{"id":"8c49336e-8c44-4234-ba7e-5f0099b58fcd","arxiv_id":"2606.25771","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":6.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":0,"one_line_summary":"Introduces a pointwise estimator for time-varying Hölder exponent via geometry accumulation integral G_Lambda(t) with proven consistency, explicit noise threshold, and CLT.","lead":"The paper introduces an estimator for the local, time-varying Hurst exponent of a stochastic process using an integral that accumulates scale derivatives across resolutions. This could enable direct extraction of roughness changes from noisy price paths in rough volatility models without relying on global averages.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.3","headline":"No significant objection identified","rationale":"The reader's weakest assumption matches the scale-separation and local-asymptotics step that the abstract relies upon. With the construction internally consistent on its face and no contradictory steps visible, the UNVERDICTED status is unchanged; full proof verification would still be the next step but does not alter the current assessment.","tokens_in":1684,"tokens_out":268,"duration_ms":60030,"concrete_test":"Re-derive the leading asymptotic of E[G_Lambda(t)] or the pathwise scaling under the local Hölder assumption (without assuming global self-similarity) and confirm the coefficient depends only on H(t) at the evaluation point.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The scale-separation argument for noise robustness follows directly from balancing signal increments s^H against additive noise of size sigma, yielding the stated threshold Lambda* = sigma^{1/H}. The integral G_Lambda is dominated by its lower limit (Lambda^{H-1} term for H < 1), which supplies the local Hölder scaling even when H varies with t. The claimed CLT rate (log Lambda)^{-1/2} is consistent with logarithmic averaging across scales. No internal inconsistency appears in the stated construction or weakest assumption.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The paper introduces a pointwise estimator for the time-varying local Hölder exponent H(t) of a stochastic process X, defined via the scale-accumulation integral G_Λ(t) = ∫_Λ^1 |eth_s X(t)| s^{-1} ds with eth_s X(t) = (X(t+s) - X(t))/s. It claims to establish consistency of the resulting estimator, noise robustness to additive microstructure noise via an explicit separation threshold Λ* = σ^{1/H}, and a CLT with convergence rate (log Λ)^{-1/2}. The construction is presented as operating at finite resolution, delivering local rather than global estimates, and directly applicable to rough-volatility price paths.","tokens_in":1780,"tokens_out":509,"duration_ms":12512,"significance":"If the stated consistency, explicit threshold, and CLT hold under the paper's regularity conditions, the result would supply a practical tool for recovering time-local roughness parameters in rough-volatility models without requiring integrated-variance aggregation. The logarithmic rate and scale-separation argument for noise robustness are potentially useful strengths if the derivations are free of hidden dependence on the unknown H.","major_comments":[{"comment":"Abstract (and any corresponding theorem statement): the noise-robustness claim rests on the explicit threshold Λ* = σ^{1/H}. Because the target of estimation is precisely H, any concrete implementation appears to require either a pilot estimator or an iterative scheme; the manuscript must show that the final statistic remains asymptotically unaffected by this auxiliary step and that the CLT rate is preserved.","section":"Abstract"},{"comment":"The weakest assumption listed in the reader's note (existence of a well-defined scale derivative eth_s X(t) whose integrated absolute value yields the local Hölder scaling) is load-bearing for both consistency and the separation argument. The paper should state the precise regularity conditions on X (e.g., local Hölder continuity, moment bounds, or semimartingale properties) under which the integral G_Λ(t) is well-defined and the domination by the lower limit holds uniformly in t.","section":"Introduction / Main theorems"}],"minor_comments":[{"comment":"Notation: the symbol eth_s is introduced without an explicit definition in the abstract; a short parenthetical reminder would improve readability.","section":"Abstract"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for the insightful comments on the noise threshold implementation and regularity conditions. We address each point below and plan revisions to strengthen the manuscript.","responses":[{"response":"We agree that practical implementation requires addressing the dependence on unknown H. In the revision, we will add a section on a pilot estimator approach, where a preliminary consistent estimator of H (e.g., from integrated variance over the whole path) is used to set Λ*. We will prove that the error in the pilot estimator is negligible and does not impact the asymptotic consistency or the (log Λ)^{-1/2} rate of the CLT, under mild additional assumptions on the pilot's convergence rate.","revision_made":"yes","referee_comment":"[Abstract] Abstract (and any corresponding theorem statement): the noise-robustness claim rests on the explicit threshold Λ* = σ^{1/H}. Because the target of estimation is precisely H, any concrete implementation appears to require either a pilot estimator or an iterative scheme; the manuscript must show that the final statistic remains asymptotically unaffected by this auxiliary step and that the CLT rate is preserved."},{"response":"The manuscript currently assumes the process admits a scale derivative with the required scaling, but we concur that explicit conditions are needed. We will revise the introduction and main theorems to include a precise Assumption set: X is a continuous process with local Hölder exponent H(t) ∈ (α,1) for some α>0, with E[|X(t+s)-X(t)|^p] ≤ C s^{p H(t)} for p≥1, ensuring the integral G_Λ(t) is well-defined and the lower limit dominates uniformly in t. This will support both consistency and noise robustness.","revision_made":"yes","referee_comment":"[Introduction / Main theorems] The weakest assumption listed in the reader's note (existence of a well-defined scale derivative eth_s X(t) whose integrated absolute value yields the local Hölder scaling) is load-bearing for both consistency and the separation argument. The paper should state the precise regularity conditions on X (e.g., local Hölder continuity, moment bounds, or semimartingale properties) under which the integral G_Λ(t) is well-defined and the domination by the lower limit holds uniformly in t."}],"tokens_in":1371,"tokens_out":497,"duration_ms":22307,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"The main takeaway is a construction for local H(t) that integrates the absolute scale derivative against s^{-1} ds to produce G_Lambda(t), then extracts the exponent from its scaling. This moves beyond the usual global integrated-variance estimators and supplies an explicit noise cutoff Lambda* = sigma^{1/H} plus a CLT at the expected (log Lambda)^{-1/2} rate.\n\nThe approach is straightforward: the lower limit of the integral dominates for H < 1, so the local Hölder scaling comes through even when H varies with t, and the scale-separation argument for robustness follows directly from balancing s^H increments against additive noise. That part checks out without circularity in the abstract statement.\n\nThe soft spot is practical use of the threshold. Because Lambda* depends on the unknown H, any implementation needs a pilot or iteration whose impact on the final statistic and on the CLT is not visible from the given material. The paper would need to show that this step does not destroy the asymptotics or introduce extra bias. The regularity conditions on the scale derivative are also left implicit.\n\nThis is aimed at people doing high-frequency rough-volatility work who need time-local roughness rather than a single number. The claims are specific enough to referee. I would send it for review; the idea is distinct and the stress-test finds no load-bearing inconsistency.","headline":"The paper gives a pointwise Hurst estimator via scale accumulation that separates noise at an explicit threshold and looks internally consistent on the stated terms.","tokens_in":2269,"tokens_out":351,"would_cite":false,"duration_ms":10761,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.3","headline":"The geometry accumulation integral of scale derivatives yields a consistent pointwise estimator of the local Hurst exponent that filters microstructure noise above an explicit threshold.","keywords":["pointwise Hurst estimation","Hölder exponent","scale derivative","rough volatility","microstructure noise","consistency","central limit theorem","geometry accumulation"],"falsifier":"Compute the estimator on simulated paths with known local H and added noise of size sigma; it must fail to converge to the true H or lose the claimed rate once Lambda is set below sigma to the power 1/H.","tokens_in":2558,"feed_emoji":"📈","tokens_out":843,"duration_ms":21765,"temperature":0.7,"pith_summary":"The paper introduces an estimator for the time-varying Hölder exponent H(t) of a stochastic process by integrating the absolute scale derivative against ds/s from a lower cutoff Lambda to 1. It proves that this integral diverges like -log Lambda times a factor depending on H, which permits consistent inversion to recover H(t) directly from the observed path at finite resolution. The construction separates additive noise of size sigma by restricting Lambda above sigma to the power 1/H, and a central limit theorem holds at the rate one over square root of log Lambda. Unlike global estimators based on integrated variance, the method produces a localized H(t) that can be read off price paths without first removing noise by other means.","feed_headline":"Scale accumulation gives consistent pointwise Hurst estimates","feed_subtitle":"The integral of absolute scale derivatives recovers local H(t) from noisy paths above threshold sigma to the 1/H and obeys a CLT at rate one","key_machinery":"The geometry accumulation integral G_Lambda(t), which sums the absolute scale derivatives eth_s X(t) weighted by ds/s and thereby encodes the local Hölder exponent through its logarithmic divergence rate.","core_discovery":"For a process whose local regularity is governed by a Hölder exponent H, the geometry accumulation integral G_Lambda(t) = integral from Lambda to 1 of |eth_s X(t)| s^{-1} ds satisfies G_Lambda(t) ~ c(H) (-log Lambda) as Lambda tends to zero, where eth_s X(t) denotes the forward difference quotient at scale s; inverting this relation produces a consistent estimator of H(t). The same integral remains asymptotically unaffected by additive noise of amplitude sigma provided Lambda exceeds sigma^{1/H}, and the normalized fluctuation around the mean converges in distribution at rate (log Lambda)^{-1/2}.","pith_inferences":["The same scale-separation principle could be applied to detect changes in local regularity over time in real-time financial data streams.","Extensions to multivariate or jump-augmented processes would require only that the scale derivative still isolate the Hölder component.","Direct comparison on high-frequency tick data against wavelet or increment-ratio estimators would test whether the integral form reduces computational cost while retaining the explicit noise threshold.","If the local H(t) varies smoothly, the estimator could serve as input to adaptive rough-volatility pricing models that adjust to intraday roughness."],"forward_implications":["The estimator converges in probability to the true local H(t) as the lower scale cutoff Lambda tends to zero.","Consistency is preserved under additive microstructure noise whenever the cutoff satisfies Lambda greater than sigma to the power 1/H.","A central limit theorem holds for the estimator with convergence rate (log Lambda)^{-1/2}.","The procedure operates directly on discrete observations at finite resolution without requiring a preliminary denoising step.","It recovers a time-localized function H(t) rather than a single global parameter extracted from integrated variance."],"fun_headline_variants":["Scale accumulation estimates pointwise Hurst","Geometry integral yields local H(t) estimates","Pointwise H(t) from scale derivative accumulation","Noise-robust Hurst via scale separation threshold","Accumulated scales recover consistent time-varying H"],"cache_read_input_tokens":64,"weakest_assumption_plain":"The process must possess a scale derivative whose absolute value, when integrated against ds/s, produces an asymptotic that is controlled solely by the local Hölder exponent and can be cleanly separated from additive noise at the stated threshold.","fun_headline_variants_meta":{"raw":{"variants":["Scale accumulation estimates pointwise Hurst","Geometry integral yields local H(t) estimates","Pointwise H(t) from scale derivative accumulation","Noise-robust Hurst via scale separation threshold","Accumulated scales recover consistent time-varying H"]},"model":"grok-4.3","cost_usd":0.004037,"raw_usage":{"total_tokens":2044,"prompt_tokens":643,"num_sources_used":0,"completion_tokens":63,"cost_in_usd_ticks":40374500,"prompt_tokens_details":{"text_tokens":643,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":1338,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":643,"tokens_out":63,"duration_ms":10216,"temperature":1.0,"reasoning_tokens":1338,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-06-25T19:40:37.050965+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"Compute the estimator on simulated paths with known local H and added noise of size sigma; it must fail to converge to the true H or lose the claimed rate once Lambda is set below sigma to the power 1/H.","supporting_citations":[],"review_version":1}