Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Explaining the Law of Supply and Demand via Online Learning
Authors: Stratis Skoulakis
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this work, we provide a mathematical foundation on the law of supply and demand through the lens of online learning. Specifically, we demonstrate that if each seller employs a no-swap regret algorithm to set their individual selling price aiming to maximize its individual revenue the collective pricing dynamics converge to the market-clearing price p . Our findings offer a novel perspective on the law of supply and demand, framing it as the emergent outcome of an adaptive learning processes among sellers. ... In Section 4 we experimentally evaluate both no-regret and no-swap regret algorithms. |
| Researcher Affiliation | Academia | Stratis Skoulakis Aarhus University EMAIL |
| Pseudocode | Yes | Protocol 1: Pricing Game over time At each round t = 1, . . . , T Each seller i [n], (secretly) selects a price pt i Si. Each seller i [n], gets utility Ui(pt i, pt i) (see Definition 3). Each seller i [n], learns pt i and uses this information to select its next price pt+1 i . |
| Open Source Code | Yes | Question: Does the paper provide open access to the data and code, with sufficient instructions to faithfully reproduce the main experimental results, as described in supplemental material? Answer: [Yes] Justification: We have included, the code used for our experimental evaluations along with detailed instructions on how to run it. |
| Open Datasets | No | We first consider the family of instances of the pricing game constructed to establish Theorem 2 (see also Appendix A). This instance is composed by n = 2 sellers and buyers where (s1, s2) = (0, 0) and (b1, b2) = (5, λ). As a result, the highest market clearing price is p = λ. ... Next we consider a more natural set-up with n = 100 sellers and sellers. We consider the linear demand curve D(p) = 20p + 100 for p [0, 5] and three different supply curves, Slinear(p) := 20p, Squad(p) := p2/0.25 and Slinear(p) := p/5, Ssqrt(p) := 100 p p/5 (see Figure 4). |
| Dataset Splits | No | The paper defines parameters for generating instances of pricing games and specific demand/supply curve functions for its experiments, but does not describe any training/test/validation splits for these generated instances. |
| Hardware Specification | Yes | All experiments were conducted in Apple M4 Pro and the Hedge algorithm was run with step-size γ = 0.1. |
| Software Dependencies | No | The paper mentions using the "Hedge [20] no-regret algorithm" and the "no-swap regret algorithm proposed by Blum and Mansour [7]" and specifies a step-size γ = 0.1 for Hedge, but it does not provide specific version numbers for any software libraries, programming languages, or other tools used in the implementation. |
| Experiment Setup | Yes | We consider as the set of prices the [0, 5] interval with 0.2 discretization-there are the following 30 possible prices {0, 0.2, . . . , 4.8, 5}. ... This instance is composed by n = 2 sellers and buyers where (s1, s2) = (0, 0) and (b1, b2) = (5, λ). ... Next we consider a more natural set-up with n = 100 sellers and sellers. We consider the linear demand curve D(p) = 20p + 100 for p [0, 5] and three different supply curves, Slinear(p) := 20p, Squad(p) := p2/0.25 and Slinear(p) := p/5, Ssqrt(p) := 100 p p/5 (see Figure 4). ... All experiments were conducted in Apple M4 Pro and the Hedge algorithm was run with step-size γ = 0.1. |