{"id":"cd126eee-4a56-4199-8355-306745ef1031","arxiv_id":"2502.07518","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"high","formal_verification":"none","parameter_count":6,"one_line_summary":"A new moneyness-dependent Hurst exponent is inserted into an implied volatility formula that is claimed to beat SABR and fSABR, but the key H=1/2 at-the-money result is built into the formula rather than discovered.","lead":"This paper proposes a new formula for implied volatility that lets a memory parameter change with how far an option's strike price is from the stock price, and reports that it fits recent U.S. options data better than two standard models. It matters because the authors claim the formula is free of arbitrage and can be read as a measure of market efficiency, but the supporting proof and tests have important gaps.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The free-arbitrage certificate in Annex 4.3(iii) is invalid: the proof assumes the underlying can only stay flat or rise, which is false for any non-degenerate risk-neutral martingale, and the mBm martingale premise is not established.","rationale":"The reader's weakest assumption identifies exactly this node: the no-arbitrage proof depends on a martingale/monotonicity premise. The manuscript contains an explicit false statement (ST can only stay same or increase), an omitted discount in Eq. (21), and no derivation of the mBm martingale property despite relying on it for conditions iii-v. I also checked the other claimed support: Table 1 is an in-sample Optuna fit with no code, data, significance tests, or out-of-sample split, so the outperformance claim is weak; but the theoretical 'free arbitrage' claim is the more fundamental advertised result. The H=1/2 ATM property is built into Eq. (5), so it cannot validate the model. I find no independent support that would rescue the certificate: no machine-checked proof, no numerical arbitrage scan, and the parameter constraints in Section 4.2 are derived only from first-derivative signs, not from convexity of call prices. Therefore the reader's REJECT is correct and should stand.","tokens_in":14423,"tokens_out":10237,"duration_ms":97894,"concrete_test":"Derive the conditional expectation for the mBm specification of Di Sciorio [10] (or any H(t) path used in calibration) and check whether e^{-r(T-t)} E[ST | Ft] = St for every t0. The valid lower bound would need Jensen applied to a true martingale, not pathwise domination. The paper's martingale premise is also not established: it asserts that H(t)->1/2 as t->T for the mBm 'ensures St is a martingale,' but convergence at the terminal time does not imply H(t)=1/2 on [t0,T), and fBm/mBm with H != 1/2 is not a semimartingale; the cited [10] is not a proof of the needed property. Eq. (21) additionally drops the discount factor present in Eq. (12). Condition (i), convexity of C in K, is only asserted via sign conditions in Eq. (17) and a 'computational study' (Fig. 4) with no proof over the parameter domain. Since the no-arbitrage property is the advertised contribution, this is the load-bearing failure.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a closed-form implied volatility (IV) model, called the AdS model, in which IV is written as a function of moneyness S/K, with a moneyness-dependent Hurst exponent H(S/K). The functional form is designed so that H(1)=1/2 and H decreases away from the money. The authors calibrate the model with Optuna on 30-day option data for many U.S. indices and stocks, compare its in-sample fit against SABR and fSABR using MSE, MAE, and curvature errors, and claim that the resulting volatility surface satisfies the five no-arbitrage conditions of Zaugg et al. The paper also claims an empirical regularity that H peaks at the at-the-money point and that this is a sign of market efficiency there.","tokens_in":14732,"tokens_out":4954,"duration_ms":50158,"significance":"If the claims were established, the paper would offer a very simple, closed-form IV surface with an explicit moneyness-dependent Hurst exponent, calibratable by standard global optimization. The explicit formula and the broad empirical exercise across many tickers are useful ingredients, and the use of Optuna with a clearly stated objective is a reproducible calibration strategy. However, the central advertised contribution is the no-arbitrage certificate, and that certificate is not valid: the proof of condition (iii) in Annex 4.3 relies on a false pathwise monotonicity statement and drops the discount factor, the martingale premise for multifractional Brownian motion is asserted rather than established, and the convexity condition (i) is only checked computationally rather than proved. In addition, the headline 'H=1/2 at the money' is built into the definition of H, not discovered from data, and the empirical comparison is in-sample and not uniformly favorable to the proposed model. The significance is therefore conditional on a substantial reworking of the proof and of the empirical design.","major_comments":[{"comment":"The proof of the lower bound (s-K)+ <= C(T,K;p) is invalid. The text states that '(ST-K)+ >= (s-K)+ because the underlying price ST can only stay the same or increase over time.' This is false for any non-degenerate risk-neutral martingale: a non-constant martingale takes values below its starting point with positive probability. Moreover, the proof first writes C(T,K;p)=e^{-r(T-t)}E^Q[(ST-K)+] and then, two lines later, uses C(T,K;p)=E^Q[(ST-K)+], silently dropping the discount factor. Even with a correct Jensen argument, the discount factor must be handled, and the constant s must be the initial spot price. As written, condition (iii) is not proved, and the no-arbitrage certificate collapses at this point.","section":"Annex 4.3, condition (iii), Eq. (21)"},{"comment":"The martingale premise for the underlying is asserted, not established. The paper states that lim_{t->T} H(t)=1/2 for a multifractional Brownian motion 'ensures that St is a martingale,' but convergence of the Hurst parameter only at the terminal time does not imply H(t)=1/2 on [t0,T), and fractional Brownian motion with H != 1/2 is not a semimartingale. The cited reference [10] is not a proof of the required property within the manuscript. Since the risk-neutral pricing formula in condition (iii) requires a genuine martingale under Q, this is a load-bearing gap.","section":"Annex 4.3, after condition (iv)"},{"comment":"The convexity of the call price in K is not proved. Section 4.2 derives conditions on the first derivative of sigma with respect to 1/K, and Eq. (17) only imposes monotonicity of sigma. The proof of condition (i) then states a second-derivative inequality and refers to 'the computational study conducted' in Figure 4. A numerical check over a finite grid cannot certify convexity over the whole parameter domain, and no analytic argument for C_KK>0 is provided. This matters because the calibration domain D in Eq. (7) is asserted to be compatible with the no-arbitrage conditions, but that compatibility is not demonstrated.","section":"Annex 4.3, condition (i)"},{"comment":"The claim that H approaches 1/2 when moneyness equals 1 is a consequence of the chosen functional form, not an independent empirical discovery. In Eq. (5), H(S/K) is constructed with a normalization factor that forces H(1)=1/2, and the denominator makes H decrease away from S/K=1. The 'inverse smile effect' is therefore baked into the model specification. The empirical Figure 1 plots the fitted implied regularity, but the paper does not provide an independent estimate of H that would test the functional form against market data.","section":"Eq. (5) and Section 2, Figure 1"},{"comment":"The empirical comparison does not support the broad claim of outperformance. First, all reported errors are in-sample calibration errors; there is no out-of-sample or cross-validation analysis. Second, Table 2 shows that AdS is not uniformly better than SABR on the per-ticker MSE: for example, the MA, MCD, MSFT, ACN, and UNH rows have lower SABR MSE than AdS MSE. Third, the model contains the extra data-dependent parameter Kmin, which is absent from the search space D in Eq. (7), and no estimation protocol for Kmin is given. Since the location of the minimum of sigma strongly affects the shape of the fitted curve, the comparison with SABR/fSABR is not on equal footing.","section":"Section 2.2, Table 2 and Eq. (7)"}],"minor_comments":[{"comment":"The abstract contains the typo 'pratcical' instead of 'practical.'","section":"Abstract"},{"comment":"The sentence 'Let the option price of a call C(t,K;sigma)' is missing a verb; it should read 'Let the option price of a call be C(t,K;sigma).'","section":"Section 4, Eq. (12)"},{"comment":"The symbol C is used both for the scale parameter in the log-linear regression log sigma(n)=log C - H log n and for the call price C(T,K;p) elsewhere; this notation is confusing and should be changed.","section":"Section 1.1, Eq. (6)"},{"comment":"The derivation of the admissible interval for beta and the argument for choosing 0 infinity needs a precise statement, and the positivity condition on the second member of Eq. (19) should be proved rather than asserted.","section":"Section 4.2, Eq. (19)-(20)"},{"comment":"The panels in Figure 4 lack axis labels and a description of the computational grid; the caption should state the parameter ranges, the number of sampled points, and the rule used to declare a condition satisfied.","section":"Figure 4"}],"recommendation":"reject","confidential_remarks":"The paper's advertised contribution is the no-arbitrage certification of the proposed IV surface. That certification fails at a basic probabilistic point in Annex 4.3, and the supporting martingale premise is not established. The headline empirical regularity is built into the model's normalization, and the performance claim is in-sample. I do not see a local fix that would preserve the central claims as stated; the authors would need to rework the proof strategy, provide a genuine martingale model for the underlying, and redo the empirical analysis with out-of-sample or cross-validated comparisons and a calibrated Kmin. For a quantitative-finance journal, the current manuscript is not suitable for publication."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper's real content is the Angelini-di Sciorio (AdS) formula: a closed-form implied volatility surface in S/K with a moneyness-dependent Hurst exponent. That parametric family is new, explicit, and easy to calibrate, and the across-ticker fitting comparison with SABR/fSABR is a reasonable first look. The idea that H is pinned to 1/2 at the money is a useful normalization, but it is imposed by the functional form in Eq. (5), not discovered in the data. The empirical gains are modest—mean MSE improves from about 0.68 to 0.50—and they come from in-sample fits with no code, no data, and no significance tests. One entry in Table 2 (SPX fSABR MAE of 5.82·10^3) is obviously a typo, which does not inspire confidence in the cleaning pipeline.\n\nThe soft spot is the load-bearing one: the arbitrage-free certificate in Annex 4.3. The proof of condition iii says `ST can only stay the same or increase over time`, which is false for any non-degenerate risk-neutral martingale. The lower bound (s-K)+ ≤ C therefore does not follow from that argument, and Eq. (21) even drops the discount factor present in Eq. (12). The martingale premise is asserted, not established: H(t)→1/2 at maturity does not make the mBm a martingale on [t0,T), and the cited [10] does not prove the needed property. Condition i, convexity of the call price in K, is backed by a computational figure rather than an analytic check over the parameter domain, and condition iv is asserted more than proven. Kmin is a free, data-dependent parameter sitting outside the stated search space D; it needs to be calibrated or profiled, not treated as known.\n\nI agree with the reader's reject verdict and with the stress-test note. The arbitrage claim is the advertised contribution, and it collapses. But the parametric model itself is a defensible standalone contribution: a simple, flexible smile formula that practitioners can test. The paper deserves a serious referee, not because the current proof works, but because the formula and the empirical comparison are concrete enough that a referee can tell the authors to either prove the arbitrage conditions rigorously or drop the arbitrage language and reframe the paper as a descriptive parametric study with out-of-sample validation. The authors are clearly capable of that; the current draft is just not there yet.","headline":"A genuinely new closed-form IV smile with moneyness-dependent H, wrapped around a no-arbitrage proof that fails at a false monotonicity claim and should be cut or fixed before this is publishable.","tokens_in":15257,"tokens_out":3399,"would_cite":false,"duration_ms":33396,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["91G20","91G60","60G22"],"pacs":[],"model":"deepseek-v4-flash","headline":"A closed-form implied volatility curve that pins the Hurst exponent to 1/2 at the money and beats SABR on fit.","keywords":["implied volatility","Hurst exponent","moneyness","arbitrage-free volatility surface","SABR model","fractional Brownian motion","volatility smile","Optuna calibration"],"falsifier":"Compute the call-price bounds $(s-K)^+ \\le C(T,K) \\le s$ on any market-observed surface: pick $s$ = current spot, take a date when the market opens below $s$ (i.e., any downward day), and check whether a deep out-of-the-money call struck above the previous close prices below $(s-K)^+$; a single such violation refutes property (iii). Alternatively, directly test whether the fitted surface passes the butterfly-arbitrage condition $C_{KK} \\ge 0$ across all strikes on the 30-day slices used in Table 1.","tokens_in":14214,"feed_emoji":"📊","tokens_out":6099,"duration_ms":50228,"temperature":0.7,"pith_summary":"This paper proposes the Angelini-di Sciorio (AdS) model, a four-parameter closed-form expression for implied volatility as a function of moneyness $S/K$, with the Hurst exponent $H$ appearing explicitly inside the formula. The model's defining claim is that implied regularity peaks at the at-the-money point, $H(S/K=1)=1/2$, and decays into the in-the-money and out-of-the-money regions, an inverse smile the authors tie to deviations from market efficiency. They argue that when this volatility curve is plugged into the Black-Scholes pricing formula, the resulting call prices satisfy the five conditions for an arbitrage-free volatility surface, and they report that Optuna-calibrated fits to 30-day option data across multiple indexes beat SABR and fSABR on accuracy and curvature error metrics. If correct, the paper supplies a parsimonious, parameterizable, arbitrage-free implied volatility surface that also encodes a moneyness-dependent regularity.","feed_headline":"Closed-form IV model beats SABR and fSABR on real option data","feed_subtitle":"A four-parameter curve links moneyness to an inverse-smile Hurst exponent and claims an arbitrage-free surface.","key_machinery":"The load-bearing object is the moneyness-dependent Hurst function $H(S/K)$, a normalized inverse-smile shape that forces $H=1/2$ exactly at $S/K=1$ and dips toward $0$ in both wings, governed by the shape parameter $\\delta$. It enters the volatility curve through the exponential decay term $e^{-\\beta H(S/K)(S/K-S/K_{\\min})}$, so that memory strength modulates how quickly implied volatility falls as moneyness moves away from the minimum-volatility strike; the quadratic prefactor $(S/K-S/K_{\\min})^2$ supplies the smile's concavity. The parameter constraints $\\alpha>0$, $\\epsilon>0$, $\\beta\\in[-1,1]$, $\\delta\\in(0,1)$ are derived from concavity and convexity requirements on $H$ and $\\sigma$, and the arbitrage-free certification then rides on standard Black-Scholes derivatives plus the martingale premise.","core_discovery":"The central claim is that implied volatility can be represented without a stochastic-volatility engine: a single closed formula $\\sigma(S/K)=\\alpha(S/K - S/K_{\\min})^2 e^{-\\beta H(S/K)(S/K - S/K_{\\min})} + \\epsilon$, with $H(S/K)=\\frac{1}{2}(1+|1-S/K_{\\min}|^\\delta)/(1+|S/K-S/K_{\\min}|^\\delta)$, reproduces the smile while satisfying butterfly- and calendar-arbitrage conditions when $\\alpha, \\epsilon>0$, $\\beta\\in[-1,1]$, $\\delta\\in(0,1)$. The paper's own verification walks through Zaugg's five conditions in Annex 4.3, with the martingale property imported from a multifractional Brownian motion setup in which $H(t)\\to 1/2$ at maturity. The empirical section claims that over the fitted 30-day slices, the model's MSE, MAE, curvature-error and absolute-curvature-error metrics are smaller on average than those of SABR and fSABR, with the advantage most pronounced in the ITM and OTM tails.","pith_inferences":["Editorial inference: the same functional form could be applied to VIX futures or variance swaps, where an implied-Hurst curve is less standard, to see whether the inverse-smile regularity pattern is a feature of equity index options or a broader market property.","Editorial inference: the curvature error metrics introduce a second-derivative target; a natural next step is to test whether calibration to ACE and RMSCE changes the fitted parameters in a way that degrades raw price fit, i.e., whether the curvature advantage is bought at the cost of price accuracy.","Editorial inference: because the model is closed-form with only four parameters, it could be embedded directly into an SVI-type parametrization or used as a prior in a neural-network calibration, reusing the invertible moneyness-to-H map as a structural regularizer."],"forward_implications":["If the claim holds, one can generate an arbitrage-free volatility smile from four parameters, removing the need for numerical simulation or density recovery in smile construction.","The $H=1/2$-at-the-money constraint gives a model-free anchor: any calibration must pinch the curve to the efficient-market value at $S=K$, and deviations in the wings are explicitly interpreted as memory and market-inefficiency effects.","The parameter restrictions ($\\beta\\in[-1,1]$, $\\delta\\in(0,1)$) come out of convexity demands, so fitting within these bounds automatically keeps the surface free of static arbitrage.","Across the 42-ticker test set, the reported curvature-error metrics suggest the functional form captures the second derivative of the smile better than SABR and fSABR in extreme moneyness regions.","The closed form is cheap enough for repeated recalibration, which matters for desks that re-fit surfaces intraday."],"supporting_citations":[{"why":"Supplies the multifractional Brownian motion risk-neutral framework that underpins the martingale premise used in the arbitrage-free proof.","marker":"[10]"},{"why":"Supplies the five no-arbitrage conditions that the AdS surface is verified against.","marker":"[31]"},{"why":"Connects implied Hurst exponent to fBm and moneyness, motivating the H(S/K) form and the H=1/2 at ATM observation.","marker":"[12]"},{"why":"Provides the Optuna optimization package used to calibrate all three models.","marker":"[3]"},{"why":"Defines the SABR model, the baseline for comparison.","marker":"[17]"},{"why":"Defines the lognormal fSABR model, the second baseline.","marker":"[2]"},{"why":"Gives the method for estimating the implied regularity H from option data via the self-similarity relationship between sigma and H.","marker":"[4]"},{"why":"Provides the estimation of the scale parameter C used in the H estimation.","marker":"[6]"}],"fun_headline_variants":["Hurst-aware IV model beats SABR and fSABR","Closed-form smile with Hurst twist outperforms SABR","Moneyness-to-Hurst curve yields arbitrage-free IV model","Four-parameter IV formula edges out SABR and fSABR","Hurst exponent cracks IV smile, beats SABR models"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The arbitrage-free certification rests on the premise that the underlying price process is a martingale under the pricing measure, imported from a multifractional Brownian motion framework in which the Hurst parameter tends to 1/2 at maturity, and on a pathwise monotonicity claim about the underlying never falling below its starting level.","fun_headline_variants_meta":{"raw":{"variants":["Hurst-aware IV model beats SABR and fSABR","Closed-form smile with Hurst twist outperforms SABR","Moneyness-to-Hurst curve yields arbitrage-free IV model","Four-parameter IV formula edges out SABR and fSABR","Hurst exponent cracks IV smile, beats SABR models"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000655,"raw_usage":{"total_tokens":2993,"prompt_tokens":930,"completion_tokens":2063,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":546,"completion_tokens_details":{"reasoning_tokens":1972}},"tokens_in":546,"tokens_out":2063,"duration_ms":14670,"temperature":1.0,"reasoning_tokens":1972,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-08T12:28:26.137129+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the call-price bounds $(s-K)^+ \\le C(T,K) \\le s$ on any market-observed surface: pick $s$ = current spot, take a date when the market opens below $s$ (i.e., any downward day), and check whether a deep out-of-the-money call struck above the previous close prices below $(s-K)^+$; a single such violation refutes property (iii). Alternatively, directly test whether the fitted surface passes the butterfly-arbitrage condition $C_{KK} \\ge 0$ across all strikes on the 30-day slices used in Table 1.","supporting_citations":[{"cited_title":"Di Sciorio","cited_arxiv_id":null,"evidence_quote":"Supplies the multifractional Brownian motion risk-neutral framework that underpins the martingale premise used in the arbitrage-free proof."},{"cited_title":"Zaugg, A","cited_arxiv_id":null,"evidence_quote":"Supplies the five no-arbitrage conditions that the AdS surface is verified against."},{"cited_title":"Flint and E","cited_arxiv_id":null,"evidence_quote":"Connects implied Hurst exponent to fBm and moneyness, motivating the H(S/K) form and the H=1/2 at ATM observation."},{"cited_title":"Akiba, S","cited_arxiv_id":null,"evidence_quote":"Provides the Optuna optimization package used to calibrate all three models."},{"cited_title":"Hagan, A","cited_arxiv_id":null,"evidence_quote":"Defines the SABR model, the baseline for comparison."},{"cited_title":"Akahori, X","cited_arxiv_id":null,"evidence_quote":"Defines the lognormal fSABR model, the second baseline."},{"cited_title":"Angelini and S","cited_arxiv_id":null,"evidence_quote":"Gives the method for estimating the implied regularity H from option data via the self-similarity relationship between sigma and H."},{"cited_title":"Bianchi, A","cited_arxiv_id":null,"evidence_quote":"Provides the estimation of the scale parameter C used in the H estimation."}],"review_version":1}