- The paper introduces a stochastic control approach to dynamically optimize fees in AMMs, eliminating over 90% of the frictionless loss.
- It demonstrates that the optimal fee depends solely on instantaneous volatility, remaining independent of liquidity providers' wealth and risk-aversion.
- Simulation results confirm that a pro-cyclical fee schedule consistently outperforms static strategies, yielding a reliable excess growth rate.
Optimal Dynamic Fees for AMMs via Stochastic Control: Summary and Analysis
Model Foundations and Structural Innovations
This paper introduces a structural framework for dynamic fee optimization in constant-product Automated Market Makers (AMMs), treating the fee-setting process as a stochastic control problem. Building on the loss-versus-rebalancing (LVR) framework [26, 27], the approach models the liquidity provider's (LP) wealth relative to a continuously rebalanced benchmark portfolio. The key insight is that the trading fee directly modulates two opposing economic forces: it increases revenue from uninformed traders but simultaneously discourages such volume, while wider fees deter arbitrageurs by expanding the no-arbitrage band, thereby reducing adverse selection losses.
Crucially, the fee only affects the drift of relative wealth and does not enter its diffusion term. This yields a reduction: the optimal control problem is ergodic and one-dimensional, parameterized solely by volatility, with the fee as a pure feedback policy in instantaneous variance. The solutions are characterized through ergodic Hamilton-Jacobi-Bellman (HJB) and Poisson equations, offering a transparent analytic and numerical recipe.
Analytical Results: Fee Independence and Pro-cyclicality
The model yields several structural results:
- Wealth- and risk-aversion independence: The growth-optimal fee is entirely independent of both the LP’s wealth and risk-aversion parameter under Constant Relative Risk Aversion (CRRA) utility. The optimal policy is a function solely of instantaneous volatility.
- Static fee optimality under constant volatility: If volatility is stationary, the optimal fee collapses to a constant, which formally rationalizes fixed fee tiers in stable markets.
- Pro-cyclical fee schedule: In dynamic regimes, the optimal fee is strictly increasing in instantaneous variance, confirming that liquidity providers should increase fees in turbulent markets. This result sharpens existing heuristics and corrects models that mistakenly assume adverse selection increases with the fee.
- Numerical solution and invariance: When volatility is modeled as a Cox-Ingersoll-Ross (CIR) process, the optimal policy results from pointwise maximization and subsequent solution of the scalar Poisson equation. Notably, under logarithmic utility, the optimal fee is invariant to price jumps.
Robustness and Extensions
The methodological reduction enables robust extensions:
- Jump-diffusion reference prices: The optimal fee remains invariant to price jumps, as the impact is symmetric across both AMM and the rebalancing benchmark.
- Competition among venues: Even in competitive equilibria, the volatility dependence of the fee persists, with aggregate fee levels compressed by market competition but the pro-cyclical structure retained.
- Gas costs and discrete updates: Incorporating on-chain update costs yields an impulse-control problem. Empirically, small dead-bands around the frictionless target allow infrequent updates, mitigating gas costs with negligible performance loss.
The paper calibrates the model under realistic blockchain and asset volatility parameters and simulates performance across thousands of volatility paths. The optimal dynamic fee uniformly dominates every static and heuristic alternative, achieving a modest but strictly positive excess growth rate improvement. The dominance is pathwise, not just in expectation, a nontrivial property deriving from the structural reduction.
Practical Protocol Design and Implementation
The optimal fee is implementable via a lookup table parameterized by external volatility estimates. The model recommends tying the fee update to manipulation-resistant signals rather than the AMM's own price—preventing adversarial feedback loops. Uninformed flow parameters should be periodically re-estimated—but with conservative governance rather than aggressive automated adaptation, given the circularity induced by the fee policy itself.
Given that well-tuned static fees capture almost all available value in stable regimes, the case for dynamic fees is strongest for token pairs exhibiting high volatility dispersion and low elasticity in uninformed demand.
The framework can be extended to concentrated-liquidity AMMs, where adverse-selection risk becomes range-dependent, adding a new state variable but retaining the stochastic control structure. Strategic feedback between fee policy and liquidity provision, and auction-mechanism-based fee updates, are cited as promising research directions.
Simulation Results and Quantitative Benchmarks
The calibration validates several key claims:
- At long-run volatility (60% annual), the frictionless LVR rate is 4.5% per year. The optimal dynamic fee eliminates 93.1% of this loss.
- The dynamic fee’s pro-cyclical schedule rises from 27 bps (volatility 30%) to 47 bps (volatility 80%), nearing the protocol cap at extreme volatility.
- Out-of-sample simulations show the optimal dynamic fee exceeds the best static by 5.7 bps annualized, with strictly nonnegative improvement on every path.
- Continuous updating incurs minimal gas costs; dead-band strategies recover near-optimal performance with reduced update frequency.
Implications and Future Directions
The findings reinforce the relevance of volatility-linked fee schedules, providing a rigorous foundation for practitioners and offering structural clarity on comparative statics and policy design. In DeFi, a deterministic volatility feedback fee minimizes adverse selection and can be deployed with minimal state dependence, enhancing robustness and capital efficiency.
Future research should address range-dependent adverse selection in concentrated liquidity and endogenize liquidity provision dynamics. Further exploration of fee auctions and strategic arbitrageur models will yield comprehensive designs for next-generation AMMs.
Conclusion
This paper presents a rigorous stochastic control framework for dynamic fee setting in AMMs, deriving a structurally robust, volatility-linked, pro-cyclical fee schedule independent of LP wealth and risk aversion. The reduced-form ergodic problem yields a tractable, practical policy with pathwise dominance over static and heuristic alternatives. The methodology provides a clear foundation for protocol-level fee policies and motivates a rich set of theoretical extensions for future research (2606.21769).