pith. sign in

arxiv: 2605.16064 · v2 · pith:MDN3OWE6new · submitted 2026-05-15 · 💻 cs.GT · cs.AI· econ.TH

Misspecified Estimate-then-Optimize Leads to Supra-Competitive Prices

Jackie Baek , Vivek F. Farias , Farrell Wu This is my paper

Pith reviewed 2026-06-30 19:13 UTC · model grok-4.3

classification 💻 cs.GT cs.AIecon.TH
keywords algorithmic pricingmisspecified demandsupra-competitive pricesNash equilibriumestimate-then-optimizefluid limit analysiscollusive outcomes
0
0 comments X

The pith

Misspecified estimate-then-optimize pricing leads to supra-competitive prices above the Nash equilibrium when firms explore similar price ranges.

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

The paper examines firms that repeatedly fit a demand model to their own price and sales history, omitting competitors' prices, then set the myopic profit-maximizing price. It shows that when this process begins with an exploration phase of independent random prices, the resulting dynamics can drive prices above the competitive Nash level. The analysis identifies the condition under which this occurs: initial explorations must lie in similar ranges on the same side of the Nash price. A fluid-limit ordinary differential equation tracks how prices evolve under repeated misspecification. Simulations calibrated to a real rental market show the same pattern persists with finite horizons, heterogeneous products, and nonlinear demand.

Core claim

When firms apply a myopic estimate-then-optimize rule with a demand model that omits competitors' prices, and the system is initialized by independent random price explorations in similar ranges on the same side of the Nash price, the prices converge to levels above the Nash equilibrium, potentially reaching monopoly prices under symmetric exploration, as characterized by the fluid-limit ODE analysis.

What carries the argument

The fluid-limit ordinary differential equation that governs the price trajectory under repeated misspecified estimation and optimization.

If this is right

  • Supra-competitive prices arise when firms initially explore within similar price ranges on the same side of the Nash price.
  • Prices can reach monopoly levels under symmetric exploration.
  • Supra-competitive outcomes arise robustly in simulations beyond theoretical assumptions, including under finite horizons, heterogeneous products, and nonlinear logit demand.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the exploration phase covers ranges on opposite sides of the Nash price, prices may instead converge to the Nash level.
  • Markets where firms rely on own-history data without competitor variables could sustain elevated prices even without explicit coordination.
  • Adding competitor prices to the fitted demand model would likely eliminate the supra-competitive convergence path shown in the fluid limit.

Load-bearing premise

The demand model fitted by each firm omits competitors' prices, and the outcome depends on the specific ranges of the initial independent random price explorations.

What would settle it

Running the pricing dynamics with initial explorations on opposite sides of the Nash price and checking whether prices converge to the Nash equilibrium rather than above it.

Figures

Figures reproduced from arXiv: 2605.16064 by Farrell Wu, Jackie Baek, Vivek F. Farias.

Figure 1
Figure 1. Figure 1: Left: the shaded regions depict the best-response cones in the (µ1, µ2) plane, where µi is firm i’s average exploration price. We show that the terminal prices are supra-competitive whenever (µ1, µ2) lies in the shaded region. The angle θ depends on the demand parameters but is always at least 45◦ , so the cones cover more than one quarter of the feasible exploration-mean space. Right: Final price under sy… view at source ↗
Figure 2
Figure 2. Figure 2: Best-response cones in the duopoly case, shown in the [PITH_FULL_IMAGE:figures/full_fig_p012_2.png] view at source ↗
Figure 2
Figure 2. Figure 2: Best-response cones in the duopoly case, shown in the [PITH_FULL_IMAGE:figures/full_fig_p013_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: ODE-implied terminal price P ODE 1 (α; µ, σ2 expI2×2) in the duopoly case. Each panel fixes (α, Σexp) and varies the exploration means (µ1, µ2). White marks the Nash price p NE = 2/3, while red and blue indicate terminal prices above and below Nash. Thin lines mark the best-response boundaries defining the two cones. Horizons sharpen cone-like regions. As the horizon α increases (left to right), the heatma… view at source ↗
Figure 3
Figure 3. Figure 3: ODE-implied terminal price P ODE 1 (τ ; µ, σ2 expI2×2) in the duopoly case. Each panel fixes (τ, Σexp = σ 2 expI2×2) and varies the exploration means (µ1, µ2). White marks the Nash price p NE = 2/3, while red and blue indicate terminal prices above and below Nash. Thin lines mark the best￾response boundaries defining the two cones. Horizons sharpen cone-like regions. As the horizon τ increases (left to rig… view at source ↗
Figure 4
Figure 4. Figure 4: ODE and stochastic mean terminal-price heatmaps. The left panel is the deterministic [PITH_FULL_IMAGE:figures/full_fig_p015_4.png] view at source ↗
Figure 4
Figure 4. Figure 4: ODE and stochastic mean terminal-price heatmaps. The left panel is the deterministic [PITH_FULL_IMAGE:figures/full_fig_p016_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: ODE map and terminal-price histograms from stochastic simulations at [PITH_FULL_IMAGE:figures/full_fig_p016_5.png] view at source ↗
Figure 5
Figure 5. Figure 5: ODE map and terminal-price histograms from stochastic simulations at [PITH_FULL_IMAGE:figures/full_fig_p017_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Terminal rent changes relative to Nash as the exploration mean multiplier [PITH_FULL_IMAGE:figures/full_fig_p019_6.png] view at source ↗
Figure 6
Figure 6. Figure 6: Terminal rent changes relative to Nash as the exploration mean multiplier [PITH_FULL_IMAGE:figures/full_fig_p020_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Finite-time dynamics of terminal rent changes relative to Nash. We fix [PITH_FULL_IMAGE:figures/full_fig_p020_7.png] view at source ↗
Figure 7
Figure 7. Figure 7: Finite-time dynamics of terminal rent changes relative to Nash. We fix [PITH_FULL_IMAGE:figures/full_fig_p021_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: ODE-implied mean terminal price under interval sampling. For each boundary pair [PITH_FULL_IMAGE:figures/full_fig_p042_8.png] view at source ↗
Figure 8
Figure 8. Figure 8: ODE-implied mean terminal price under interval sampling. For each boundary pair [PITH_FULL_IMAGE:figures/full_fig_p034_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Mean terminal price under center–dispersion sampling. For each anchor price [PITH_FULL_IMAGE:figures/full_fig_p043_9.png] view at source ↗
Figure 9
Figure 9. Figure 9: Mean terminal price under center–dispersion sampling. For each anchor price [PITH_FULL_IMAGE:figures/full_fig_p035_9.png] view at source ↗
read the original abstract

We study whether simple algorithmic pricing systems can systematically produce collusive-like prices in multi-firm markets. We consider firms that price using a myopic estimate-then-optimize rule: each repeatedly fits a demand model to its own price and sales history and sets the price that maximizes estimated profit. This demand model is misspecified, omitting competitors' prices. We analyze the dynamics of this rule when it is initialized by an exploration phase of independent random prices. We characterize when this pipeline converges to supra-competitive prices above the Nash equilibrium, via a fluid-limit ordinary differential equation analysis. We show that supra-competitive prices arise when firms initially explore within similar price ranges on the same side of the Nash price. Moreover, prices can be substantially above the Nash price; we show that prices can reach monopoly levels under symmetric exploration. Simulations calibrated to a real multifamily rental market confirm that supra-competitive outcomes arise robustly beyond our theoretical assumptions, including under finite horizons, heterogeneous products, and nonlinear logit demand.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper studies firms that repeatedly fit a misspecified demand model (omitting competitors' prices) to their own price-sales history and myopically optimize estimated profit. Initialized by an independent random-price exploration phase, the dynamics are analyzed via a fluid-limit ODE whose attractors determine long-run prices. The central claim is that supra-competitive prices (including monopoly levels under symmetric exploration) arise when firms explore in similar ranges on the same side of the Nash price; calibrated simulations are said to confirm robustness beyond the ODE assumptions.

Significance. If the ODE characterization and its simulation validation hold, the work identifies a concrete, non-collusive mechanism by which simple algorithmic pricing can produce supra-competitive outcomes in oligopoly. The fluid-limit approach ties outcomes to initial exploration distributions in a falsifiable way, and the use of real-market calibration strengthens applicability. This is relevant to algorithmic collusion debates and market-design questions.

major comments (2)
  1. [Fluid-limit ODE analysis] Fluid-limit ODE section: the claim that the ODE attractors characterize the discrete stochastic process rests on the approximation that estimation noise vanishes and new observations track the continuous trajectory. Because prices are chosen endogenously from the current estimate, the data-generating process is not exogenous; it is unclear whether finite-sample variance or path dependence can cause escape from the predicted basins. The abstract invokes simulations for robustness, but this does not substitute for a direct argument or numerical check that the discrete algorithm converges to the ODE equilibria rather than other limits.
  2. [Simulation section] Simulation calibration and validation: the abstract states that simulations confirm supra-competitive outcomes under finite horizons, heterogeneous products, and logit demand, yet no information is given on the number of Monte Carlo replications, how initial exploration ranges are sampled in finite samples, or whether the realized long-run prices match the specific ODE attractors (as opposed to merely exceeding Nash). Without these details the simulations cannot be assessed as confirming the ODE characterization.
minor comments (1)
  1. [Introduction] The introduction could more explicitly contrast the misspecification (omission of rivals' prices) with standard Bertrand-Nash assumptions to clarify the source of the bias.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive comments. We address each major comment below and describe the revisions that will be made to strengthen the manuscript.

read point-by-point responses

  1. Referee: [Fluid-limit ODE analysis] Fluid-limit ODE section: the claim that the ODE attractors characterize the discrete stochastic process rests on the approximation that estimation noise vanishes and new observations track the continuous trajectory. Because prices are chosen endogenously from the current estimate, the data-generating process is not exogenous; it is unclear whether finite-sample variance or path dependence can cause escape from the predicted basins. The abstract invokes simulations for robustness, but this does not substitute for a direct argument or numerical check that the discrete algorithm converges to the ODE equilibria rather than other limits.

    Authors: We agree that the link between the fluid-limit ODE and the discrete process merits a more explicit justification, especially given the endogenous data generation. In the revision we will add a dedicated subsection that invokes standard results from stochastic approximation theory to delineate the conditions under which the ODE attractors govern the long-run behavior. We will also include new numerical experiments that run the exact discrete algorithm alongside the ODE trajectory and report the frequency with which the discrete process reaches the predicted basins rather than escaping. revision: yes

  2. Referee: [Simulation section] Simulation calibration and validation: the abstract states that simulations confirm supra-competitive outcomes under finite horizons, heterogeneous products, and logit demand, yet no information is given on the number of Monte Carlo replications, how initial exploration ranges are sampled in finite samples, or whether the realized long-run prices match the specific ODE attractors (as opposed to merely exceeding Nash). Without these details the simulations cannot be assessed as confirming the ODE characterization.

    Authors: We accept that the simulation section lacks the necessary methodological detail. The revised manuscript will report the exact number of Monte Carlo replications, describe the precise sampling distribution used for initial exploration ranges, and add quantitative comparisons (e.g., tables or histograms) showing that the empirical terminal prices align with the specific ODE equilibria rather than merely lying above the Nash price. revision: yes

Circularity Check

0 steps flagged

No circularity: fluid-limit ODE derives from stated misspecified dynamics

full rationale

The paper's central result characterizes convergence of the estimate-then-optimize process to supra-competitive prices via a fluid-limit ODE whose trajectories and attractors are obtained directly from the myopic pricing rule applied to the misspecified demand model and the initial exploration distribution. This is a standard mean-field approximation of the discrete stochastic updates and does not reduce any claimed prediction to a fitted parameter or self-citation by construction. The ODE is not obtained by renaming a known result or smuggling an ansatz; the attractors emerge from solving the derived differential equation under the given misspecification. No load-bearing step relies on prior self-citations whose content is unverified or tautological. The derivation remains self-contained against the paper's own model equations.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The claim rests on the domain assumption of demand misspecification and the modeling choice of fluid-limit approximation for discrete pricing updates.

axioms (1)
  • domain assumption Demand model omits competitors' prices
    This misspecification is the driver of the supra-competitive convergence.

pith-pipeline@v0.9.1-grok · 5705 in / 1090 out tokens · 43811 ms · 2026-06-30T19:13:13.528732+00:00 · methodology

Review history (2 revisions) →

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 1 Pith paper

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

  1. Should Demand Models Incorporate Competitor Prices? Oblivious Learning and Algorithmic Collusion

    cs.GT 2026-06 unverdicted novelty 7.0

    In stylized competitive markets with noisy demand and iterated least squares learning, oblivious demand models yield transient collusive patterns that dissipate under sufficient exploration, informed sellers strictly ...

Reference graph

Works this paper leans on

37 extracted references · 27 canonical work pages · cited by 1 Pith paper · 1 internal anchor

  1. [1]

    Abada and X

    I. Abada and X. Lambin. Artificial intelligence: Can seemingly collusive outcomes be avoided? Management Science, 69 0 (9): 0 5042--5065, 2023. doi:10.1287/mnsc.2022.4623

    work page doi:10.1287/mnsc.2022.4623 2023

  2. [2]

    Abada, X

    I. Abada, X. Lambin, and N. Tchakarov. Collusion by mistake: Does algorithmic sophistication drive supra-competitive profits? European Journal of Operational Research, 318 0 (3): 0 927--953, 2024. doi:10.1016/j.ejor.2024.06.006

    work page doi:10.1016/j.ejor.2024.06.006 2024

  3. [3]

    Aouad and A

    A. Aouad and A. V. den Boer. Algorithmic collusion in assortment games. Available at SSRN 3930364 , 2021. doi:10.2139/ssrn.3930364. URL https://ssrn.com/abstract=3930364

    work page doi:10.2139/ssrn.3930364 2021

  4. [4]

    E. R. Arunachaleswaran, N. Collina, S. Kannan, A. Roth, and J. Ziani. Algorithmic collusion without threats. In R. Meka, editor, 16th Innovations in Theoretical Computer Science Conference ( ITCS 2025) , volume 325 of Leibniz International Proceedings in Informatics ( LIPIcs ) , pages 10:1--10:21, Dagstuhl, Germany, 2025. Schloss Dagstuhl -- Leibniz-Zentr...

    work page doi:10.4230/lipics.itcs.2025.10 2025

  5. [5]

    Asker, C

    J. Asker, C. Fershtman, and A. Pakes. Artificial intelligence, algorithm design, and pricing. AEA Papers and Proceedings , 112: 0 452--456, 2022. doi:10.1257/pandp.20221059

    work page doi:10.1257/pandp.20221059 2022

  6. [6]

    Assad, R

    S. Assad, R. Clark, D. Ershov, and L. Xu. Algorithmic pricing and competition: Empirical evidence from the German retail gasoline market. Journal of Political Economy, 132 0 (3): 0 723--771, 2024. doi:10.1086/726906

    work page doi:10.1086/726906 2024

  7. [7]

    Aviv and A

    Y. Aviv and A. Pazgal. Pricing of short life-cycle products through active learning. Working paper, Olin School of Business, Washington University in St. Louis, 2002

    2002

  8. [8]

    arXiv preprint arXiv:2202.05946 , year=

    M. Banchio and G. Mantegazza. Artificial intelligence and spontaneous collusion. arXiv preprint arXiv :2202.05946 , 2022. URL https://arxiv.org/abs/2202.05946

    work page arXiv 2022

  9. [9]

    Online optimization algorithms in repeated price competition: Equilibrium learning and algorithmic collusion.arXiv preprint arXiv:2412.15707, 2024.https://arxiv.org/abs/2412.15707

    M. Bichler, J. Durmann, and M. Oberlechner. Online optimization algorithms in repeated price competition: Equilibrium learning and algorithmic collusion, 2024. URL https://arxiv.org/abs/2412.15707

    work page arXiv 2024

  10. [10]

    Z. Y. Brown and A. MacKay. Competition in pricing algorithms. American Economic Journal: Microeconomics, 15 0 (2): 0 109--156, 2023. doi:10.1257/mic.20210158

    work page doi:10.1257/mic.20210158 2023

  11. [11]

    Calder-Wang and G

    S. Calder-Wang and G. H. Kim. Algorithmic pricing in multifamily rentals: Efficiency gains or price coordination? Available at SSRN 4403058 , 2024. doi:10.2139/ssrn.4403058. URL https://ssrn.com/abstract=4403058

    work page doi:10.2139/ssrn.4403058 2024

  12. [12]

    Calvano, G

    E. Calvano, G. Calzolari, V. Denicol \`o , and S. Pastorello. Artificial intelligence, algorithmic pricing, and collusion. American Economic Review, 110 0 (10): 0 3267--3297, 2020. doi:10.1257/aer.20190623

    work page doi:10.1257/aer.20190623 2020

  13. [13]

    L. Chen, A. Mislove, and C. Wilson. An empirical analysis of algorithmic pricing on Amazon Marketplace . In Proceedings of the 25th International Conference on World Wide Web , pages 1339--1349, 2016. doi:10.1145/2872427.2883089

    work page doi:10.1145/2872427.2883089 2016

  14. [14]

    W. L. Cooper, T. Homem-de-Mello , and A. J. Kleywegt. Learning and pricing with models that do not explicitly incorporate competition. Operations Research, 63 0 (1): 0 86--103, 2015. doi:10.1287/opre.2014.1341

    work page doi:10.1287/opre.2014.1341 2015

  15. [15]

    A. V. den Boer, J. M. Meylahn, and M. P. Schinkel. Artificial collusion: Examining supracompetitive pricing by Q -learning algorithms. Amsterdam Law School Research Paper No. 2022-25; Amsterdam Center for Law & Economics Working Paper No. 2022-06; available at SSRN 4213600, 2024. URL https://ssrn.com/abstract=4213600

    2022

  16. [16]

    Douglas, F

    C. Douglas, F. Provost, and A. Sundararajan. The illusion of collusion. arXiv preprint arXiv :2411.16574 , 2024. URL https://arxiv.org/abs/2411.16574

    work page arXiv 2024

  17. [17]

    V. F. Farias and B. Van Roy. Dynamic pricing with a prior on market response. Operations Research, 58 0 (1): 0 16--29, 2010. doi:10.1287/opre.1090.0729

    work page doi:10.1287/opre.1090.0729 2010

  18. [18]

    S. Fish, Y. A. Gonczarowski, and R. I. Shorrer. Algorithmic collusion by large language models. arXiv preprint arXiv :2404.00806 , 2024. URL https://arxiv.org/abs/2404.00806

    work page arXiv 2024

  19. [19]

    E. J. Green and R. H. Porter. Noncooperative collusion under imperfect price information. Econometrica, 52 0 (1): 0 87--100, 1984. doi:10.2307/1911462

    work page doi:10.2307/1911462 1984

  20. [20]

    K. T. Hansen, K. Misra, and M. M. Pai. Frontiers: Algorithmic collusion: Supra-competitive prices via independent algorithms. Marketing Science, 40 0 (1): 0 1--12, 2021. doi:10.1287/mksc.2020.1276

    work page doi:10.1287/mksc.2020.1276 2021

  21. [21]

    M. Hettich. Algorithmic collusion: Insights from deep learning. Available at SSRN 3785966 , 2021. doi:10.2139/ssrn.3785966. URL https://ssrn.com/abstract=3785966

    work page doi:10.2139/ssrn.3785966 2021

  22. [22]

    Keppo, Y

    J. Keppo, Y. Li, G. Tsoukalas, and N. Yuan. AI pricing, agent heterogeneity, and collusion. Available at SSRN 5386338 , 2025. doi:10.2139/ssrn.5386338. URL https://ssrn.com/abstract=5386338

    work page doi:10.2139/ssrn.5386338 2025

  23. [23]

    A. P. Kirman. Learning by firms about demand conditions. In R. H. Day and T. Groves, editors, Adaptive Economic Models, pages 137--156. Academic Press, New York, 1975

    1975

  24. [24]

    A. P. Kirman. On mistaken beliefs and resultant equilibria. In R. Frydman and E. S. Phelps, editors, Individual Forecasting and Aggregate Outcomes: ``Rational Expectations'' Examined, pages 147--168. Cambridge University Press, Cambridge, 1986

    1986

  25. [25]

    A. P. Kirman. Learning in oligopoly: Theory, simulation, and experimental evidence. In A. P. Kirman and M. Salmon, editors, Learning and Rationality in Economics, pages 127--178. B. Blackwell, Oxford, UK and Cambridge, MA, 1995

    1995

  26. [26]

    T. Klein. Autonomous algorithmic collusion: Q -learning under sequential pricing. The RAND Journal of Economics , 52 0 (3): 0 538--558, 2021. doi:10.1111/1756-2171.12383

    work page doi:10.1111/1756-2171.12383 2021

  27. [27]

    M. A. Lariviere and E. L. Porteus. Stalking information: Bayesian inventory management with unobserved lost sales. Management Science, 45 0 (3): 0 346--363, 1999. doi:10.1287/mnsc.45.3.346

    work page doi:10.1287/mnsc.45.3.346 1999

  28. [28]

    Conjectural Variations in Competitive Dynamic Pricing: A Learning Foundation via Experimentation Design and Feedback Structure

    B. Light and W. Wang. Conjectural variations in competitive dynamic pricing: A learning foundation via experimentation design and feedback structure, 2026. URL https://arxiv.org/abs/2602.12888

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  29. [29]

    Lin and \"O

    M. Lin and \"O . Sar ta c . Competition in pricing algorithms: Stability, exploration, and supracompetitive outcomes. Available at SSRN 5958355, December 2025. URL https://ssrn.com/abstract=5958355

    2025

  30. [30]

    Loots and A

    T. Loots and A. V. den Boer. Data-driven collusion and competition in a pricing duopoly with multinomial logit demand. Production and Operations Management, 32 0 (4): 0 1169--1186, 2023. doi:10.1111/poms.13919

    work page doi:10.1111/poms.13919 2023

  31. [31]

    Maskin and J

    E. Maskin and J. Tirole. A theory of dynamic oligopoly, II : Price competition, kinked demand curves, and edgeworth cycles. Econometrica, 56 0 (3): 0 571--599, 1988. doi:10.2307/1911701

    work page doi:10.2307/1911701 1988

  32. [32]

    J. M. Meylahn and A. V. den Boer. Learning to collude in a pricing duopoly. Manufacturing & Service Operations Management, 24 0 (5): 0 2577--2594, 2022. doi:10.1287/msom.2021.1074

    work page doi:10.1287/msom.2021.1074 2022

  33. [33]

    Robinson

    J. Robinson. An iterative method of solving a game. Annals of Mathematics, 54 0 (2): 0 296--301, 1951

    1951

  34. [34]

    H. A. Simon. Dynamic programming under uncertainty with a quadratic criterion function. Econometrica, 24 0 (1): 0 74--81, 1956

    1956

  35. [35]

    R. S. Sutton and A. G. Barto. Reinforcement Learning: An Introduction. MIT Press, Cambridge, MA, 1998

    1998

  36. [36]

    Z. Yang, X. Lei, and P. Gao. Regulating discriminatory pricing in the presence of tacit collusion. Available at SSRN 4633784 , 2023. doi:10.2139/ssrn.4633784. URL https://ssrn.com/abstract=4633784

    work page doi:10.2139/ssrn.4633784 2023

  37. [37]

    Z. Yang, P. Gao, and Z. Wang. Driven to collusion: Competitive pricing under independent demand model and price imitation, 2026. URL https://ssrn.com/abstract=6197578. Available at SSRN 6197578

    2026