ParlayMarket: AMM for Joint Contracts

• ParlayMarket is an automated market making framework that supports parlay-style (joint-outcome) contracts with coherent probabilistic pricing and unified liquidity pools.
• It employs a pairwise exponential-family model and an LMSR-style cost function to ensure convergence, control pricing error, and manage risk via stochastic gradient updates.
• ParlayMarket demonstrates practical efficacy in prediction, sports, and derivatives markets by efficiently aggregating information and supporting scalable, multi-event contract structures.

ParlayMarket is an automated market making (AMM) framework designed to support parlay-style (joint-outcome) contracts with coherent probabilistic pricing, unified liquidity pools, and explicit convergence guarantees for joint distribution learning under trade-induced updates. It provides structural support for base (single-event) and joint (multi-event) contracts—especially relevant in prediction, sports, and derivatives markets—enabling information aggregation and risk management spanning exponentially many outcomes via a parametric model with polynomially-sized state (Rana et al., 23 Mar 2026).

1. Fundamentals of Parlay-Style Markets

Parlay contracts, central to ParlayMarket, provide payoffs based on the simultaneous occurrence of multiple binary or categorical events. Each $k$-leg parlay contract $S$ has payoff

$f_S(x) = \prod_{i \in S} x_i$

where $x_i \in \{0,1\}$ represents the realized outcome of base event $i$, and $S$ is a nonempty subset of $M$ base events. Singleton $S$ yields base contracts; general $|S|>1$ correspond to higher-order parlays.

Existing prediction and derivatives AMMs typically only support singleton contracts, enforcing independence or requiring ad hoc liquidity splitting, resulting in inconsistent pricing, fragmented liquidity, and poor identifiability of cross-event dependence (Rana et al., 23 Mar 2026). ParlayMarket addresses these deficiencies by enabling coherent pricing and unified liquidity across the full parlay contract lattice.

2. Market-Making and Pricing Mechanism

ParlayMarket exploits a pairwise exponential-family (Ising-type) model over joint event outcome space $x \in \{0,1\}^M$: $S$0 with $S$1, $S$2 the log-partition, and contract prices given by

$S$3

for every subset $S$4.

The AMM cost function is an LMSR-style log-partition-based function: $S$5 where $S$6 encodes the net vector of outstanding contracts, $S$7 is the market depth parameter, and the infinitesimal marginal price for contract $S$8 is $S$9.

Incoming trade requests $f_S(x) = \prod_{i \in S} x_i$0 (reflecting trader beliefs) are processed by stochastic gradient steps on a composite cross-entropy loss

$f_S(x) = \prod_{i \in S} x_i$1

with update

$f_S(x) = \prod_{i \in S} x_i$2

where

$f_S(x) = \prod_{i \in S} x_i$3

and

$f_S(x) = \prod_{i \in S} x_i$4

with $f_S(x) = \prod_{i \in S} x_i$5 the sufficient statistics vector $f_S(x) = \prod_{i \in S} x_i$6 (Rana et al., 23 Mar 2026).

3. Theoretical Guarantees and Convergence

The system is designed so that repeated trading dynamics converge, under mild convexity and Lipschitz regularity, to a unique fixed point minimizing the divergence objective

$f_S(x) = \prod_{i \in S} x_i$7

whose minimizer $f_S(x) = \prod_{i \in S} x_i$8 matches all moment constraints $f_S(x) = \prod_{i \in S} x_i$9, i.e., the I-projection of the true joint $x_i \in \{0,1\}$0 onto the pairwise exponential family.

Discrete-time SGD dynamics converge geometrically fast: $x_i \in \{0,1\}$1 where $x_i \in \{0,1\}$2 is the model's strong convexity parameter, $x_i \in \{0,1\}$3 the step size, and $x_i \in \{0,1\}$4 the per-step noise, so limiting mean parameter error is $x_i \in \{0,1\}$5 (Rana et al., 23 Mar 2026).

Pricing and monetary loss are tightly controlled:

• Pricing error for any $x_i \in \{0,1\}$6 is $x_i \in \{0,1\}$7 for $x_i \in \{0,1\}$8.
• Instantaneous market-maker loss on trade $x_i \in \{0,1\}$9 is $i$0.
• Aggregate per-round loss scales as $i$1 (polynomial in number of base markets).

A crucial structural finding is that inclusion of higher-order parlays ($i$2) supplies “curvature” to the loss's Hessian, boosting $i$3 and accelerating convergence (especially for weakly correlated base events, where absence of parlays results in degenerate curvature and inefficient learning) (Rana et al., 23 Mar 2026).

4. Parlay Wagering Theory and Kelly Optimization

For independent base events under multiplicative parlay pricing, optimal Kelly betting strategies for singles, doubles, and higher-order parlays factorize as products of one-event Kelly optimizers. For a single event with state prices $i$4 and outcome probabilities $i$5, the implicit-cash Kelly solution is

$i$6

where $i$7 (Long, 27 Mar 2026).

For $i$8 independent events, the joint (parlay) Kelly stake on $i$9 is

$S$0

yielding wealth

$S$1

This factorization implies the “active-leg criterion”: only parlays whose every leg is active in the single-event Kelly solution receive positive stake.

Forbidding parlays and restricting to singles is a low-order truncation: growth-rate loss is $S$2 for edge size $S$3, and optimal singles-only stakes deviate from isolated Kelly stakes only at cubic order, explaining empirical near-proportionality (Whitrow asymptotics) (Long, 27 Mar 2026).

5. ParlayMarket as a Constant Function Market Maker

Within the CFMM (Constant Function Market Maker) framework, ParlayMarket can be realized with the “minimum token” parlay payoff

$S$4

yielding a trading function

$S$5

with $S$6 as the invariant. The marginal price for any token is constant in the interior, and liquidity is uniform up to the hard boundary, beyond which slippage is infinite and trading halts (Angeris et al., 2021). This design allows a smart-contract implementation where arbitrary swaps are permitted so long as post-trade reserves satisfy the invariant.

6. Empirical Properties and Practical Implementation

Empirically, ParlayMarket achieves:

• Geometric convergence in simulated settings (≈22 rounds per 10× error reduction for models with $S$7 to $S$8, with observed per-market loss decaying exponentially in $S$9).
• Flat aggregate AMM loss per round across $M$0, consistent with quadratic scaling.
• In historical market replays (e.g., NBA parlay data), standalone ParlayMarket AMMs recorded high net PnL, competitive Sharpe ratios, and superior risk-adjusted returns compared to independence-only baselines, which systematically misprice joint outcomes (Rana et al., 23 Mar 2026).

Key engineering requirements:

• For state-price-based parlay wagering: solve single-event Kelly optimizations, assemble the full contract book via outer product, and restrict menu to active tickets according to the active-leg criterion.
• AMM implementation with pairwise exponential family and log-partition cost scaling is computationally tractable (parameter space $M$1, with no high-dimensional convex programming needed).

7. Limitations and Extensions

Current ParlayMarket implementations are limited by model misspecification for higher-order dependencies and the restriction to binary event representation (Rana et al., 23 Mar 2026). Extensions include:

• Preprocessing layers to identify latent binary clauses in richer outcome spaces.
• Expansion to higher-order exponential-family interactions as capital budgets increase.
• Adaptive tuning of liquidity ($M$2) and learning rates ($M$3) to address temporal volatility.

These directions can potentially reduce identifiability error and capture more complex dependence in multi-event markets, subject to feasibility constraints imposed by transaction costs and capital efficiency.

---

References

• "ParlayMarket: Automated Market Making for Parlay-style Joint Contracts" (Rana et al., 23 Mar 2026)
• "Optimal Parlay Wagering and Whitrow Asymptotics: A State-Price and Implicit-Cash Treatment" (Long, 27 Mar 2026)
• "Replicating Market Makers" (Angeris et al., 2021)