agent-based amm simulation

In decentralized finance (DeFi), arbitrage against blue-chip asset pairs in automated market makers (AMMs) constitutes a primary revenue source but also induces substantial impermanent loss (IL) due to informed order flow. The central challenge lies in suppressing informed arbitrage while preserving liquidity for uninformed trades. Method: This paper introduces the first analytical model of AMM arbitrage dynamics as a stochastic walk with state-dependent rewards, yielding a tractable theoretical framework for fee optimization. Through rigorous stochastic process analysis, sensitivity analysis, and quantification of value retention, we derive closed-form relationships between fee rates and pool asset value preservation. Results: We formally prove the existence of an optimal fee rate and characterize its structural properties—balancing revenue generation against IL mitigation. This work establishes the first analytically rigorous and practically actionable paradigm for designing AMM fee mechanisms.

Automated Market Makers (AMMs) in DeFi expose liquidity providers (LPs) to arbitrage losses—primarily impermanent loss—due to price lag relative to external markets. Method: This paper proposes an adaptive AMM curve design grounded in the Glosten-Milgrom market microstructure model. It derives, for the first time, the optimal curve differential equation minimizing arbitrage loss; constructs an on-chain, oracle-free real-time market price estimator; and establishes an equivalence between static curve dynamics and optimality conditions. Implementation integrates Kalman filtering, stochastic differential equation modeling, and Uniswap v4’s hook-based programmability for on-chain execution. Contribution/Results: Experiments demonstrate substantial reduction in LP losses, faster price convergence, improved capital efficiency, and strong robustness under high volatility and adversarial conditions.

Current automated market maker (AMM) designs lack a unified taxonomy and standardized evaluation criteria, resulting in elevated financial risk, suboptimal capital efficiency, and poor cross-domain adaptability. To address this, we propose the first systematic AMM taxonomy framework, integrating mechanism design, game-theoretic analysis, and software engineering principles to establish a verifiable and extensible modeling and comparative paradigm. Leveraging this framework, we design three AMM prototypes—each formally aligned with core token issuance and exchange requirements—thereby bridging the disciplinary gap between economic modeling and systems implementation. Our contributions include: (i) a rigorous, modular classification schema enabling principled AMM analysis; (ii) executable, specification-driven prototypes supporting formal verification; and (iii) a structured design methodology for developers, facilitating robust, multi-scenario deployment and advancing the engineering of sustainable cryptographic economies. (149 words)

This work addresses two fundamental limitations of automated market makers (AMMs): deadweight loss under informed order flow and insufficient fee revenue under uninformed order flow. To this end, we propose an on-chain, censorship-resistant sealed-bid auction mechanism that dynamically selects “pool managers” to adjust the trading fee rate of constant-product AMMs in real time and selectively capture arbitrage opportunities. Our approach is the first to integrate on-chain auctions with AMM governance, enabling adaptive fee pricing and strategic allocation of arbitrage rights. We formally prove that, under reasonable price sensitivity and game-theoretic equilibrium assumptions, the equilibrium liquidity provision strictly dominates that of fixed-fee AMMs. Moreover, the mechanism supports frictionless liquidity provider (LP) entry and exit, coupled with dynamic rent extraction, thereby significantly enhancing fee revenue and mitigating MEV exposure.

This paper investigates optimal trading and statistical arbitrage in constant-product market makers (CPMs), focusing on the trade-off between exchange rate risk and execution cost. Methodologically, it introduces a novel stochastic convexity-based modeling framework for execution cost—formalizing the transaction function’s convexity as the core cost metric—and reveals its linear dependence on trade size and nonlinear dependence on liquidity depth and spot exchange rates. Building upon this, the paper unifies centralized exchanges and CPMs into a single exchange-rate formation model and designs a real-time trading strategy that explicitly incorporates stochastic convexity costs. The approach integrates stochastic optimization, convex analysis, multi-market equilibrium modeling, and empirical econometrics. Out-of-sample evaluation demonstrates that the proposed strategy significantly reduces average execution costs and enhances the stability of arbitrage profits. Moreover, the convexity cost model exhibits high fidelity and strong robustness in replicating actual AMM behavior.

研究使用强化学习解决去中心化金融市场中流动性提供者如何动态调整资本分配的问题，以优化收益并减少风险。

This study addresses the lack of a systematic understanding of market efficiency and return structures in automated market makers (AMMs) under the dynamic interaction between liquidity providers (LPs) and arbitrageurs. The authors develop a dynamic equilibrium framework for constant function market makers (CFMMs) that integrates slippage, trading fees, price impact, noise trading, endogenous gas fees, and time-varying volatility to capture strategic interactions among heterogeneous participants. They uncover an inherent bid–ask asymmetry in CFMMs, demonstrate that liquidity provision is strictly dominated in a pure arbitrage environment, and establish a non-degenerate interior equilibrium incorporating execution costs, which explains the inverted-U relationship between liquidity supply and volatility. Combining dynamic game-theoretic modeling, closed-form solutions, and on-chain data calibration, the paper empirically validates asymmetric price impact and non-monotonic optimal liquidity provision, offering theoretical foundations for AMM mechanism design and LP strategies.

This work addresses the losses incurred by liquidity providers in automated market makers (AMMs) due to adverse selection, commonly quantified by loss-versus-rebalancing (LVR). To mitigate this issue, the paper proposes a partially active AMM mechanism that partitions liquidity reserves into active and passive components, with only the active portion participating in trades. The proportion of active liquidity is dynamically adjusted at the beginning of each block. Drawing inspiration from index tracking optimization, this approach simultaneously reduces LVR and constrains deviations of asset weights from a target portfolio allocation. Theoretical analysis and empirical experiments demonstrate that, compared to conventional constant-function market makers (CFMMs), the proposed mechanism significantly enhances liquidity provider wealth while effectively balancing the trade-off between adverse selection costs and portfolio drift.

This paper investigates equilibrium behavior and endogenous price formation in complex financial markets under multi-agent strategic interaction. We propose a synchronized multi-agent reinforcement learning (MARL) framework that integrates a high-fidelity limit-order-book simulator—extended from ABIDES-Gym—with an augmented Kyle model. To accommodate heterogeneous agents, we decouple state observation from kernel-level interruptions, enabling parallel decision-making; execution optimization is embedded directly into the strategic policy game to achieve endogenous liquidity modeling. Our contributions are threefold: (i) the first MARL framework to rigorously preserve market microstructure constraints—including price-time priority; (ii) successful replication of gradual price discovery dynamics; and (iii) causal identification of how execution strategies shape market-maker behavior and price evolution. Empirical evaluation confirms model validity and supports reproducible equilibrium analysis.

This study addresses the significant loss in liquidity value (LVR) experienced by liquidity providers (LPs) in concentrated liquidity automated market makers like Uniswap v3, which stems from adverse selection and is inadequately compensated by existing fee mechanisms. To tackle this challenge, the authors propose and evaluate a dynamic fee mechanism that integrates volatility and order flow toxicity metrics. For the first time, they develop a multi-agent simulation framework grounded in realistic on-chain microstructure, incorporating a Heston stochastic volatility market model, block propagation delays, and heterogeneous participant behaviors—including MEV searchers and smart routing strategies. Experimental results demonstrate that the proposed mechanism substantially increases LP fee revenue under price lag risk, enabling their hedged PnL to turn positive, thereby validating the efficacy of dynamically compensating for LVR rather than attempting its complete elimination.