auctions.Handbook of Econometrics6A, 3847–3965. Bulow, J. & and P. Klemperer (1996). Auctions versus negotiations.American Economic Review86, 180–194. Campo, S., E. Guerre, I. Perrigne & Q. Vuong (2011). Semiparametric esti- mation of ﬁrst-price auctions with risk-averse bidders.Review of Economic...

work page 2007

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.

[1] [1]

Abrantes-Metz, R. & P. Bajari (2009). Screens for conspiracies and their mul- tiple applications.Antitrust 24, 66–71. An, Y., Y. Hu & M. Shum (2010).Estimatingﬁrst-priceauctionswithan unknown number of bidders: a misclassiﬁcation approach.Journal of Econometrics157, 328–

work page 2009

[2] [341]

Regression Discontinuity Designs

Athey, S. & P.A. Haile (2007). Nonparametric approaches to auctions.Handbook of Econometrics6A, 3847–3965. Bulow, J. & and P. Klemperer (1996). Auctions versus negotiations.American Economic Review86, 180–194. Campo, S., E. Guerre, I. Perrigne & Q. Vuong (2011). Semiparametric esti- mation of ﬁrst-price auctions with risk-averse bidders.Review of Economic...

work page 2007