{"id":"14f09470-0dfe-4faf-a568-58757f7c7d46","arxiv_id":"2611.16790","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"The paper derives quasi-likelihood estimators for drift parameters in a singular SDE with Gaussian noise and a strongly consistent estimator of the mean-reversion ratio for a broad class of Gaussian processes.","lead":"This paper develops statistical methods for estimating drift parameters in a stochastic process with singular mean-reverting drift and Gaussian noise. The work provides explicit estimators for both drift coefficients under Riemann-Liouville fractional Brownian noise and a strongly consistent estimator of the mean-reversion ratio for a broader class of Gaussian processes.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek/deepseek-v4-flash","headline":"Quasi-likelihood estimator assumes known H and σ; the paper does not address how to estimate them or quantify the impact of misspecification.","rationale":"The reader's verdict identifies the same concern: 'The quasi-likelihood estimation in Section 3 relies on the assumption that the Hurst parameter H and the volatility sigma are known. In practice, these would need to be estimated, which could introduce additional uncertainty.' The paper itself does not address this issue at all—it is not mentioned in the limitations or discussion. The simulation study fixes H and σ, so it provides no evidence on the robustness of the estimators to misspecification. This is a load-bearing concern because the central claim of Section 3 is that the estimators are explicit and closed-form, but their practical value depends on whether H and σ can be reliably estimated or whether the estimators are robust to misspecification. The paper does not provide any guidance or analysis on this point. Therefore, the concern is valid and significant. The verdict should remain CONDITIONAL, as the reader already noted this weakness. The concrete test would provide evidence on whether this concern actually lands in practice.","tokens_in":25824,"tokens_out":619,"duration_ms":6768,"concrete_test":"Perform a simulation study where H and σ are estimated from the data (e.g., via the method of moments or maximum likelihood based on the quadratic variation for σ, and via the scaling behavior of increments for H) before plugging them into the quasi-likelihood estimators. Compare the resulting bias and RMSE of the drift estimates to the case where H and σ are known. If the additional estimation error substantially degrades performance (e.g., RMSE increases by more than 50% for some parameter configurations), then the claim that the estimators are practically useful is weakened.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central claim of Section 3 is that the quasi-likelihood estimators (29) provide explicit closed-form estimators of a and b. However, the construction assumes that both the Hurst parameter H and the volatility σ are known (see sentence [b86]: 'where σ > 0 and H ∈ (1/2, 1) are assumed to be known'). In practice, these parameters are unknown and would need to be estimated from the same data. The paper does not discuss how to estimate H and σ, nor does it analyze the sensitivity of the drift estimators to misspecification of these nuisance parameters. The simulation study fixes H and σ at their true values, so it does not reveal the impact of estimation error in H and σ on the drift estimates. Without such analysis, the practical applicability of the quasi-likelihood estimators is severely limited. The reader's verdict already flags this as a weakness, but the paper itself does not address it at all—it is not merely an acknowledged limitation but a missing piece that undermines the claim of a complete estimation procedure.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek/deepseek-v4-flash","summary":"The paper studies drift parameter estimation for a positive stochastic process with singular mean-reverting drift driven by Gaussian noise. For Riemann-Liouville fractional Brownian noise with known Hurst parameter H>1/2, the authors apply an observable fractional transformation to obtain a semimartingale with drift linear in the unknown parameters, then use the quasi-likelihood approach to derive explicit closed-form estimators of both drift coefficients a and b. For a broader class of Gaussian driving processes satisfying local Hölder regularity >1/2 and a sublinear growth condition, they construct a strongly consistent estimator of the ratio θ=a/b. The theoretical results are complemented by a simulation study investigating the effects of observation horizon and discretization step on finite-sample accuracy.","tokens_in":26116,"tokens_out":1483,"duration_ms":11958,"significance":"If the results hold, the paper provides useful estimation methods for a class of stochastic differential equations with singular drift and Gaussian noise, which are relevant in financial modeling (e.g., fractional CIR-type models). The quasi-likelihood approach avoids the need for Girsanov-type absolute continuity conditions, and the ratio estimator applies to a broad class of Gaussian processes. The simulation study is thorough and reveals important finite-sample biases that depend on horizon and discretization step, which the authors analyze. However, the practical applicability is limited by the assumption that H and σ are known for the quasi-likelihood estimators, and the paper does not address estimation of these nuisance parameters or sensitivity to misspecification.","major_comments":[{"comment":"The quasi-likelihood estimators (29) are derived under the assumption that the Hurst parameter H and the volatility σ are known (sentence [b86]: 'where σ > 0 and H ∈ (1/2, 1) are assumed to be known'). In practice, these parameters are unknown and would need to be estimated from the same data. The paper does not discuss how to estimate H and σ, nor does it analyze the sensitivity of the drift estimators to misspecification of these nuisance parameters. The simulation study fixes H and σ at their true values, so it does not reveal the impact of estimation error in H and σ on the drift estimates. Without such analysis, the practical applicability of the quasi-likelihood estimators is severely limited. This is not merely an acknowledged limitation but a missing piece that undermines the claim of a complete estimation procedure.","section":"Section 3, Eq. (7) and surrounding text","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The paper identifies a bias that depends linearly on the discretization step δ and a horizon-specific component ca(T), cb(T). However, the analysis of this bias is purely empirical; there is no theoretical derivation or explanation of the bias structure. The claim that the bias can be separated into two components with ka, kb < 0 is based on simulation observations but is not supported by any theoretical analysis. This is a significant gap because the bias is not negligible even for small δ, and the paper does not provide any guidance on how to correct for it or how to choose δ to control it.","section":"Section 5.2, Eq. (34) and Table 3","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The ratio estimator ˆθ_T is shown to be strongly consistent under the assumption that G is λ-Hölder continuous for some λ > 1/2 and that G*_t / t → 0 a.s. as t → ∞. While several examples are given, the paper does not provide any rate of convergence or asymptotic distribution for ˆθ_T. The simulation study shows that the estimator can have substantial bias for moderate T, especially for small a and large H. Without a theoretical understanding of the convergence rate, it is difficult to assess the practical utility of the estimator or to construct confidence intervals.","section":"Section 4, Theorem 5","authors_acknowledge":null,"authors_acknowledge_p":null}],"minor_comments":[{"comment":"The abstract states 'an observable fractional transformation reduces the model to a semimartingale with drift linear in the unknown parameters.' It would be helpful to mention that this transformation requires knowledge of H.","section":"Abstract","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The definitions of Y_t and Z_t involve fractional derivatives. It would be helpful to explicitly state that these are well-defined under the Hölder continuity assumptions on X and X^{-1}, which are justified in the text, but a brief remark connecting to the earlier regularity results would improve readability.","section":"Section 3.1, Eq. (8)-(9)","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The description of the Lubich convolution quadrature and the recursion for w_k is somewhat terse. A reference to a standard source for the BDF2 generating function would be helpful for readers unfamiliar with the method.","section":"Section 5.1, paragraph on fractional derivative","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The analysis of bias divided by δ is interesting, but the interpretation of the limiting values ka and kb as 'the leading terms of the discretization error' is not clearly justified. The paper should either provide a theoretical derivation or acknowledge that this is an empirical observation.","section":"Section 5.2, Figure 5 and Table 3","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The relative bias and RMSE are reported in percent, which is appropriate. However, the standard errors are also reported in percent, which is fine, but the table would benefit from a note explaining that the standard errors are for the relative bias, not for the RMSE.","section":"Section 5.4, Table 4","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"Reference [10] is listed as 'Volterra volatility models: Option pricing and hedging, Springer Finance, Springer, Cham, 2026.' This appears to be a forthcoming book; if it is not yet published, the reference should be updated to reflect its status (e.g., 'to appear' or 'in press').","section":"References","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The paper uses the notation ~X for the auxiliary process in Section 4.1, but the tilde appears as a combining character in the PDF, which may cause readability issues. Consider using a different notation, such as X^{(0)} or X^0.","section":"General","authors_acknowledge":null,"authors_acknowledge_p":null}],"recommendation":"major_revision"},"author_rebuttal":null,"desk_editor":{"model":"deepseek/deepseek-v4-flash","letter":"This paper gives two estimation methods for drift parameters in a CIR-type SDE driven by Gaussian noise. The first is a quasi-likelihood estimator for Riemann–Liouville fBm with known H and σ, derived via a fractional transformation that yields a Brownian semimartingale. The second is a ratio estimator for a/b that works under much weaker assumptions (local Hölder > 1/2 and sublinear growth) and is proved strongly consistent. The math is rigorous, the simulation study is thorough and honest about biases, and the paper is well written.\n\nThe soft spot is the one the stress-test flags: the quasi-likelihood estimator assumes H and σ are known, and the paper does not discuss how to estimate them or what happens if they are misspecified. The authors state this assumption clearly, so they are not hiding it, but they also do not address it. For a reader who wants to use these estimators on real data, this is a real gap. The ratio estimator does not have this problem—it only needs continuous observation—so that part is more practically useful.\n\nA minor point: the simulation study for the quasi-likelihood estimator uses a burn-in of 0.5 and discards early data, which is sensible but means the estimator is not really using the full trajectory. This is not a flaw, but worth noting since the theoretical derivation assumes observation from 0.\n\nWho is this for? Researchers working on statistical inference for fractional SDEs, especially CIR-type models with memory. The ratio estimator is the more broadly applicable result; the quasi-likelihood part is a nice theoretical contribution but less ready for practice.\n\nRecommendation: This paper deserves peer review. The math is solid, the ratio estimator is a clean result, and the simulation study is transparent. The missing discussion of H and σ estimation is a limitation that should be addressed in revision, but it is not fatal. I would send it to a good statistics journal.","headline":"Solid inference for a singular SDE: a clean ratio estimator and a quasi-likelihood construction, the latter assuming known H and σ.","tokens_in":27179,"tokens_out":564,"duration_ms":10643},"pith_extraction":{"msc":["62M09","62F12","60H10","60G22"],"pacs":[],"model":"deepseek/deepseek-v4-flash","headline":"Closed-form estimators for drift in singular SDEs with Gaussian noise","keywords":["singular stochastic differential equation","Riemann-Liouville fractional Brownian motion","quasi-likelihood estimation","drift parameter estimation","Gaussian processes","mean-reversion","Cox-Ingersoll-Ross model","ratio estimator"],"falsifier":"Simulate data from the SDE (1) with a known Gaussian driving process that does NOT satisfy the Hölder condition λ > 1/2 (e.g., standard Brownian motion with H=1/2) and check whether the ratio estimator ˆθ_T still converges to a/b; if it does, the Hölder condition is not necessary, and if it does not, the condition is essential.","tokens_in":27319,"feed_emoji":"📈","tokens_out":1471,"duration_ms":11817,"temperature":0.7,"pith_summary":"This paper develops two complementary approaches for estimating the drift parameters of a positive stochastic process with singular mean-reverting drift driven by Gaussian noise. The first approach, applicable when the noise is a Riemann-Liouville fractional Brownian motion with known Hurst parameter, uses a fractional transformation to convert the model into a Brownian semimartingale, from which explicit quasi-likelihood estimators of both drift coefficients a and b are derived. The second approach, applicable to a much broader class of Gaussian driving processes, constructs a strongly consistent estimator of the ratio a/b that requires only mild Hölder regularity and sublinear growth of the noise. The theoretical results are supported by a simulation study examining the effects of observation horizon and discretization step on finite-sample accuracy.","feed_headline":"Closed-form drift estimators for singular SDEs with memory","feed_subtitle":"Fractional transformation yields explicit quasi-likelihood estimates of both drift parameters; a simple ratio consistently estimates a/b","key_machinery":"The key machinery in Section 3 is the Riemann-Liouville fractional derivative of order H - 1/2, applied to the observed process X and its reciprocal 1/X, which transforms the fractional-noise-driven equation into a semimartingale driven by standard Brownian motion. This transformation (equations (8)-(9)) produces observable processes Y and Z that satisfy a linear drift equation (12), enabling the use of Heyde's martingale quasi-likelihood method to obtain explicit estimators (29). The key machinery in Section 4 is the ratio estimator ˆθ_T = (∫_0^T X_s ds) / (∫_0^T $X_s^{{-1}}$ ds), whose consistency is proved via comparison with an auxiliary Bessel-type equation (b=0) and a quadratic inequality argument that exploits the mean-reverting structure.","core_discovery":"For the stochastic differential equation dX_t = (a/X_t - b X_t) dt + dG_t with Gaussian noise G, the paper shows that when G is a Riemann-Liouville fractional Brownian motion with Hurst parameter H > 1/2, an observable fractional transformation (the Riemann-Liouville fractional derivative of order H - 1/2 applied to X and 1/X) produces processes Y and Z such that the transformed equation becomes a Brownian semimartingale with drift linear in a and b. This allows the construction of explicit closed-form quasi-likelihood estimators for a and b that coincide with maximum likelihood estimators when the latter exist. For a general class of Gaussian driving processes satisfying only local Hölder continuity of order > 1/2 and sublinear growth (G_t/t → 0 a.s.), the ratio estimator ∫_0^T X_s ds / ∫_0^T $X_s^{{-1}}$ ds is strongly consistent for θ = a/b, which has a natural interpretation as the equilibrium level of the associated squared CIR-type process.","pith_inferences":["The ratio estimator's simplicity and broad applicability suggest it could be used as a diagnostic or starting value for more complex estimation procedures, even when the full quasi-likelihood method is not applicable.","The decomposition of bias into horizon and discretization components (equation 34) provides a template for understanding finite-sample behavior of estimators in fractional SDEs that could be extended to other models.","The fractional transformation technique may be generalizable to other singular SDEs with Riemann-Liouville fractional Brownian noise, potentially yielding explicit estimators for other parameter combinations.","The strong consistency of the ratio estimator under only Hölder continuity and sublinear growth suggests that similar ratio-based estimators might work for other nonlinear SDEs where the drift has a balancing structure."],"forward_implications":["Practitioners can estimate the drift parameters of fractional CIR-type models without numerical optimization, using the closed-form formulas (29) for Riemann-Liouville fractional Brownian noise.","The ratio estimator ˆθ_T provides a simple, consistent estimator of the equilibrium level a/b that works for many Gaussian noises beyond fractional Brownian motion, including tempered and multifractional variants.","The simulation study provides practical guidance on how discretization step and observation horizon affect estimator accuracy, showing that bias has two components: one from finite horizon and one proportional to step size.","The quasi-likelihood approach avoids the need for Girsanov transformations or absolute continuity conditions, making it applicable even when standard maximum likelihood is unavailable.","The results open the door to statistical inference for a class of singular SDEs that appear in finance (interest rate models) and other fields where memory effects are present."],"supporting_citations":[{"why":"Supplies the martingale quasi-likelihood approach that is the foundation for the estimation method in Section 3.","marker":"[13]"},{"why":"Provides the general quasi-likelihood framework used to construct the quasi-score and estimators.","marker":"[14]"},{"why":"Provides existence, uniqueness, positivity, and Hölder continuity results for the solution of the SDE, which are used throughout.","marker":"[10]"},{"why":"Establishes existence and uniqueness for the fractional CIR-type process that underlies equation (1).","marker":"[27]"},{"why":"Provides the theory of Riemann-Liouville fractional integrals and derivatives used in the transformation and in proving nondegeneracy.","marker":"[31]"},{"why":"Establishes existence and uniqueness results for singular equations driven by fractional Brownian motion.","marker":"[17]"}],"fun_headline_variants":[],"cache_read_input_tokens":0,"weakest_assumption_plain":"The quasi-likelihood estimation in Section 3 assumes that the Hurst parameter H and the volatility σ are known, and that the driving noise is a Riemann-Liouville fractional Brownian motion with H > 1/2.","fun_headline_variants_meta":{"raw":{"variants":[]},"model":"deepseek/deepseek-v4-flash","effort":"low","cost_usd":8.8e-05,"raw_usage":{"cost":8.7906e-05,"is_byok":false,"cost_details":{"upstream_inference_cost":8.7906e-05,"upstream_inference_prompt_cost":8.7542e-05,"upstream_inference_completions_cost":3.64e-07},"total_tokens":964,"prompt_tokens":962,"completion_tokens":2,"prompt_tokens_details":{"audio_tokens":0,"video_tokens":0,"cached_tokens":0,"cache_write_tokens":0},"completion_tokens_details":{"audio_tokens":0,"image_tokens":0,"reasoning_tokens":0}},"tokens_in":962,"tokens_out":2,"cost_source":"openrouter_usage","duration_ms":3449,"image_pages":[],"temperature":1.0,"billed_attempts":1,"reasoning_tokens":0,"transport_provider":"openrouter","cache_read_input_tokens":0,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-11-13T12:44:46.929596+00:00","model_set":{"reader":"deepseek/deepseek-v4-flash"},"pipeline_version":"pith-pipeline@v1.4.1-daily-deepseek-or-medium","input_kind":null,"input_chars":null,"input_total_chars":null,"passes_missing":null,"published_standing":null,"published_confidence":null,"objection_audit":null,"run_record":null,"basis":{"kind":"not_recorded","source":null,"chars_given":null,"chars_total":null},"receipt":null,"falsifier":"Simulate data from the SDE (1) with a known Gaussian driving process that does NOT satisfy the Hölder condition λ > 1/2 (e.g., standard Brownian motion with H=1/2) and check whether the ratio estimator ˆθ_T still converges to a/b; if it does, the Hölder condition is not necessary, and if it does not, the condition is essential.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the martingale quasi-likelihood approach that is the foundation for the estimation method in Section 3."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the general quasi-likelihood framework used to construct the quasi-score and estimators."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides existence, uniqueness, positivity, and Hölder continuity results for the solution of the SDE, which are used throughout."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Establishes existence and uniqueness for the fractional CIR-type process that underlies equation (1)."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the theory of Riemann-Liouville fractional integrals and derivatives used in the transformation and in proving nondegeneracy."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Establishes existence and uniqueness results for singular equations driven by fractional Brownian motion."}],"lane_accuracy":{"status":"not yet measured","lane":{"model":"deepseek/deepseek-v4-flash","input_kind":null},"url":"https://pith.science/accuracy"},"review_version":1}