Sample Complexity of Posted Pricing for a Single Item

Selling a single item to $n$ self-interested buyers is a fundamental problem in economics, where the two objectives typically considered are welfare maximization and revenue maximization. Since the optimal mechanisms are often impractical and do not work for sequential buyers, posted pricing mechanisms, where fixed prices are set for the item for different buyers, have emerged as a practical and effective alternative. This paper investigates how many samples are needed from buyers' value distributions to find near-optimal posted prices, considering both independent and correlated buyer distributions, and welfare versus revenue maximization. We obtain matching upper and lower bounds (up to logarithmic factors) on the sample complexity for all these settings.

Paper

Similar papers

Peer review

Reviewer L1MB8/10 · confidence 4/52024-07-10

Summary

This paper studies the sample complexity of posted pricing problems. In such problems, a sequence of $n$ buyers arrive with valuations drawn from fixed distributions, and the goal is to post prices that maximize either the revenue or the welfare. In particular, the item is sold to the first buyer who accepts its proposed price. Notably, in the welfare maximizing case, this is the prophet inequality problem. Given access to the distributions of the buyers’ valuations, it is possible to find the optimal online strategy via dynamic programming. This paper studies this problem when the access to the distribution is given via samples. The authors provide matching sample complexity bounds for the revenue and welfare maximization tasks, and for both the independent and correlated setting.

Strengths

- Posted pricing is an important problem, vastly studied in both the EC and NeurIPS/ICML communities, with countless applications - Studying its sample complexity is well-motivated - the paper significantly improves on the state-of-the-art [Guo et al. COLT’21], removing the dependence on $n$ - The paper is well-written

Weaknesses

The paper is a strong NeurIPS submission

Questions

None

Rating

8

Confidence

4

Soundness

4

Presentation

4

Contribution

4

Limitations

No potential negative societal impact of this work

Reviewer U8bp6/10 · confidence 4/52024-07-15

Summary

This paper studies the problem of learning approximately revenue (welfare) optimal posted-prices single-item auctions from samples. The authors consider the setting where the valuation distributions of the agents are independent and the setting where the distributions are correlated. For both settings and both objectives they derive (almost) matching lower and upper bounds on the sample complexity of the learning task. All the results restrict the valuation distributions to be bounded in $[0,1]$ and consider additive approximations of the optimal posted-prices auctions.

Strengths

-The problem of learning posted-prices auctions is of interest to the community. -The results derived are almost tight. -The technical results require some work (even though, conceptually, the approach relies heavily on prior work).

Weaknesses

-The conceptual ideas of the work feel a bit incremental, given the long line of work in the setting of learning optimal auctions. -The results are limited to a very simple setting of single item auctions. -The running time of the algorithm for correlated distributions is exponential. -The presentation could be improved a bit. For example, I found the use of the word "policy" a bit confusing. Usually, this would imply some dynamic decision process, whereas here there is only a single-shot setting. It would be good to use different symbols for the revenue/welfare objectives. Line 61: instead of $V$ use something that denotes a set of samples. Line 104: what is the value-to-go? The proof sketch of Theorem 5 is a bit confusing. Lines 157-158 "single item" is repeated. The notation $r_i, \hat{r}_i, r^*_i$ in Section 2.1 is a bit confusing. In Section 2.2 you can add a quick comment that choosing a price randomly will not help. In Line 23 "know" appears twice. In Line 173 drop "a". It might be useful to include the definition of the pseudo-dimension at least in the appendix.

Questions

-What settings other than single-item auctions can the results be extended to? -For other types of distributions such as regular/MHR, can you derive optimal bounds? -For the bound on the pseudo-dimension, can't you use a direct argument like Morgenstern and Roughgarden (2016) for $t$-level auctions? I am not sure I understand the need to go through the dual class. -Can you comment a bit on the differences of your results/approach and [GHTZ'21]?

Rating

6

Confidence

4

Soundness

4

Presentation

3

Contribution

2

Limitations

Addressed.

Reviewer gD9M6/10 · confidence 4/52024-07-15

Summary

The paper studies a setup of a series posted-price auctions that tries to sell a single item up to the first price acceptance. The authors seeking for the number of samples from buyer value distributions (sample complexity) to be able to setup near-optimal posted-price in the auctions. They consider two objectives (social welfare and revenue maximization) and different dependencies of buyer distributions (independent and correlated). The paper contributes proven both upper and lower on the sample complexity for the all four settings: the bounds are close up to logarithmic factors.

Strengths

- Clear math contribution - Well-structured proofs

Weaknesses

- No conclusions: the paper does not have a concluding discussion - No experimental evidence: asymptotic bounds are clear measures for growing (shrinking) parameters, but, for practice, constant factors are more important. It would be nice to understand how they work in practical cases.

Questions

- What are the factors (multiplicative constants) in the upper and lower bounds in Theorems 1-4? What is behind O(...) and \Omega(...)?

Rating

6

Confidence

4

Soundness

4

Presentation

2

Contribution

3

Limitations

the authors adequately addressed the limitations

Reviewer mFS47/10 · confidence 4/52024-07-15

Summary

The paper considers the problem of statistically estimating the optimal posted price mechanism for selling one item to multiple buyers. Here, it is assumed that the buyers appear sequentially, a price is presented to each of them and the sale is completed if the price posted to them is lower than their valuation. The mechanism aims to optimize one of two targets -- the welfare, defined here as the value of the bidder who wins the auction and revenue of the auctioneer which is the price paid by the winning bidder. It is assumed that the valuations of the bidders are drawn from a distribution and the paper aims to analyze the sample complexity of determining near-optimal posted prices for each of these settings where it is assumed that the learner receives iid samples from the bid distributions. Depending on the degree of dependence between the valuations of the bidders, the paper considers two settings -- firstly, the independent setting where the values are determined independently for each bidder and the dependent setting where the valuations are correlated across the bidders. In the independent setting, it is shown that there exists a separation between the sample complexities of welfare and revenue maximization. They show that essentially $1 / \epsilon^2$ samples suffice for welfare maximization whereas revenue maximization necessarily \emph{requires} $\Omega (n / \epsilon^2)$ samples where $n$ is the number of bidders and $\epsilon$ an additive error parameter. In this case, they obtain near-optimal characterization of the sample complexity while the upper and lower bounds from prior work are $n / \epsilon^2$ and $1 / \epsilon^2$ respectively. In contrast, when the bids are correlated, the paper shows that $\Omega (n / \epsilon^2)$ is \emph{required} for both revenue and welfare maximization. To partially address this pessimistic bound, the paper also considers the setting where the class of possible posted prices is restricted. The particular restriction considered in the paper is by restricting the change points of the price schedule; that is, the price is only allowed to change at $k$ points in the sequence of prices. Here, it is shown that sample complexities independent of $n$ are obtainable and instead only depend on $k$. The proof of the upper bound independent of $n$ for welfare maximization in the independent valuation setting is quite interesting. Essentially, the paper analyzes the dynamic programming algorithm for welfare maximization. Observing that the (welfare of the) posted prices have a closed-form solution in terms of the welfare of the next step, the paper proves that the sub-optimality of the posted prices may be bounded by a sum of error terms which satisfy a backwards martingale structure. The use of standard Martingale concentration bounds then yield the required bound. This proof is elegant and appears to be novel. Unfortunately, such martingale structure does not hold in the presence of dependencies between the valuations. Instead, they adopt a classical uniform convergence based approach and show that the statistical complexity may be controlled by the pseudo-dimension of the policy class. Overall, this is a nice paper that obtains nearly optimal statistical characterizations in several fundamental settings, making several interesting contributions. The proofs of the results, the welfare maximization for independent valuations in particular, are elegant and well-presented. My main concerns are with the assumptions underlying the learning problem. The paper assumes that one obtains independent samples from the \emph{valuations} of the bidders. It is not clear how such samples are obtained -- it is not clear how one may obtain such samples in practice. More exposition on this point would be helpful. Furthermore, it would be helpful if the authors could comment on alternative settings circumventing the $\Omega (n / \epsilon^2)$ sample complexities incurred by the paper -- perhaps, by bounding the degree of dependence between the bidders?

Strengths

See main review

Weaknesses

See main review

Questions

See main review

Rating

7

Confidence

4

Soundness

3

Presentation

3

Contribution

3

Limitations

See main review

Reviewer mFS42024-08-08

Thank you for the response! It would be great if the authors could include the above discussion in subsequent versions of the paper. I will retain my current evaluation.

Reviewer U8bp2024-08-09

I would like to thank the authors for their thorough response. After reading their comments and the rest of the reviews, I've decided to increase my score to 6.

Program Chairsdecision2024-09-25

Decision

Accept (spotlight)

© 2026 NYSGPT2525 LLC