{"id":"9d8d132b-240a-4aad-8782-0929bd62efdc","arxiv_id":"2508.02971","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"An automated market maker's adverse-selection cost (LVR) equals the funding fee of a replicating portfolio of perpetual continuous-installment puts, enabling forward-looking LVR estimation and liquidity-band design.","lead":"This paper models an automated market maker's liquidity position as a portfolio of perpetual American continuous-installment options, and proves that a measure of arbitrage loss, loss-versus-rebalancing, equals the options' continuous funding fees. It then shows how to choose liquidity boundaries so that future loss-versus-rebalancing is nearly constant and predictable.","discovery_kind":"unification","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Section 6.1 constant-LVR construction sets V'(S) equal to the delta of a long CI put, which is negative, while CFAMM delta is the positive token-0 reserve; no feasible construction is proved.","rationale":"The reader's weakest_assumption is on target: Section 6.1 asserts without proof that a CFAMM band can replicate a single CI option's delta. My review sharpens this into a concrete internal inconsistency: the delta of a long CI put is negative, while a CFAMM LP's delta equals the positive token-0 reserve. Hence Eq. (13) cannot hold for any CFAMM with positive reserves; the intended short-put version needs a sign correction and a reserve-positivity proof. This matters because the paper's second headline result and the LP-facing design rules (Section 8.3) depend on it. The central Theorem 5 identity, by contrast, is an asymptotic statement supported by Lemmas 3–4 and the numerical error analysis; its main weakness is the abstract's exact-equals wording, which the reader already flagged. Because the Section 6.1 issue is addressable by a construction (or by softening the claim), the CONDITIONAL verdict stands; no change is needed.","tokens_in":17475,"tokens_out":28525,"duration_ms":299411,"concrete_test":"Using a row from Table 1 (e.g., r=0.05, σ_eff=60%, τ=1 month, K=100, q≈49), compute the proposed CFAMM value function V(S)=P_q(S;K) and check (i) whether X_q(S;K) is positive on (S_l,S_u); (ii) whether y(S)=V(S)−S·X_q(S;K) is nonnegative on the band. Repeat with V(S)=−P_q(S;K)+C for a constant C chosen so reserves are nonnegative. If the literal Eq. (13) fails the sign test, or if the short-put version violates y(S)≥0 for some band in Table 1, the constant-LVR positions are not realizable as CFAMMs and the construction must be revised or removed.","verdict_should_be":"UNCHANGED","load_bearing_attack":"In Section 6.1 the paper posits a concentrated-liquidity CFAMM whose delta satisfies X(S)=V'(S)≡X_q(S;K*) on [a,b]=[S_l,S_u]. But X_q(S;K*) is the delta of a long perpetual American CI put, which is negative on the continuation region, whereas any CFAMM LP position has positive delta equal to the token-0 reserve x(S). Equation (13) as written is therefore unsatisfiable by a CFAMM with positive reserves. The intended object must be a short put (delta −X_q) plus a cash constant, and even then one must prove that the induced reserve curves x(S)=−X_q(S;K*) and y(S)=V(S)−S·x(S) are nonnegative on the band and define a valid constant-function invariant. The paper supplies neither the sign correction nor an existence proof, so the 'approximately constant, price-independent LVR' design rules in Sections 6.1 and 8.3 rest on an unverified premise.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes an option-theoretic representation of constant-function AMM (CFAMM) liquidity provision. It models an LP position as a portfolio of perpetual American continuous-installment (CI) put options, using the limit of large installment rate q to obtain a step-function delta. The authors claim two key results: (a) the instantaneous funding fee of the delta-replicating CI portfolio equals the instantaneous LVR of the CFAMM position, and hence integrated fees equal integrated LVR; and (b) a CFAMM position whose delta matches that of a single CI put suffers approximately constant, price-independent LVR over a forward window. The paper also studies discrete-strike replication error and calibrates the constant volatility parameter from the ATM implied volatility term structure, with error bounds.","tokens_in":17636,"tokens_out":5344,"duration_ms":62858,"significance":"If the central claims hold, the paper offers a genuinely novel interpretation of LVR as the funding stream of perpetual CI options, and a concrete recipe for LPs to choose band width and position shape with predictable forward adverse-selection cost. The use of external closed forms from Ciurlia and Caperdoni and the LVR formula from Milionis et al. is transparent, and the limiting algebra in the appendix is internally consistent. The discrete replication study and the volatility calibration with real ATM IV data are useful practical complements. However, the second advertised result currently rests on an unproven and, as written, sign-inconsistent construction, and the first result is stated as an exact identity while the proof establishes only a q→∞ limit; these issues are load-bearing for the paper's title claims.","major_comments":[{"comment":"The construction of a CFAMM band whose delta equals X_q(S;K*) is unsatisfiable as stated. For a long perpetual American CI put, the delta X_q(S;K*) lies between -1 and 0 on the continuation region, whereas a CFAMM LP's delta equals the token-0 reserve x(S), which is strictly positive inside the band (as in Eq. (2) for the concentrated CPAMM). Setting V'(S)=X_q(S;K*) therefore forces a negative delta on [a,b], which no CFAMM with positive token reserves can implement. The intended object is presumably a short put whose delta is -X_q plus a cash constant, but even with that sign correction the paper must prove that the resulting reserve curves x(S) and y(S)=V(S)-S x(S) are nonnegative on the band and that they lie on a level set of some constant-function invariant F(x,y)=k. No such existence proof is supplied. Since Theorem 6, Table 1, Section 8.3, and the advertised 'constant future LVR' design rules all inherit this premise, this is a load-bearing gap.","section":"Section 6.1, Eqs. (13)–(15)"},{"comment":"The abstract states that LVR is 'analytically identical' to the CI funding fees, but the proof establishes only the limiting identity lim_{q→∞} dFee^q_t = dLVR_t. In the proof of Theorem 5, the line before 'Hence' is an explicit limit, and no finite-q equality is shown. The residual at finite q is not addressed by Section 7, which measures delta-replication error on a discrete strip, not the funding-fee residual of the continuous portfolio. Theorem 6's bound |ϵ(t)|≤rK* applies only to the special Section 6.1 construction, which itself lacks an existence proof. The theorem and abstract should either be restated as a limit result as q→∞ or be accompanied by a finite-q bound on |dFee^q_t - dLVR_t|. Because the exact identity is advertised as the first key result, this conflation of limit and equality is a central issue.","section":"Abstract and Theorem 5 (proof in Appendix B)"},{"comment":"The recursive strike construction S_ℓ(q;K_{i+1})=S_u(q;K_i) requires that the functions K↦S_ℓ(q;K) and K↦S_u(q;K) allow a finite tiling of any interval [a,b] with prescribed endpoints. The lemma asserts existence and uniqueness implicitly, but no monotonicity or surjectivity argument is given for the tiling maps. If the construction fails for some parameter ranges, the convergence argument for the discrete portfolio and the funding-fee limit would not apply to those cases. This is less severe than the two points above, but it should be addressed for rigor.","section":"Appendix B, Lemma 4"}],"minor_comments":[{"comment":"The description of boundary behavior is reversed relative to Section 2.3.2: it says 'when S_t=S_l, the option is dropped, or when S_t=S_u, it is exercised,' while the earlier definition states that S_l is the exercise boundary and S_u is the dropping boundary. Please align the two passages.","section":"Section 8.1"},{"comment":"The name 'Millionis et al. [16]' in the related-work section is a misspelling of 'Milionis'; please correct for consistency with the reference list.","section":"Section 1 and references"},{"comment":"The step-delta limit is stated with X∞(S;K) = -1_{S<K}. At the point S=K this convention gives 0. The boundary value at a single price is immaterial for the integrals, but the authors should state the convention explicitly to avoid confusion in the dominated-convergence argument.","section":"Lemma 1 and Eq. (26)"},{"comment":"The text states that 'both errors are below 10^{-3}', but the figures plot log-error and the reader cannot verify the exact numerical range from the axis labels alone; adding a brief quantitative sentence in the caption or text with the maximum observed RMSE would improve reproducibility.","section":"Section 7, Figure 4"},{"comment":"The fixed-point nature of Eq. (19) is described in words but not labeled as a fixed-point equation in the equation display. A short sentence clarifying that σ_eff^2 appears on both sides would help readers.","section":"Section 8.2, Eq. (19)"}],"recommendation":"major_revision","confidential_remarks":"The central delta-replication and limiting algebra appear sound, and the paper's first result is salvageable as an asymptotic identity. The main concern is the Section 6.1 constant-LVR construction: as written it requires a negative CFAMM delta and no existence proof is provided, so the second advertised contribution is not established. I recommend major revision rather than rejection because the flaw is local and fixable within the manuscript's scope: correct the sign convention, prove realizability of the delta profile by an actual constant-function invariant, or clearly restrict the claims. The abstract should also be aligned with the q→∞ limit that the proof actually delivers."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The core result here is legitimate and new: in the q→∞ limit, the funding fees of a delta-replicating strip of perpetual CI puts equal LVR. The limit argument is internally consistent, uses published closed forms rather than fitted parameters, and the discrete-error numerics are honest. That part of the paper is worth taking seriously.\n\nThe soft spots are real, though. The abstract and Theorem 5 state an exact equality, but the proof is only in the limit q→∞. The finite-q error analysis in Section 7 measures delta-replication error, not the funding-fee residual, so the practical claim of a predictable fee stream is only quantified for the special case. And the special case in §6.1 is worse than just approximate: it sets V'(S) equal to X_q(S;K*), the delta of a long perpetual CI put. That delta is negative in the continuation region, while any CFAMM's delta is its positive token-0 reserve. As written, Eq. (13) is unsatisfiable. You need to use a short put, i.e., delta −X_q, and then prove the implied reserve functions x(S) and y(S) are nonnegative and define a valid constant-function invariant. No such construction or existence proof appears. The empirical IV calibration uses ETH data but provides no data or code, so I cannot check that part at all.\n\nStill, the main theoretical unification—LVR as CI funding fees in the limit—appears sound and is a genuine contribution. The flaws are fixable: make the limit explicit, correct the sign in §6.1 and add a feasibility argument or soften the claim, and either release the calibration data or clearly mark it as illustrative.\n\nFor a serious referee, this paper deserves the time. The central identity is novel and the derivations are checkable. I would send it to review, but I'd expect the constant-LVR section to be challenged before acceptance.","headline":"The LVR-funding-fee identity is a real asymptotic result, but the constant-LVR construction in §6.1 has a sign problem that makes it infeasible as stated.","tokens_in":18213,"tokens_out":5329,"would_cite":true,"duration_ms":59230,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["91G20","91G80"],"pacs":[],"model":"deepseek-v4-flash","headline":"A constant-function AMM's loss-versus-rebalancing cost is exactly the funding fee stream of a delta-replicating portfolio of perpetual American continuous-installment puts, and matching a single such put's delta yields nearly constant…","keywords":["automated market maker","loss-versus-rebalancing","continuous-installment options","perpetual American options","adverse selection","liquidity provision","delta replication","implied volatility calibration"],"falsifier":"Take any concrete constant-function invariant (for example, constant-product with concentrated liquidity), compute its delta $X(S)=V'(S)$ on a band, and check whether $X'(S)$ can be made to equal $\\beta_p\\gamma_p(\\gamma_p-1)S^{\\gamma_p-2}$ for some $(r,\\sigma,K^*,q)$ over $[a,b]$ with $a=S_\\ell$, $b=S_u$; if no parameter choice matches at all price levels, the exact-replication premise fails. A second check is to simulate a GBM path, deploy the band calibrated in Section 6.1, and test whether cumulative LVR minus $qT$ ever leaves the claimed bound $rK^*$.","tokens_in":17271,"feed_emoji":"♾️","tokens_out":9036,"duration_ms":75962,"temperature":0.7,"pith_summary":"The paper establishes that loss-versus-rebalancing (LVR)—the adverse-selection cost a liquidity provider pays when an AMM's quoted price lags the market—is identical to the continuous funding fees of a portfolio of perpetual American continuous-installment (CI) put options that delta-replicates the position. As the installment rate $q$ tends to infinity, each CI put's delta collapses to a step function, so a weighted continuum of puts can replicate any smooth AMM value profile, and the funding stream from the single active option converges to the instantaneous LVR formula $\\frac12\\sigma^2S_t^2X'(S_t)\\,dt$. The paper also shows that an AMM band whose delta matches one CI put suffers an almost constant, price-path-independent LVR equal to the put's funding fee up to a bounded residual. This yields a practical rule: calibrate the band from implied-volatility term structure to obtain a predictable forward adverse-selection cost.","feed_headline":"AMM's adverse-selection cost is exactly option funding fees","feed_subtitle":"Perpetual continuous-installment option fees exactly reproduce the adverse-selection cost of AMM liquidity provision.","key_machinery":"The load-bearing object is the perpetual American continuous-installment (CI) put: an option with no expiry whose holder pays a constant funding rate $q$ until either exercising at payoff $\\max(K-S,0)$ or dropping the option, with exercise and dropping boundaries $S_\\ell(q;K)$ and $S_u(q;K)$ set by value-matching and smooth-pasting. In the limit $q\\to\\infty$ these boundaries collapse to the strike $K$, the delta becomes the step function $-\\mathbf{1}_{\\{S<K\\}}$, and the identity $\\lim_{q\\to\\infty}q(S_u-S_\\ell)=\\sigma^2K^2/2$ is what makes the active option's funding stream converge exactly to the CFAMM's instantaneous LVR. This combination converts a static, time-invariant option valuation into the dynamic, path-dependent cost of AMM liquidity provision.","core_discovery":"The central discovery is a pair of exact and approximate identities linking AMM adverse selection to CI option funding fees. Theorem 5 states that for any CFAMM position with delta $X(S)$, the delta-replicating CI-option portfolio $\\Pi$ has instantaneous funding income $dFee_t$ satisfying $\\lim_{q\\to\\infty}dFee_t = dLVR_t = \\frac12\\sigma^2S_t^2 X'(S_t)\\,dt$, and hence $Fee|_0^T = LVR|_0^T$ for every horizon $T>0$. The proof rests on the limiting identity $\\lim_{q\\to\\infty}q(S_u(q;K)-S_\\ell(q;K))=\\sigma^2K^2/2$, which converts the funding fee of the uniquely active strike into LVR. Theorem 6 then shows that if a concentrated-liquidity band is chosen with $X(S)=X_q(S;K^*)$ on $[a,b]$ with $a=S_\\ell(q,K^*)$, $b=S_u(q,K^*)$, the instantaneous LVR rate equals $q\\,dt$ plus a residual bounded by $rK^*$, so LVR over any window is approximately flat and equal to the funding fee. This recasts LVR as a tradable funding premium rather than an unhedgeable drift.","pith_inferences":["If on-chain or exchange-traded perpetual CI options became liquid, their quoted funding rates would directly reveal the market's expectation of future LVR for any AMM band, turning LVR into a price rather than a post-hoc statistic.","The exact-delta assumption in Theorem 6 is probably not strictly necessary: the Section 7 error analysis suggests that approximate matching of the single-put delta to within a small tolerance should keep LVR within a correspondingly small band around $q$; a quantitative stability theorem would be a natural follow-up.","The fixed-point volatility calibration could be applied to other AMM invariants (e.g., stableswap or log-normal curves) by replacing the CI-put delta with the corresponding gamma profile; this would extend 'constant-LVR' band design beyond concentrated constant-product pools.","Because the residual bound $|\\epsilon(t)|\\le rK^*$ grows with the strike, the constant-LVR construction is most accurate for low-strike, low-rate environments; a testable implication is that wider bands at higher rates will show LVR drift that scales with $rK^*$, observable in backtests."],"forward_implications":["A liquidity provider can, in principle, exactly hedge LVR by selling a strip of CI puts: the funding fees received equal the adverse-selection loss, leaving the hedged position delta-neutral.","With finite $q$ and finitely many strikes, the delta-replication error can be pushed below $10^{-3}$ in the tested ranges ($q\\ge 8$, $\\Delta K\\le 4$), so the theoretical decomposition is practically implementable.","By choosing a liquidity band that matches a single CI put's delta, an LP converts a stochastic, volatility-dependent adverse-selection cost into a nearly fixed cost of $q$ per unit time, with residual at most $rK^*$.","Using ATM implied-volatility term structure, an LP can solve the fixed-point equation $\\sigma^2_{\\rm eff}=w(\\bar\\tau(\\sigma^2_{\\rm eff}))/\\bar\\tau(\\sigma^2_{\\rm eff})$ to calibrate $q$ and the band boundaries to a desired expected holding period $\\bar\\tau$.","The error bounds in Theorem 8 show that when the total variance curve is approximately linear, the volatility calibration error is small, so the forward LVR estimate is robust to variation in the first-exit time."],"supporting_citations":[{"why":"Supplies the closed-form valuation, delta, and exercise/dropping boundaries of perpetual American continuous-installment puts used throughout.","marker":"[5]"},{"why":"Defines loss-versus-rebalancing and provides the instantaneous LVR formula that Theorem 5 proves equal to CI funding fees.","marker":"[16]"},{"why":"Provides the real-analysis convergence tools (dominated convergence, Leibniz rule) used to pass the q→∞ limit through the strike integral.","marker":"[11]"},{"why":"Supplies the stochastic-calculus machinery (Dynkin's formula, generator ODE) behind the closed-form expected first-exit time in Theorem 7.","marker":"[18]"}],"fun_headline_variants":["AMM LVR equals CI option funding fees","LVR is exact funding income of a perpetual option","AMM adverse selection is just option theta","New theorem: LVR identical to option funding","AMM losses recast as tradable funding premium"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument that a single CI put can be replicated by a CFAMM band assumes there exists a constant-function invariant whose delta exactly equals the put's delta at every price in the band; the paper does not prove that any concrete invariant realizes this profile, so the constant-LVR construction may be unattainable in practice.","fun_headline_variants_meta":{"raw":{"variants":["AMM LVR equals CI option funding fees","LVR is exact funding income of a perpetual option","AMM adverse selection is just option theta","New theorem: LVR identical to option funding","AMM losses recast as tradable funding premium"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000419,"raw_usage":{"total_tokens":2232,"prompt_tokens":1096,"completion_tokens":1136,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":712,"completion_tokens_details":{"reasoning_tokens":1064}},"tokens_in":712,"tokens_out":1136,"duration_ms":11603,"temperature":1.0,"reasoning_tokens":1064,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T17:37:30.022843+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take any concrete constant-function invariant (for example, constant-product with concentrated liquidity), compute its delta $X(S)=V'(S)$ on a band, and check whether $X'(S)$ can be made to equal $\\beta_p\\gamma_p(\\gamma_p-1)S^{\\gamma_p-2}$ for some $(r,\\sigma,K^*,q)$ over $[a,b]$ with $a=S_\\ell$, $b=S_u$; if no parameter choice matches at all price levels, the exact-replication premise fails. A second check is to simulate a GBM path, deploy the band calibrated in Section 6.1, and test whether cumulative LVR minus $qT$ ever leaves the claimed bound $rK^*$.","supporting_citations":[{"cited_title":"A note on the pricing of perpetual continuous-installment options","cited_arxiv_id":null,"evidence_quote":"Supplies the closed-form valuation, delta, and exercise/dropping boundaries of perpetual American continuous-installment puts used throughout."}],"review_version":2}