Paper: arXiv 2509.06510

Authors: Philippe Bergault, Sébastien Bieber, Leandro Sánchez-Betancourt

Abstract

We study the problem of optimal liquidity withdrawal for a representative liquidity provider (LP) in an automated market maker (AMM). LPs earn fees from trading activity but are exposed to impermanent loss (IL) due to price fluctuations. While existing work has focused on static provision and exogenous exit strategies, we characterise the optimal exit time as the solution to a stochastic control problem with an endogenous stopping time. Mathematically, the LP’s value function is shown to satisfy a Hamilton-Jacobi-Bellman quasi-variational inequality, for which we establish uniqueness in the viscosity sense. To solve the problem numerically, we develop two complementary approaches: a Euler scheme based on operator splitting and a Longstaff-Schwartz regression method. Calibrated simulations highlight how the LP’s optimal exit strategy depends on the oracle price volatility, fee levels, and the behaviour of arbitrageurs and noise traders. Our results show that while arbitrage generates both fees and IL, the LP’s optimal decision balances these opposing effects based on the pool state variables and price misalignments. Lastly, we find the optimal fee level for the representative LP when they play the exit strategy we derived. This work contributes to a deeper understanding of dynamic liquidity provision in AMMs and provides insights into the sustainability of passive LP strategies under different market regimes.

Complexity vs Empirical Score

  • Math Complexity: 9.0/10
  • Empirical Rigor: 7.0/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper presents advanced mathematical theory, including Hamilton-Jacobi-Bellman quasi-variational inequalities and viscosity solutions, indicating high math complexity. It also demonstrates empirical rigor through calibrated simulations, two numerical methods (Euler scheme and Longstaff-Schwartz), and detailed sensitivity analyses on fee levels, volatility, and trader behavior.

Research Flowchart

  flowchart TD
  A["Research Goal<br>Find optimal exit time for LPs<br>in AMMs given IL and fees"] --> B["Model Formulation<br>Stochastic Control Problem<br>Endogenous Stopping Time"]
  B --> C{"Data & Calibration"}
  C --> C1["Oracle Price Volatility"]
  C --> C2["Fee Levels"]
  C --> C3["Trader Behavior<br>Arbitrageurs & Noise"]
  B --> D["Computational Methods"]
  D --> D1["HJB QVIE<br>Operator Splitting"]
  D --> D2["Longstaff-Schwartz<br>Regression"]
  D --> E["Simulation & Analysis"]
  E --> F["Key Findings & Outcomes"]
  F --> F1["Optimal exit balances<br>fees vs. impermanent loss"]
  F --> F2["Strategy depends on<br>pool state & misalignment"]
  F --> F3["Derived optimal fee level<br>for the representative LP"]

Evidence Card

What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card

Proposes

The paper characterises the optimal exit time for a liquidity provider in a constant‑function market, proves the value function is the unique viscosity solution of a Hamilton‑Jacobi‑Bellman quasi‑variational inequality, and provides two numerical solution methods (an Euler scheme with operator split

Source sentence
We characterise the optimal exit time of a liquidity provider from a CFM. Mathematically, we show that the value function of the corresponding optimal stopping problem is the unique viscosity solution to a Hamilton–Jacobi–Bellman quasi-variational inequality (HJB QVI) that delineates the optimal exit region. 2.

Assumes 6

  • The external price follows a Brownian motion with volatility σ. model
    Source sentence
    We introduce a probability measure ˆP such that W is a Brownian motions and ˆN a, ˆN b are Poisson processes with intensity a0 > 0.
  • Arbitrage and noise‑trader order arrivals are independent Poisson processes with intensities a0, a1, a2. model
    Source sentence
    We introduce a probability measure ˆP such that W is a Brownian motions and ˆN a, ˆN b are Poisson processes with intensity a0 > 0. Under this probability measure, W, ˆN a and ˆN b are independent.
  • The AMM uses a constant‑function market with trading function f(x,y)=xy. market
    Source sentence
    We endow the space (Ω, F) with the ˆP−augmented canonical filtration F := (Ft)t∈T generated by (χt)t∈T. We study an automated market maker (AMM) for a pair of assets X and Y (e.g., USDC and ETH) with trading function f(x, y) = x y. This type of AMM is referred to as constant function market (CFM).
  • Liquidity providers earn fees proportional to trading volume and incur impermanent loss due to price fluctuations. market
    Source sentence
    LPs earn fees from trading activity but are exposed to impermanent loss (IL) due to price fluctuations. However, LPs also face risks, most notably the so-called impermanent loss (IL), which arises due to price fluctuations between the pooled assets and external market prices.
Show 2 more
  • Risk‑aversion can be incorporated by maximising expected utility of PnL with constant absolute risk aversion ψ. model
    Source sentence
    Appendix A. Risk-averse liquidity provider Here we provide details for the problem of a risk averse LP wishing to maximise the expected utility of her PnL. More precisely, we study the problem sup τ∈T E  −exp  −ψ �P X τ + SτP Y τ + Rτ   , (A.1) where ψ > 0 represents the absolute risk aversion of the LP.
  • The model parameters (e.g., fee level r, order‑flow intensities) are calibrated to Binance and Uniswap V2 data for the ETH‑USDC pair. data
    Source sentence
    We use the model parameters from Aqsha et al. (2025) who used market data from Binance and Uniswap V2 in the pair ETH-USDC between 1 January 2022 and 30 April 2022 to calibrate model parameters, that is, we take S0 = Z0 = 2820 $, σ = 0.0569 S0 $ · day−1/2, Y0 = 50, 000 ETH, and X0 = Y0 Z0 $.

Data

  • Binance & Uniswap V2 ETH‑USDC market data · Binance / Uniswap crypto-onchain public
    2022‑01‑01 to 2022‑04‑30 · ETH‑USDC pair
    Source sentence
    We use the model parameters from Aqsha et al. (2025) who used market data from Binance and Uniswap V2 in the pair ETH-USDC between 1 January 2022 and 30 April 2022 to calibrate model parameters, that is, we take S0 = Z0 = 2820 $, σ = 0.0569 S0 $ · day−1/2, Y0 = 50, 000 ETH, and X0 = Y0 Z0 $.
  • Synthetic calibrated simulations synthetic unknown
    Source sentence
    Calibrated simulations highlight how the LP's optimal exit strategy depends on the oracle price volatility, fee levels, and the behaviour of arbitrageurs and noise traders. To test the Longstaff–Schwartz algorithm against a well-known Euler scheme, we employ the following model parameters: n = 1,440, m = 5,000, T = 1 day, d = 3, σ = 100 $ · day−1/2, a0 = 4 day−1, a1 = 8 day−1, a2 = 0.04 $−1 · day−1, ξ = 1, X0 =

Recognised sources: Binance

Costs

Modeled Costs are part of the reported headline results.

Fees are modelled as a parameter r; the performance criterion (fees minus impermanent loss) increases approximately linearly with r beyond a threshold.

Source sentence
(b) The performance criterion (defined as fees collected minus impermanent loss), which is non-negative in expectation since exiting the pool immediately is always admissible, increases approx-imately linearly with the fee level beyond a certain threshold. This is because, once fees are high enough, LPs optimally choose to remain in the pool until the terminal time.

Finds 5

  • Both average fees collected and average impermanent loss at the optimal exit time are concave functions of oracle price volatility. performance
    Source sentence
    Our results reveal several key findings: (a) Both the average fees collected and the average impermanent loss at the optimal exit time are concave functions of the volatility of the oracle price.
  • The performance criterion (fees minus impermanent loss) grows approximately linearly with the fee level once fees are high enough, prompting LPs to stay in the pool until the terminal time. performance
    Source sentence
    (b) The performance criterion (defined as fees collected minus impermanent loss), which is non-negative in expectation since exiting the pool immediately is always admissible, increases approx-imately linearly with the fee level beyond a certain threshold. This is because, once fees are high enough, LPs optimally choose to remain in the pool until the terminal time.
  • Higher arbitrage activity raises fee revenue but also increases realised impermanent loss up to the maximum loss implied by price volatility. performance
    Source sentence
    (c) Higher activity by arbitrageurs and noise traders leads to higher fee revenues. However, arbi-trageurs also increase the realised impermanent loss at the exit time, up to the maximum loss implied by the oracle price volatility.
  • Increased noise‑trader activity raises fees while leaving average impermanent loss unchanged. performance
    Source sentence
    In contrast, an increase in noise trader activity raises the fees collected while leaving impermanent loss unchanged on average.
  • LPs remain in the pool when fee income outweighs potential impermanent loss, which occurs more often under low volatility, high fees, or high trading activity. theoretical
    Source sentence
    LPs stay in the pool when the fee income outweighs the potential impermanent loss, which occurs more frequently when volatility is low, fees are high, or trading activity is high.

Checked by others

Supported by

  • Optimal Decisions for Liquid Staking: Allocation and Exit Timing (2025)
    On the question: When should a liquidity provider exit a pool given fee accumulation and impermanent loss?
    Compare the findings

    This paper: The performance criterion (fees minus impermanent loss) grows approximately linearly with the fee level once fees are high enough, prompting LPs to stay in the pool until the terminal time.

    That paper: Numerical analyses show that fee accumulation can make remaining in the pool beneficial despite price declines.

    Both findings agree that high enough fees make staying in the pool optimal; B shows performance grows linearly with fee level, supporting A's claim that fee accumulation can outweigh price declines.

  • Model-free Hedging of Impermanent Loss in Geometric Mean Market Makers (2023)
    On the question: How do fee levels impact impermanent loss and liquidity provider behavior in AMMs?
    Compare the findings

    This paper: LPs remain in the pool when fee income outweighs potential impermanent loss, which occurs more often under low volatility, high fees, or high trading activity.

    That paper: In simulations, higher fees lead to lower impermanent loss.

    Both findings agree that higher fees reduce impermanent loss; B further notes this effect is stronger with low volatility, high trading activity, or when fee income exceeds loss.

  • Inefficiency of CFMs: hedging perspective and agent-based simulations (2023)
    On the question: How does noise‑trader activity affect fees and loss for liquidity providers?
    Compare the findings

    This paper: Increased noise‑trader activity raises fees while leaving average impermanent loss unchanged.

    That paper: The loss magnitude is independent of the arrival rate of noise traders.

    Both findings agree that loss magnitude (impermanent loss) stays unchanged with more noise‑trader activity; B adds that fees increase, while A notes independence from arrival rate.

Contradicted by

  • Model-free Hedging of Impermanent Loss in Geometric Mean Market Makers (2023)
    On the question: How do changes in fees affect impermanent loss for liquidity providers?
    Compare the findings

    This paper: Increased noise‑trader activity raises fees while leaving average impermanent loss unchanged.

    That paper: In simulations, higher fees lead to lower impermanent loss.

    Finding A: higher fees → lower impermanent loss. Finding B: fees raised by noise‑trader activity → impermanent loss unchanged. The effect on IL differs.

Cited by 2 papers in the archive (1 by independent authors)

Code

No repository by the authors found in the paper.

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.

Cited References (114)

  1. Price-aware automated market makers: Models beyond brownian prices and static liquidity — → 2405.03496
  2. Automated market making: the case of pegged assets — → 2411.08145
  3. Arbitrageurs’ profits, LVR, and sandwich attacks: batch trading as an AMM design response (2023) — link
  4. An economic model of a decentralized exchange with concentrated liquidity (2025)
  5. An analysis of Uniswap markets (2020)
  6. The replicating portfolio of a constant product market (2020)
  7. The replicating portfolio of a constant product market with bounded liquidity (2021)
  8. Lectures on BSDEs, stochastic control, and stochastic differential games with financial applications (2016)
and 106 more
  1. Optimal Decisions for Liquid Staking: Allocation and Exit Timing — → 2507.14810
  2. Fundamentals of Stochastic Filtering (2009)
  3. Controlled Markov Processes and Viscosity Solutions (2006)
  4. Mean field games with partial information for algorithmic trading (2018) — link
  5. Equilibrium reward for liquidity providers in automated market makers — → 2503.22502
  6. Martingales `a temps discret (1972)
  7. Pathwise stochastic control with applications to robust filtering (2020)
  8. Portfolio choice, portfolio liquidation, and portfolio transition under drift uncertainty (2019)
  9. Optimal fees for geometric mean market makers (2021)
  10. Improved price oracles: Constant function market makers (2020)
  11. Strategic liquidity provision in uniswap v3 (2021) — link
  12. The need for fees at a dex: How increases in fees can increase dex trading volume (2022)
  13. Liquidity provision by automated market makers (2020)
  14. The economics of automated market makers (2022)
  15. Coexisting exchange platforms: Limit order books and automated market makers (2025)
  16. On the quality of cryptocurrency markets: Centralized versus decentralized exchanges (2021) — link
  17. Decentralized exchange: The uniswap automated market maker (2025)
  18. Automated market making and arbitrage profits in the presence of fees — → 2305.14604
  19. Parameter uncertainty in the Kalman–Bucy filter (2019)
  20. Liquidity in competitive dealer markets (2021)
  21. Optimal Dynamic Fees in Automated Market Makers — → 2506.02869
  22. The adoption of blockchain-based decentralized exchanges (2021) — link
  23. Execution and statistical arbitrage with signals in multiple automated market makers (2023)
  24. Optimal exiting for liquidity provision in constant function market makers (2025)
  25. Decentralized Finance and Automated Market Making: Predictable Loss and Optimal Liquidity Provision (2024)
  26. Arbitrage on Decentralized Exchanges — → 2507.08302
  27. Model-Free Hedging of Impermanent Loss in Geometric Mean Market Makers with Proportional Transaction Fees (2024)
  28. Model-free hedging of impermanent loss in geometric mean market makers — → 2303.11118
  29. Automated market making and loss-versus-rebalancing (2022) — link
  30. A myersonian framework for optimal liquidity provision in automated market makers — → 2303.00208
  31. A Structural Model of Automated Market Making (2025)
  32. Liquidity provision of utility indifference type in decentralized exchanges — → 2502.01931
  33. Decentralised finance and automated market making: Execution and speculation — → 2307.03499
  34. Optimal execution of portfolio transactions (2001)
  35. Liquidity provision with adverse selection and inventory costs (2023)
  36. The Financial Mathematics of Market Liquidity: From optimal execution to market making (2016)
  37. Algorithmic market making in dealer markets with hedging and market impact (2023)
  38. Convex analysis and variational problems (1999)
  39. Internalisation by electronic FX spot dealers (2019)
  40. A mean-field game of market-making against strategic traders (2023)
  41. Optimal trading strategy and supply/demand dynamics (2013)
  42. Dynamic programming approach to principal–agent problems (2018)
  43. On the impossibility of informationally efficient markets (1980)
  44. Liquidity and market structure (1988)
  45. Detecting Toxic Flow — → 2312.05827
  46. Nash Equilibrium between Brokers and Traders — → 2407.10561
  47. Unwinding Stochastic Order Flow: When to Warehouse Trades — → 2310.14144
  48. Strategic insider trading equilibrium: a filter theory approach (2012)
  49. Double-execution strategies using path signatures (2022)
  50. Trading foreign exchange triplets (2020)
  51. A Mean Field Game between Informed Traders and a Broker — → 2401.05257
  52. A survey of nonsymmetric Riccati equations (2002)
  53. Non—Blow—Up Conditions for Riccati—type Matrix Differential and Difference Equations (2000)
  54. Optimal signal-adaptive trading with temporary and transient price impact (2022)
  55. Optimal trading: The importance of being adaptive (2021)
  56. Mean-field games with differing beliefs for algorithmic trading (2020)
  57. Trading algorithms with learning in latent alpha models (2019)
  58. Merton’s optimal investment problem with jump signals (2021) — link
  59. Stochastic calculus and applications (2015)
  60. Enhancing trading strategies with order book signals (2018)
  61. Option Volume Imbalance as a predictor for equity market returns (2022) — link
  62. Unwinding Toxic Flow with Partial Information — → 2407.04510
  63. Optimal execution with stochastic delay (2021)
  64. Brokers and Informed Traders: dealing with toxic flow and extracting trading signals (2022)
  65. Liquidation in target zone models (2018)
  66. Closed-loop N ash competition for liquidity (2023)
  67. Continuous-time Stochastic Control and Optimization with Financial Applications (2009)
  68. Portfolio liquidation games with self-exciting order flow (2022)
  69. Liquidity in competitive dealer markets (2021)
  70. Investing for the long run when returns are predictable (2000)
  71. Optimal Trade Execution under Endogenous Order Flow (2023)
  72. Algorithmic trading with model uncertainty (2017)
  73. Continuous-time stochastic control and optimization with financial applications (2009)
  74. Incorporating signals into optimal trading (2019)
  75. Incorporating order-flow into optimal execution (2016)
  76. Algorithmic and high-frequency trading (2015)
  77. Buy low, sell high: A high frequency trading perspective (2014)
  78. Matrix Riccati equations in control and systems theory (2012)
  79. Fast and Slow Optimal Trading with Exogenous Information (2023)
  80. Market microstructure theory (1998)
  81. Mean-variance hedging for partially observed drift processes (2001)
  82. Solvability of differential Riccati equations and applications to algorithmic trading with signals (2022)
  83. Automated market makers: Mean-variance analysis of lps payoffs and design of pricing functions — → 2212.00336
  84. Uniswap v4 core [draft] (2023)
  85. Uniswap v3 core (2021)
  86. Automated market makers designs beyond constant functions (2023)
  87. am-amm: An auction-managed automated market maker — → 2403.03367
  88. Strategic Bonding Curves in Automated Market Makers (2024)
  89. Optimal Portfolio in Partially Observed Stochastic Volatility Models (2001)
  90. Equilibrium in Functional Stochastic Games with Mean-Field Interaction — → 2306.05433
  91. Optimal Fees for Liquidity Provision in Automated Market Makers — → 2508.08152
  92. Optimal exit from Uniswap v3 and best expected return for a liquidity provider (2025)
  93. Weak dynamic programming principle for viscosity solutions (2011)
  94. Optimal stochastic control, stochastic target problems, and backward SDE (2012)
  95. High Frequency Trading in a Limit Order Book (2008)
  96. Informed speculation with imperfect competition (1989)
  97. Continuous auctions and insider trading (1985)
  98. Buy Low, Sell High: A High Frequency Trading Perspective (2014) — link
  99. Market making with alpha signals (2020)
  100. Riccati Differential Equations (1972)
  101. PORTFOLIO OPTIMIZATION IN A DEFAULT MODEL UNDER FULL/PARTIAL INFORMATION (2015)
  102. Optimal Market Making under Partial Information with General Intensities (2020)
  103. Mean Field Game Systems with Common and Latent Processes (2018)
  104. Applied Stochastic Control of Jump Diffusions (2007) — link
  105. Über die partiellen Differenzengleichungen der mathematischen Physik (1928) — link
  106. Valuing American options by simulation: A simple least-squares approach (2001)

20 of 114 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.