pith. sign in

arxiv: 2603.22596 · v2 · pith:ASHROOFYnew · submitted 2026-03-23 · 💻 cs.CE · econ.GN· q-fin.EC

ParlayMarket: Automated Market Making for Parlay-style Joint Contracts

Ranvir Rana , Viraj Nadkarni , Niusha Moshrefi , Pramod Viswanath This is my paper

Pith reviewed 2026-05-21 10:21 UTC · model grok-4.3

classification 💻 cs.CE econ.GNq-fin.EC
keywords automated market makingparlay contractsprediction marketsjoint distributionsconvergenceliquidity provisioninformation aggregation
0
0 comments X

The pith

ParlayMarket's automated market maker converges under repeated trading to the best approximation of the true joint distribution within its model class.

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

The paper presents ParlayMarket, an automated market maker that supports parlay-style joint contracts inside one unified liquidity pool while keeping prices coherent across base events and their combinations. Its central result is that repeated trading causes the system's dynamics to reach a unique fixed point that gives the closest match to the actual joint outcome probabilities allowed by the model. Parameter errors stay bounded at this point because each trade update mixes informative signals with noise in a self-balancing way. Pricing mistakes and the market maker's total monetary loss then grow in proportion to those parameter errors, so overall losses remain controlled and increase at most quadratically with the number of base markets. Parlay trades contribute directly to tighter convergence by supplying explicit constraints on joint outcomes that marginal trades alone cannot provide.

Core claim

Under repeated trading, the AMM dynamics converge to a unique fixed point corresponding to the best approximation to the true joint distribution within the model class. Parameter error remains bounded at stationarity due to a balance between signal and noise in trade-induced updates. Pricing error and monetary loss scale with this parameter error, implying that aggregate market-maker loss remains controlled and grows at most quadratically in the number of base markets. Parlay trades play a structural role by improving identifiability of dependence structure and reducing steady-state error relative to markets that use only marginal trades.

What carries the argument

Convergence of the AMM dynamics to a unique fixed point that best approximates the true joint distribution, produced by repeated trade-induced updates that balance signal and noise.

If this is right

  • Parameter error stays bounded once the system reaches stationarity.
  • Both pricing error and monetary loss scale directly with the bounded parameter error.
  • Aggregate market-maker loss grows at most quadratically in the number of base markets.
  • Parlay trades reduce steady-state error by supplying direct constraints on joint outcomes.

Where Pith is reading between the lines

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

  • The same scaling behavior could support reliable liquidity for correlated forecasts in other prediction-market settings.
  • Designers of multi-outcome platforms could test whether adding joint contracts measurably improves dependence recovery compared with separate markets.
  • The quadratic loss bound suggests the approach remains practical even when the number of base events grows moderately.

Load-bearing premise

The trading process and update rules actually drive the system to the claimed unique fixed point without extra assumptions on trader behavior or market completeness.

What would settle it

A controlled simulation or historical replay in which the system either fails to converge to the described fixed point or shows market-maker losses growing faster than quadratically with the number of base markets.

Figures

Figures reproduced from arXiv: 2603.22596 by Niusha Moshrefi, Pramod Viswanath, Ranvir Rana, Viraj Nadkarni.

Figure 1
Figure 1. Figure 1: ParlayMarket enables 2M markets with O(M2 ) capital. Theoretical curves show the asymp￾totic loss scaling implied by the analysis in Section 5. Empirical curves are obtained from controlled simulation of correlated binary markets under the Gaussian score model described later in Sections 3 and 6, and confirm the same polynomial-vs.-exponential separation in practice. 2 Background and Related Work 2.1 Marke… view at source ↗
Figure 2
Figure 2. Figure 2: In request-for-quote mechanism, market makers short on the requested parlay. They have [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: The RFQ trilemma. 4.2 Strawman Independence-Based Parlay AMM The shortcomings of RFQ motivate a natural alternative: replace dealer-mediated quoting with a pooled-liquidity architecture that offers immediate algorithmic quotes for exposed parlays. Hence the natural alternative is to use an automated market makers (AMMs) to parlay markets. A straw￾man approach is to assume independence across base events. I… view at source ↗
Figure 4
Figure 4. Figure 4: The ParlayMarket is the only non-oracle model whose loss decays exponentially with the [PITH_FULL_IMAGE:figures/full_fig_p016_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Price convergence over time (mean ±1σ across 1000 simulations). LMSR market-price MAE vs. true probabilities. 2 3 4 5 6 7 8 9 10 Number of base markets M 10 −5 10 −4 10 −3 10 −2 Mean loss True oracle Ising AMM (ours) Pairwise oracle Gaussian oracle Independent LMSR (a) Mean LMSR loss per round per market ¯ℓM. 2 3 4 5 6 7 8 9 10 Number of base markets M 10 −4 10 −3 10 −2 C V a R95 loss True oracle Ising AMM… view at source ↗
Figure 6
Figure 6. Figure 6: The ParlayMarket fails to perform better than even independent AMMs in the absence of [PITH_FULL_IMAGE:figures/full_fig_p017_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Per-market loss under varying noise-trader fractions ( [PITH_FULL_IMAGE:figures/full_fig_p018_7.png] view at source ↗
read the original abstract

Prediction markets are powerful mechanisms for information aggregation, but existing designs are optimized for single-event contracts. In practice, traders frequently express beliefs about joint outcomes - through parlays in sports, conditional forecasts across related events, or scenario bets in financial markets. Current platforms either prohibit such trades or rely on ad hoc mechanisms that ignore correlation structure, resulting in inefficient prices and fragmented liquidity. We introduce ParlayMarket, the first automated market-making design that supports parlay-style joint contracts within a unified liquidity pool while maintaining coherent pricing across base markets and their combinations. Our main result is a convergence characterization of the resulting system. Under repeated trading, the AMM dynamics converge to a unique fixed point corresponding to the best approximation to the true joint distribution within the model class. We show that (i) parameter error remains bounded at stationarity due to a balance between signal and noise in trade-induced updates, and (ii) pricing error and monetary loss scale with this parameter error, implying that aggregate market-maker loss remains controlled and grows at most quadratically in the number of base markets. These results establish explicit limits on the information-retrieval error achievable through the trading interface. Importantly, parlay trades play a structural role in this convergence: by providing direct constraints on joint outcomes, they improve identifiability of dependence structure and reduce steady-state error relative to markets that rely only on marginal trades. Empirically, we show both in controlled simulations and in replay on historical Kalshi parlay data that this design achieves the intended scaling while remaining effective in realistic market settings.

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 / 2 minor

Summary. The paper introduces ParlayMarket, an automated market maker supporting parlay-style joint contracts in a unified liquidity pool for coherent pricing across base events and combinations. The central claim is a convergence characterization: repeated trading drives the AMM dynamics to a unique fixed point that is the best approximation to the true joint distribution within the model class. Parameter error remains bounded at stationarity from the signal-noise balance in updates; pricing error and monetary loss scale with this error, so aggregate market-maker loss is controlled and grows at most quadratically in the number of base markets. Parlay trades structurally improve identifiability of dependence structure and reduce steady-state error relative to marginal-only markets. The claims are supported by controlled simulations and replay experiments on historical Kalshi parlay data.

Significance. If the convergence and scaling results hold under the paper's trading model, this provides a principled mechanism for handling correlated outcomes in prediction markets without liquidity fragmentation. The explicit error bounds, the quadratic loss scaling, and the demonstrated benefit of parlay trades for identifiability would be a notable contribution to automated market design, offering concrete limits on achievable information retrieval through the trading interface.

major comments (2)
  1. [Convergence characterization] Convergence characterization section: The uniqueness of the fixed point and the subsequent quadratic loss bound rest on the update dynamics behaving as a convergent stochastic approximation whose attractor is exactly the projection onto the model class. The manuscript must supply explicit conditions on the distribution of trader beliefs and the relative frequency of parlay versus marginal trades that guarantee persistent excitation of all joint outcomes; absent these conditions, the effective gradient may lie in a proper subspace, permitting multiple stationary points or convergence to a point whose approximation error is not controlled by the claimed scaling.
  2. [Loss scaling result] Loss scaling result (referenced after the fixed-point claim): The statement that aggregate market-maker loss grows at most quadratically in the number of base markets is load-bearing for the main practical conclusion. The derivation should be stated as a precise theorem, including the dependence of any constants on the model dimension, and it must be shown that the quadratic bound follows directly from the bounded parameter error without additional assumptions that reintroduce linear or higher growth.
minor comments (2)
  1. [Empirical evaluation] In the replay experiments on Kalshi data, specify how the ground-truth joint distribution is constructed or approximated when computing approximation error, as this choice directly affects the measured scaling.
  2. [Notation and model definition] Ensure that the definition of the model class and the divergence minimized by the fixed point are stated with the same notation in both the theoretical analysis and the simulation setup.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed comments. We address each major comment below. We agree that the convergence result benefits from explicit conditions on trader beliefs and trade frequencies, and we will add these to the manuscript. We will also formalize the loss scaling result as a precise theorem.

read point-by-point responses

  1. Referee: [Convergence characterization] Convergence characterization section: The uniqueness of the fixed point and the subsequent quadratic loss bound rest on the update dynamics behaving as a convergent stochastic approximation whose attractor is exactly the projection onto the model class. The manuscript must supply explicit conditions on the distribution of trader beliefs and the relative frequency of parlay versus marginal trades that guarantee persistent excitation of all joint outcomes; absent these conditions, the effective gradient may lie in a proper subspace, permitting multiple stationary points or convergence to a point whose approximation error is not controlled by the claimed scaling.

    Authors: We agree that the uniqueness claim requires explicit conditions to ensure persistent excitation. The manuscript currently relies on an implicit assumption of positive probability for all trade types under repeated trading. We will revise by adding Assumption 3.1: trader beliefs are drawn i.i.d. from a distribution with full support on the joint probability simplex, and the relative frequency of parlay trades is bounded below by a fixed α > 0. Under these conditions, the effective gradient spans the full parameter space, the stochastic approximation converges to the unique KL-projection onto the model class, and the claimed error bounds hold. We will include a short proof sketch in the revised convergence section. revision: yes

  2. Referee: [Loss scaling result] Loss scaling result (referenced after the fixed-point claim): The statement that aggregate market-maker loss grows at most quadratically in the number of base markets is load-bearing for the main practical conclusion. The derivation should be stated as a precise theorem, including the dependence of any constants on the model dimension, and it must be shown that the quadratic bound follows directly from the bounded parameter error without additional assumptions that reintroduce linear or higher growth.

    Authors: We will restate the loss scaling claim as a self-contained theorem (new Theorem 4.2). The theorem states: Let ε denote the stationary parameter error (bounded by the signal-noise balance). Then the aggregate market-maker loss L satisfies L ≤ C(d) ε², where the dimension-dependent constant satisfies C(d) ≤ K d² for a universal K independent of d. The proof shows that (i) each parlay price deviation is Lipschitz continuous in the parameter vector with a Lipschitz constant independent of d, and (ii) the number of relevant parlay combinations that contribute to loss is O(d²) due to the low-order interaction structure of the model class. The quadratic dependence on ε follows directly from the quadratic nature of the loss in price deviations; no additional assumptions are introduced that would produce linear or higher growth in d. revision: yes

Circularity Check

0 steps flagged

No detectable circularity in provided derivation outline

full rationale

The abstract describes a convergence characterization in which repeated trading drives AMM parameters to a unique fixed point that is the best approximation to the true joint distribution within the model class, with subsequent bounds on parameter error, pricing error, and quadratic loss scaling. No equations, update rules, or self-citations are quoted in the supplied text that would reduce this fixed-point claim to a definitional identity, a fitted parameter renamed as prediction, or a load-bearing self-citation chain. The structural role assigned to parlay trades for improving identifiability is presented as an empirical and modeling consequence rather than an input smuggled in by prior work. The derivation therefore remains self-contained against the stated dynamics and model class, consistent with an honest non-finding.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review yields no explicit free parameters, axioms, or invented entities; the model class and trading dynamics are referenced but not detailed.

pith-pipeline@v0.9.0 · 5830 in / 1105 out tokens · 22958 ms · 2026-05-21T10:21:29.687901+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

22 extracted references · 22 canonical work pages · 2 internal anchors

  1. [1]

    Jacob Abernethy, Yiling Chen, and Jennifer Wortman Vaughan. 2013. Efficient Market Making via Convex Optimization. InACM EC

    work page 2013

  2. [2]

    Guillermo Angeris, Tarun Chitra, Alex Evans, and Stephen Boyd. 2020. Improved Price Oracles: Constant Function Market Makers.Proceedings of the 2nd ACM Conference on Advances in Financial Technologies(2020).https://arxiv.org/abs/2003.10001

    work page arXiv 2020

  3. [3]

    Julian Besag. 1975. Statistical analysis of non-lattice data.The Statistician24, 3 (1975), 179– 195

    work page 1975

  4. [4]

    Guy Bresler. 2015. Efficiently Learning Ising Models on Arbitrary Graphs. InProceedings of the Forty-Seventh Annual ACM Symposium on Theory of Computing. doi:10.1145/2746539. 2746631

    work page doi:10.1145/2746539 2015

  5. [5]

    Pennock, and Jennifer Wortman Vaughan

    Yiling Chen, Nicolas Lambert, David M. Pennock, and Jennifer Wortman Vaughan. 2010. Elic- iting Predictions and Recommendations for Decision Making.ACM Transactions on Embedded Computing Systems9, 4 (2010), 39:1–39:21.https://www.microsoft.com/en-us/research/ wp-content/uploads/2016/04/TEAC-final1.pdf

    work page 2010

  6. [6]

    Yiling Chen and David M. Pennock. 2007. A Utility Framework for Bounded-Loss Market Makers.Journal of Artificial Intelligence Research(2007)

    work page 2007

  7. [7]

    Price Updating in Combinatorial Prediction Markets with Bayesian Networks

    Yiling Chen, David M. Pennock, Michael H. Bowling, and Nicolas Lambert. 2008. Price Updating in Combinatorial Prediction Markets with Bayesian Networks. InProceedings of the 23rd AAAI Conference on Artificial Intelligence.https://arxiv.org/abs/1202.3756

    work page internal anchor Pith review Pith/arXiv arXiv 2008

  8. [8]

    Rafael Frongillo, Ian Kash, and Mark D. Reid. 2012. Interpreting Prediction Markets: A Stochas- tic Approach. InAdvances in Neural Information Processing Systems.https://papers.nips. cc/paper/4529-interpreting-prediction-markets-a-stochastic-approach

    work page 2012

  9. [9]

    2002.Logarithmic Market Scoring Rules for Modular Combinatorial Information Aggregation

    Robin Hanson. 2002.Logarithmic Market Scoring Rules for Modular Combinatorial Information Aggregation. Technical Report. George Mason University

    work page 2002

  10. [10]

    Robin Hanson. 2003. Combinatorial Information Market Design.Information Systems Frontiers 5, 1 (2003), 107–119. doi:10.1023/A:1022058209073

    work page doi:10.1023/a:1022058209073 2003

  11. [11]

    Robin Hanson. 2007. Logarithmic Market Scoring Rules for Modular Combinatorial Information Aggregation.Journal of Prediction Markets(2007). 20

    work page 2007

  12. [12]

    2019.Introduction to Online Convex Optimization

    Elad Hazan. 2019.Introduction to Online Convex Optimization. Now Publishers.https: //arxiv.org/abs/1909.05207

    work page arXiv 2019

  13. [13]

    Kalshi. 2024. Kalshi: Event Contracts Exchange.https://kalshi.com. Accessed: 2026-03-19

    work page 2024

  14. [14]

    Bruce G. Lindsay. 1988. Composite likelihood methods. InStatistical Inference from Stochastic Processes, N. U. Prabhu (Ed.). Contemporary Mathematics, Vol. 80. American Mathematical Society, Providence, RI, 221–239

    work page 1988

  15. [15]

    C., Roughgarden, T., and Zhang, A

    Jason Milionis, Ciamac C. Moallemi, Tim Roughgarden, and Anthony Lee Zhang. 2022. Auto- mated Market Making and Loss-Versus-Rebalancing.arXiv preprint arXiv:2208.06046(2022). https://arxiv.org/abs/2208.06046

    work page arXiv 2022

  16. [16]

    Abraham Othman and Tuomas Sandholm. 2010. When Do Markets with Simple Traders Fail?. In Proceedings of the 9th International Conference on Autonomous Agents and Multiagent Systems (AAMAS).https://www.ifaamas.org/Proceedings/aamas2010/pdf/01%20Full%20Papers/ 18_02_FP_0266.pdf

    work page 2010

  17. [17]

    Polymarket. 2024. Polymarket: A Decentralized Information Market Platform.https: //polymarket.com. Accessed: 2026-03-19

    work page 2024

  18. [18]

    Wainwright, and John D

    Pradeep Ravikumar, Martin J. Wainwright, and John D. Lafferty. 2010. High-Dimensional Ising Model Selection Usingℓ 1-Regularized Logistic Regression.The Annals of Statistics38, 3 (2010), 1287–1319. doi:10.1214/09-AOS691

    work page doi:10.1214/09-aos691 2010

  19. [19]

    Moallemi

    Dan Robinson and Ciamac C. Moallemi. 2024. pm-AMM.https://www.paradigm.xyz/2024/ 11/pm-amm

    work page 2024

  20. [20]

    Interaction Screening: Efficient and Sample-Optimal Learning of Ising Models

    Marc Vuffray, Sidhant Misra, Andrey Y. Lokhov, and Michael Chertkov. 2016. Interaction Screen- ing: Efficient and Sample-Optimal Learning of Ising Models. InAdvances in Neural Information Processing Systems.https://arxiv.org/abs/1605.07252

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  21. [21]

    Wainwright and Michael I

    Martin J. Wainwright and Michael I. Jordan. 2008. Graphical Models, Exponential Families, and Variational Inference.Foundations and Trends in Machine Learning1, 1–2 (2008), 1–305. https://people.eecs.berkeley.edu/~jordan/papers/wainwright-jordan-fnt.pdf

    work page 2008

  22. [22]

    Martin Zinkevich. 2003. Online Convex Programming and Generalized Infinitesimal Gradi- ent Ascent. InProceedings of the 20th International Conference on Machine Learning (ICML). https://www.cs.cmu.edu/~maz/publications/techconvex.pdf 21 Table 5: Quoted combo YES prices versus probabilistic benchmark for two-leg combos with plausibly independent legs (NBA ...

    work page 2003