{"id":"e3f1868b-453e-4182-9613-ed0113f601a9","arxiv_id":"2411.08145","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":10,"one_line_summary":"A new AMM model uses nested Ornstein-Uhlenbeck dynamics with filtering to quote stablecoin and liquid-staking token pairs, and beat geometric Brownian motion-based AMMs in simulation.","lead":"The paper builds an automated market maker for stablecoin and liquid-staking pairs, modeling their exchange rate as a two-level nested mean-reverting process and solving for optimal quotes. If its simulations hold up, it could improve how liquidity providers price and hedge pegged crypto assets.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The empirical claim of outperformance rests on an uncalibrated synthetic order-flow model: changing the ad hoc intensity parameters in Section 4.4 can reverse the PnL comparison, so the central claim is not yet supported.","rationale":"The paper's mathematical core is coherent: the nested OU representation, the MLE procedure, the Kalman-filter Riccati equation, and the quadratic-Hamiltonian control approximation are internally consistent, and applying this machinery to pegged-asset AMM design is a legitimate contribution. Credit is due for using real Pyth price series and for being transparent about simulation assumptions. My concern is not with the mathematics but with the load-bearing empirical inference. The comparison in Section 4.4 is a joint test of the nested OU model and an arbitrary order-flow model. Since the trade-flow parameters are stated without estimation and the authors themselves describe the values as \"assumed,\" the plotted frontiers cannot establish the claimed outperformance. A calibrated order-flow experiment would settle this. I therefore do not move the reader's conditional verdict; the paper should not be accepted as fully empirically validated without either such calibration or a substantial sensitivity analysis. I agree with the reader that the uncalibrated trade-flow model is the weakest point; my formulation adds that the same concern applies to both synthetic and \"real-price\" runs, since the latter still use synthetic order flow.","tokens_in":25149,"tokens_out":7585,"duration_ms":82390,"concrete_test":"Obtain a timestamped, size-marked dataset of actual swaps for a USDC/USDT pool (e.g., Curve 3pool or a Uniswap v3 pool) and for wstETH/WETH over a comparable three-month window, using the pool's executed marginal prices relative to the Pyth reference price as the markup δ. Fit the marked point process intensities Λ0,1(z,δ) and Λ1,0(z,δ), including the trade-size distribution, by maximum likelihood, then re-run the Section 4.4 simulations replacing the ad hoc λ, a, b, and fixed trade size with the fitted kernels. If the NOU AMM no longer dominates the GBM AMM on the mean-standard deviation frontier, the central empirical claim is not supported. A supplementary split-sample or rolling-window calibration would also remove the look-ahead in parameter estimates.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central assertion that the NOU AMM \"leverages the multi-level nested OU dynamics to achieve levels of average excess P&L that outperform those of the GBM AMM by far\" (Section 4.4) is tested only against a synthetic trade-flow model. In Section 4.4, trade arrivals are marked point processes with logistic intensities Λ(z,δ)=λ/(1+exp(a+bδ)), with λ=250 day^-1, a=0, b=10,000, and fixed sizes 100,000 USDT or 40 WETH. These parameters are not estimated or justified from any real market, and no sensitivity analysis is reported. The excess P&L functional in Eq. (3), and hence the optimal markups in Eq. (10), depend directly on this assumed intensity through H and the inverse map. Therefore Figures 10-12, including the \"real price\" experiments, evaluate the strategies under an arbitrary demand curve; real data enter only as price paths, not as order flow. If actual liquidity-taking is less elastic or arrives with different sizes, the comparison can change. This makes the empirical headline conditional on an unverified auxiliary model rather than on the nested OU contribution itself. An additional attribution gap is that no one-factor OU AMM is benchmarked, so even a confirmed outperformance over GBM would not show that the multi-level nested structure, rather than mere mean reversion, is responsible.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a multi-level nested Ornstein-Uhlenbeck model for the exchange rate of pegged crypto-assets such as stablecoin pairs and liquid-staking-token pairs. It derives the Gaussian law and covariance of the observed price, develops a maximum-likelihood calibration procedure and a Kalman-filter-type estimator for the unobserved mean-reversion target, and then embeds the filtered dynamics in a CARA expected-utility market-making problem. The HJB equation is approximated via a quadratic Hamiltonian, leading to a Riccati system for the value-function coefficients and closed-form approximate markups. The resulting NOU AMM is compared numerically with the GBM AMM of the authors' earlier work on simulated prices and on historical Pyth price paths for USDC/USDT and wstETH/WETH.","tokens_in":25503,"tokens_out":7517,"duration_ms":75972,"significance":"The theoretical core is a useful and mostly self-contained contribution: the covariance formula, the filtering equations, and the quadratic-Hamiltonian reduction to Riccati ODEs are carefully derived and go beyond the standard one-factor OU market-making setup. The paper also explicitly addresses a real gap, namely AMM design for stationary, mean-reverting pegged-asset pairs rather than geometric-Brownian-motion pairs. If the empirical claims were supported, the framework would be of practical value for stablecoin and liquid-staking-token pools. However, the numerical validation is currently conditional on an uncalibrated synthetic order-flow model, and the attribution of the performance to the multi-level nested structure is not established.","major_comments":[{"comment":"The central outperformance claim is evaluated under an uncalibrated synthetic order-flow model. The marked point process intensities are fixed as lambda_{0,1}=lambda_{1,0}=250 day^{-1}, a=0, b=10,000, with fixed trade sizes of 100,000 USDT or 40 WETH, and these values are not estimated from any data set. The objective in Eq. (3) and the optimal markups in Eq. (10) depend directly on these choices through H and the inverse map bar-delta, so Figures 10-12 report PnL conditional on an arbitrary demand curve; real price data enter only as price paths, not as order flow. The abstract's claim that the AMM 'consistently yields efficient quotes' and Section 4.4's claim of outperformance 'by far' are therefore not yet supported. A sensitivity analysis over b, lambda, and trade size, and ideally a calibration or comparison with real trade-flow data, is needed before these conclusions can be drawn.","section":"Section 4.4, Eq. (3) and Eq. (10)"},{"comment":"The synthetic-price validation is circular: prices are simulated from the same multi-level nested OU dynamics (Table 3) that the NOU AMM assumes, so Figure 10 checks internal consistency rather than predictive performance. The real-price simulations in Figures 11-12 are less circular, but they still inherit the synthetic order-flow assumption described above, so they do not break the dependence of the headline result on the unverified demand model.","section":"Section 4.4, synthetic price experiment"},{"comment":"The comparison to a GBM AMM does not identify the source of any outperformance. A one-factor OU AMM benchmark is missing; without it, the paper cannot support the claim that the multi-level nested structure, rather than mean reversion per se, is responsible for the results. This attribution matters because the paper's stated novelty is specifically the multi-level nested OU dynamics. Adding a one-factor OU AMM to Figures 10-12, or otherwise isolating the contribution of the second mean-reversion level, is necessary to support the 'leverages the multi-level nested OU dynamics' claim.","section":"Section 4.4, benchmark choice"}],"minor_comments":[{"comment":"After filtering, the state variable is hat-U with volatility hat-nu defined in Eq. (2), so the second-order terms in Eq. (7) and Eq. (9) should read (1/2)hat-nu^2 partial^2_{hat-U hat-U} theta and sigma hat-nu partial^2_{S hat-U} theta, not nu^2 and sigma nu. The quadratic risk term already uses hat-nu, so the displayed PDE is internally inconsistent.","section":"Eq. (7) and Eq. (9)"},{"comment":"The statement that the decrease in performance on the right side of Figure 12 is 'not statistically significant' should be supported by confidence intervals or standard errors across the 300 simulations; otherwise the assertion cannot be checked by the reader.","section":"Section 4.4, wstETH/WETH paragraph"},{"comment":"The zoomed panel in Figure 12 lacks axis labels and a legend, making it difficult to determine what is being displayed.","section":"Figure 12"},{"comment":"There are several typographical errors: 'Similary' in the proof of Proposition 1, 'diffential' in the appendix, and 'allow use to define' in Section 2.4 should read 'allow us to define.'","section":"Various"},{"comment":"The claim that the volatility parameter does not matter for the GBM AMM because the simulations span a large range of gamma deserves a one-sentence justification; as written it is not obvious why the two effects are redundant.","section":"Footnote 16"}],"recommendation":"major_revision","confidential_remarks":"The theoretical sections are solid and should be publishable once the empirical validation is reframed or strengthened. The main risk is that the current abstract and Section 4.4 overstate what the simulations establish: the order-flow model is arbitrary, and the multi-level structure is not isolated from a simpler OU benchmark. I would encourage the editor to send the revised version back to a referee familiar with the market-making literature to check the Eq. (7) typo and the revised empirical protocol."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper's core is a competent application of nested OU dynamics (Hull-White style) to stablecoin and liquid staking pairs, with MLE calibration, a Kalman-type filter, and the quadratic Hamiltonian approximation from [16] to obtain near-closed-form markups. The math hangs together: the Gaussian covariance expression, the filtering SDEs, and the Riccati ODE system are all derived carefully and are reproducible from the text. That is real, citable work.\n\nWhat is genuinely new is the packaging: multi-level nested OU for pegged crypto exchange rates, with the latent target treated as an unobservable state and fed through an optimal control problem for AMM quotes. The wstETH/WETH treatment with a yield discount is a sensible extension, and the real-data figures for USDC/USDT and wstETH/WETH are suggestive.\n\nThe soft spot is exactly where the reader and stress-test note point: Section 4.4. The trade-flow intensities (lambda = 250/day, a = 0, b = 10,000) are chosen ad hoc, not estimated, and no sensitivity analysis is reported. The synthetic-price test is partly circular because the simulator uses the same nested OU model the strategy assumes. The \"real price\" experiments still evaluate PnL under that arbitrary demand curve; real data enter only as price paths. So the abstract's \"efficient quotes\" and the Section 4.4 claim of outperforming GBM \"by far\" are not yet supported by the evidence. The claim is conditional on an unverified auxiliary model. Also, no one-factor OU AMM is benchmarked, so even confirmed outperformance over GBM would not isolate the benefit of the multi-level structure. Parameter estimates come without standard errors.\n\nThese are addressable flaws, not load-bearing ones. The optimal control derivation does not depend on the order-flow calibration; the weakness is confined to the validation section. The citation pattern is fine, and the self-citations point to genuine prior work. This is an honest, technically solid paper with an overreaching empirical section.\n\nWho benefits? Researchers working on AMM design, market making with mean reversion, or DeFi liquidity provision. A serious referee can push for order-flow calibration, sensitivity analysis, standard errors, and a one-factor OU benchmark; those requests would improve the paper without changing its structure. I would send it to peer review rather than desk-reject it.","headline":"A clean modeling-and-control framework for pegged-asset AMMs, whose headline outperformance claim is only as strong as an uncalibrated simulated order flow.","tokens_in":26030,"tokens_out":1646,"would_cite":true,"duration_ms":17790,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["91G80","60G35"],"pacs":[],"model":"deepseek-v4-flash","headline":"A two-layer mean-reverting price model lets an automated market maker quote better prices for pegged cryptoassets than a Brownian-motion model does.","keywords":["automated market making","pegged assets","stablecoins","liquid staking tokens","nested Ornstein-Uhlenbeck processes","stochastic filtering","optimal control","excess P&L"],"falsifier":"Rerun the NOU-versus-GBM comparison on the same real price series, but replace the assumed logistic arrival intensities and fixed trade sizes with intensities and a size distribution estimated from actual transaction data on a USDC/USDT pool. If the NOU AMM no longer dominates on average excess P&L at comparable standard deviations, the paper's performance claim is an artifact of the assumed flow rather than of the nested OU price model.","tokens_in":24941,"feed_emoji":"💱","tokens_out":11641,"duration_ms":100559,"temperature":0.7,"pith_summary":"The paper tries to establish that the exchange rate between two cryptoassets anchored to the same value—stablecoins pegged to the same fiat currency, or a liquid staking token against its native token—is better described by a two-layer mean-reverting process than by the Brownian-motion prices usually assumed in market making. In this model, the observed rate reverts to a hidden target that itself reverts to a long-run peg, so temporary de-pegging phases last longer and the market maker need not over-trust the peg. The authors build an automated market maker on this model, with quotes computed from the observed price and from a filtered estimate of the hidden target, and they formulate the quoting problem as maximising the liquidity provider's risk-averse expected utility of excess profit or loss relative to simply holding the pool tokens. Numerical simulations on USDC/USDT and wstETH/WETH data are used to argue that this AMM consistently produces better mean-risk trade-offs than the geometric-Brownian AMM, especially for the fast-mean-reverting wstETH/WETH pair. If correct, the framework gives pegged-asset pools a practical way to recalibrate quotes as slow movements of the peg unfold.","feed_headline":"Mean-reversion AMM outdoes Brownian quotes for pegged crypto","feed_subtitle":"Modeling the hidden peg target improves liquidity provision in stablecoin and liquid-staking pools.","key_machinery":"The central object is the multi-level nested Ornstein-Uhlenbeck process, defined by $dS_t = -\\kappa(S_t-U_t)\\,dt + \\sigma\\,dW^S_t$ and $dU_t = -\\eta(U_t-\\bar U)\\,dt + \\nu\\,dW^U_t$ with $\\kappa>\\eta>0$: the observed exchange rate pulls toward a hidden target, and the hidden target pulls toward the peg. Two pieces of machinery make this usable for market making. First, a stochastic filtering step turns the unobservable $U$ into an observable filtered state $\\hat U_t$ satisfying a diffusion driven by the innovations of $S$, with the conditional variance following a deterministic Riccati ODE. Second, the Hamilton-Jacobi-Bellman equation for the exponential-utility control problem is approximated through quadratic Hamiltonians, so the value function is a quadratic polynomial in inventory, $S$, and $\\hat U$, and the optimal markups are obtained from the solution of a small system of Riccati ODEs for time-dependent coefficients.","core_discovery":"The central claim is that exchange-rate dynamics for pegged cryptoassets should be modeled not by a single Ornstein-Uhlenbeck process nor by geometric Brownian motion, but by a multi-level nested Ornstein-Uhlenbeck system: the observed exchange rate $S_t$ mean-reverts to an unobservable process $U_t$, and $U_t$ itself mean-reverts to a fixed long-run value $\\bar U$. The paper derives the Gaussian structure, the maximum-likelihood estimator for the parameters, and a stochastic filter that produces the conditional estimate $\\hat U_t = \\mathbb{E}[U_t \\mid \\mathcal F^S_t]$. On top of this it solves a stochastic control problem for an AMM that quotes bid and ask prices in terms of observable quantities only, and it approximates the solution with a quadratic-Hamiltonian method that reduces the optimal quotes to numerically tractable Riccati equations. Tested on simulated prices and on real USDC/USDT and wstETH/WETH prices, the resulting AMM (the paper's NOU AMM) is claimed to outperform the GBM AMM from the authors' earlier work in the mean-standard-deviation plane; for the wstETH/WETH pair the paper states the NOU AMM 'achieves levels of average excess P&L that outperform those of the GBM AMM by far.'","pith_inferences":["Strictly beyond the paper: the parameter $\\nu$ sets a trust-in-peg dial—small $\\nu$ keeps the hidden target near $\\bar U$, large $\\nu$ lets the target chase the price; this could be stress-tested by simulating a sharp de-peg followed by re-pegging and watching how the filtered quotes widen and recover.","Strictly beyond the paper: because the quote rule only needs an observable price series, the same control machinery should transfer to any asset pair with an anchor, such as an ETF against its net asset value or an on-chain wrapped token against its underlying, provided the two-level OU assumption fits.","Strictly beyond the paper: the simulations use an external indicative price as the mid-price, so in a live deployment the AMM would inherit oracle risk (lag, manipulation, or stale feeds); the paper does not quantify that risk.","Strictly beyond the paper: the trade-flow model is assumed rather than estimated, so the most direct test of the performance claim is to fit the logistic arrival intensities and trade-size distribution to actual pool or centralized-exchange transaction data and rerun the NOU-vs-GBM comparison."],"forward_implications":["The optimal AMM quotes depend only on observable quantities: the pool's inventory position, the external exchange rate, and the filtered estimate of the hidden mean-reversion target.","The quadratic-Hamiltonian approximation reduces a four-dimensional HJB equation to low-dimensional ODEs, so the quoting policy can be computed quickly enough for practical use and has an explicit ergodic limit when the horizon is long.","For the wstETH/WETH pair, where mean reversion is fast, the NOU AMM's average excess P&L per unit of standard deviation is far above the GBM AMM frontier, because geometric Brownian motion is misspecified even on the short time scale of inventory round trips.","For USDC/USDT, both strategies perform respectably, but the NOU AMM dominates the GBM AMM in the mean-standard-deviation plane on simulated prices and on real prices.","The same framework applies to a non-rebasing liquid staking token relative to its native token after discounting the price series by an estimated constant staking yield."],"supporting_citations":[{"why":"supplies the exponential-utility objective used to define the optimal market-making problem.","marker":"[6]"},{"why":"contributes the markup-accumulation structure and marked point process trade flow that the AMM model inherits.","marker":"[11]"},{"why":"shows how stationary mean-reverting prices can be used in market making, the modelling route adapted here to pegged assets.","marker":"[12]"},{"why":"provides the GBM AMM benchmark and the Hodl-benchmark excess P&L convention used in every comparison.","marker":"[14]"},{"why":"supplies the quadratic-Hamiltonian approximation that reduces the HJB equation to Riccati ODEs.","marker":"[16]"},{"why":"provides the simplifying ansatz and intensity assumptions used to pass from the HJB equation to the approximate optimal markups.","marker":"[31]"},{"why":"is the two-factor Hull-White analogy cited for replacing a fixed mean-reversion target with a stochastic latent level.","marker":"[40]"}],"fun_headline_variants":[],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The performance comparison rests on an assumed trade-flow model—trades arrive with logistic intensities $\\lambda=250\\,\\mathrm{day}^{-1}$, $a=0$, $b=10{,}000\\,\\mathrm{USDT}^{-1}$ (or $\\mathrm{WETH}^{-1}$) and fixed trade sizes—none of which are calibrated to real order flow; if real liquidity takers respond to markups differently, the reported excess P&L gap may be misleading.","fun_headline_variants_meta":{"error":"Client error '402 Payment Required' for url 'https://api.deepseek.com/chat/completions'\nFor more information check: https://developer.mozilla.org/en-US/docs/Web/HTTP/Status/402"},"cache_creation_input_tokens":0},"created_at":"2026-08-12T21:55:31.585941+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Rerun the NOU-versus-GBM comparison on the same real price series, but replace the assumed logistic arrival intensities and fixed trade sizes with intensities and a size distribution estimated from actual transaction data on a USDC/USDT pool. If the NOU AMM no longer dominates on average excess P&L at comparable standard deviations, the paper's performance claim is an artifact of the assumed flow rather than of the nested OU price model.","supporting_citations":[{"cited_title":"Dealing with multi-currency inventory risk in foreign exchange cash markets.Risk Magazine (Cutting Edge), 2023","cited_arxiv_id":null,"evidence_quote":"contributes the markup-accumulation structure and marked point process trade flow that the AMM model inherits."},{"cited_title":"Automated market makers: Mean-variance analysis of LPs payoffs and design of pricing functions.Digital Finance, 6(2):225–247, 2024","cited_arxiv_id":null,"evidence_quote":"provides the GBM AMM benchmark and the Hodl-benchmark excess P&L convention used in every comparison."},{"cited_title":"Optimal market making.Applied Mathematical Finance, 24(2):112–154, 2017","cited_arxiv_id":null,"evidence_quote":"provides the simplifying ansatz and intensity assumptions used to pass from the HJB equation to the approximate optimal markups."},{"cited_title":"Numerical procedures for implementing term structure models i: Single- factor models","cited_arxiv_id":null,"evidence_quote":"is the two-factor Hull-White analogy cited for replacing a fixed mean-reversion target with a stochastic latent level."}],"review_version":1}