Complex Markets and Mean Field Games: Beyond Basic Models

Abstract: This work builds on the theoretical frameworks presented in "Liquidity pools as mean field games: A new framework" and "Liquidity pools as mean field games with transaction costs" by the same author, where the strategic interactions among traders in a constant-product market-making protocol were modelled using mean field games (MFG), first without transaction costs and then incorporating them. Here we present the formulation of a more complete model that integrates three types of agents: traders, liquidity providers (LPs), and arbitrageurs. While we do not establish existence results for this general model, the formulation identifies the main technical difficulties and lays the groundwork for future work. The LP acts as a dominating player in the sense of 'Mean field games with a dominating player' by Bensoussan, Chau, and Yam: its strategy influences the mean field distribution of the traders, and the equilibrium is sought as a solution to the coupled system of three problems that constitute a Major-Minor game. The arbitrageurs operate by solving the optimization problem presented in "An analysis of uniswap markets" by Angeris et al., and their impact on the LP is captured through the loss-versus-rebalancing of "Automated market making and loss-versus-rebalancing" by Milionis et al. The material in this article should be read as an open research proposal rather than a collection of closed results.

• The paper formulates a Major-Minor mean field game for constant-product AMMs in which liquidity providers dominate traders and arbitrageurs affect outcomes through loss-versus-rebalancing.
• The model couples liquidity decisions, trader slippage, aggregate order flow, pool reserves, and external-price arbitrage through nonlinear and partly path-dependent dynamics, with state augmentation restoring Markovian structure for one component.
• The paper establishes no existence, uniqueness, or equilibrium results, instead identifying multiplicative coupling, simultaneous SHJB–Fokker-Planck–adjoint solvability, and Lions-derivative regularity as priorities for analysis and numerical validation.

Overview and scope

This paper extends a line of work in which strategic interaction among traders in a constant-product automated market maker (AMM) liquidity pool is modeled as a mean field game (MFG). Earlier contributions by the same author established existence and characterization results for the traders' equilibrium, first without transaction costs and then with fees. The present article departs from that program in an important way: it does not prove new theorems. Instead, it formulates a substantially richer model incorporating three agent classes—traders, liquidity providers (LPs), and arbitrageurs—and explicitly frames itself as an open research proposal rather than a collection of closed results. The author states plainly that no existence results are established for the general model; the contribution lies in identifying the correct game-theoretic architecture (a Major-Minor MFG) and cataloguing the technical obstacles to proving equilibrium existence.

The structural novelty relative to prior work is twofold. First, arbitrageurs are introduced as agents who realign the pool price with an external market price whenever the no-arbitrage band $(1-\tau)\,m_p \leq m_u \leq (1+\tau)\,m_p$ is violated, where $\tau$ is the pool fee. Their impact on LPs enters through loss-versus-rebalancing (LVR), following Milionis, Moallemi, Roughgarden, and Zhang. Second, the LP is modeled as a dominating player in the sense of Bensoussan, Chau, and Yam: its strategy shifts the mean field distribution of the traders, so the equilibrium must be sought as a simultaneous solution of three coupled problems rather than as a single fixed point.

Agent dynamics

Traders. The $N$-trader model carries over from the earlier fee-inclusive framework. Trader $i$ controls a trading rate $\alpha_t^i$ affecting ETH inventory via $dX_t^i = \alpha_t^i\,dt + \sigma^i\,dW_t^i$, with USDT inventory dynamics derived from execution at the mid-price under the post-fee factor $\phi = 1-\tau$. Pool reserves are driven by the average of trader controls, and the price after a trade of $\Delta_t^X$ ETH follows the constant-product update with fees applied in two stages.

Liquidity provider. A single LP deposits both tokens at rate $\alpha_t^{LP}$, constrained to add liquidity proportionally ($\beta_t^{LP} = \alpha_t^{LP} P_t$) so as not to move the internal price. The paper adopts a sign convention in which negative $\tau$0 corresponds to withdrawing liquidity (increasing pool share). The LP's state comprises ETH holdings, USDT holdings, and pool share $\tau$1, with total inventory $\tau$2. Notably, LPs pay no fees on deposit or withdrawal, but their actions alter the invariant $\tau$3 and hence the slippage experienced by traders.

Arbitrageur. The arbitrageur solves the convex optimization problem of Angeris et al., whose closed-form solution is

$\tau$4

active when $\tau$5. Crucially, the arbitrageur is treated as exogenous to the LP–trader game: it optimizes against the external price without strategic interaction with the other agents, and its effect is fed into the system dynamics through the LVR channel.

Interactions among agents

Three coupling mechanisms connect the agents. The LP–price channel redefines the pool's equilibrium reserves as $\tau$6, accumulating the LP's historical provision, and yields a modified price dynamics in which the drift depends jointly on $\tau$7 and the aggregate trader flow—a multiplicative dependence that later obstructs standard existence arguments.

The LP–trader channel operates through slippage. The paper proposes embedding the slippage cost $\tau$8 directly into the trader's USDT inventory dynamics, so that slippage reduces what the trader receives but not what is delivered to the pool. This makes the LP's liquidity decisions directly relevant to trader payoffs.

The arbitrageur–LP channel is captured by LVR. Under the assumption that the external price is a martingale geometric Brownian motion under the risk-neutral measure, the pool value decomposes as $\tau$9 with $N$0 and $N$1; moreover, the arbitrageur's cumulative profit equals cumulative LVR. The author argues this decomposition remains valid with active LPs and transaction costs, and incorporates it as a running cost in the LP's objective via $N$2. Total pool reserves are then written as the sum of three contributions: the LP-induced equilibrium, the arbitrageur impact $N$3, and the negative of average trader flow.

One caveat deserves emphasis: the claim that the LVR analysis "carries over analogously" to active LPs and transaction costs is asserted rather than proven here, and the martingale assumption on the external price is inherited from the original LVR paper rather than justified in this setting.

The coupled Major-Minor formulation

In the mean field limit, the aggregate reserve impact is summarized by $N$4, where $N$5 is the law of trader controls. Both the LP's and the representative trader's running costs are expressed through a common kernel involving $N$6, making explicit how each side's payoff depends on the other's aggregate behavior.

A useful technical observation resolves one potential obstruction: because $N$7 depends on the cumulative quantity $N$8, the problem appears path-dependent; however, introducing the auxiliary state $N$9 restores a Markovian formulation over progressively measurable controls, without imposing continuity on control paths.

The equilibrium concept instantiates the General Major-Minor framework through three problems:

1. Representative agent control: minimize the trader's cost functional given the dominating player's process and an exogenous measure $i$0.
2. Mean field consistency: find $i$1 satisfying the fixed-point condition $i$2, where $i$3 is the law induced by the optimal state in Problem 1.
3. Dominating player control: minimize the LP's cost functional given the consistent $i$4.

The mean field term is carried by the joint state–control law $i$5, with marginals recovering $i$6 and $i$7. The author notes that Problems 1–3 constitute the SHJB–Fokker-Planck–Adjoint system of the Major-Minor framework, with the AMM-specific costs defined above—but stops short of deriving necessary optimality conditions or proving anything about solvability.

Technical difficulties

The paper is candid that four obstacles block a direct extension of the earlier existence proofs:

• Multiplicative nonlinear coupling: rewards depend on products of state, aggregate actions, reserves, and LP liquidity, defeating the additive decomposition required by sandwich-type conditions used previously.
• Path-dependence: reserves depend on full histories of all agents' actions through cumulative integrals, beyond the Markovian setting—partially mitigated for the LP by state augmentation, though the analogous issue for the coupled system remains.
• Simultaneous three-problem coupling: existence of equilibria for the SHJB–Fokker-Planck–Adjoint system under AMM price nonlinearities is stated as an open problem.
• Regularity of the Lions derivative: well-definedness of $i$8 under the specific AMM nonlinearities is not verified.

These are genuine gaps, not formalities: the multiplicative structure in particular means the earlier existence machinery cannot be applied without new estimates.

Proposed directions

Four research avenues are outlined. Tractable simplifications include treating the arbitrageur as exogenous with a given external price process, fixing the LP strategy at constant liquidity, or adopting myopic short-horizon traders—each potentially amenable to the techniques of the prior papers. Numerical validation would adapt the existing pipeline (fixed-point solver, $i$9-Nash test, convergence in $\alpha_t^i$0) to probe equilibrium existence and stability empirically before formal proof. Extension of the LVR analysis asks whether strategic LP optimization can quantitatively reduce arbitrage losses relative to the passive-LP benchmark—an economically substantive question the formulation raises but cannot answer. Calibration with on-chain data from Uniswap or similar protocols is proposed to validate qualitative predictions.

Limitations and open questions

The central limitation is acknowledged throughout: no existence, uniqueness, or characterization results are provided for the full model, and the LVR transfer to active-LP settings is assumed rather than demonstrated. Specific open questions include whether equilibria exist for the coupled three-problem system under AMM nonlinearities; whether the Lions derivative is sufficiently regular on the relevant measure spaces; and whether strategic LP behavior materially changes LVR relative to passive provision. The treatment of arbitrageurs as fully exogenous also leaves open a genuinely endogenous model in which arbitrageurs compete strategically among themselves.

Conclusion

This paper contributes a precise formulation of a three-agent MFG for constant-product liquidity pools, structured as a Major-Minor game with the LP as dominating player, arbitrageurs entering exogenously via the Angeris et al. optimization and the LVR decomposition, and traders forming the minor population. Its value is methodological: it identifies the exact points where existing existence theory breaks down—multiplicative reward coupling, path-dependence, simultaneous three-problem solvability, and Lions derivative regularity—and proposes concrete simplifications and numerical strategies as next steps. As the author intends, it should be read as a research proposal whose analytical payoff remains to be delivered.