{"id":"fbeced04-f82b-4f76-bab6-ee0768cb076f","arxiv_id":"2412.16067","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"The paper shows that the rough Heston calibration in El Euch and Rosenbaum (2019) is likely a numerical artifact, and provides faster, more accurate pricing methods.","lead":"This paper introduces modified fractional Adams methods and sinh-accelerated Fourier inversion to price options in the rough Heston model, and claims that a prominent calibration of that model is an artifact of numerical errors. A generalist might care because it suggests that some published evidence for rough volatility could be spurious, and it proposes a principle to avoid such 'ghost calibration'.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Ghost-calibration claim depends on short-maturity ATM skews computed with a self-referential error control; the shared asymptotic ansatz in the modified Adams method is not independently benchmarked.","rationale":"Read in good faith, the paper's goal is to correct the numerical pricing used in rough Heston calibration and to show that the Euch-Rosenbaum parameters produce a flatter, poorly fitting IV surface. That claim would be true only if the paper's own prices are accurate to well below the gap it reports. The paper is unusually transparent about the conditional nature of its numerics, and the internal consistency checks (agreement among modifications, fine-grid benchmarks, comparisons with the hybrid method) are substantial. The weakest point is not the lack of a rigorous analyticity proof per se; it is that the regime decisive for the headline claim — short-dated ATM skew — is precisely where the method's error control is heuristic and self-referential. All modifications share the same asymptotic ansatz (2.13), so the bootstrap agreement cannot exclude a common large-|ξ| bias. The ad hoc bound (3.20) is acknowledged to be imperfect, and no independent benchmark exists for T = 1/52 and T = 1/252. Because of this, the quantitative claim of 'several times lower' skew is not yet established; it is a plausible but unverified numerical finding. This does not warrant rejection — the paper's constructive methods and its negative results about CM, COS, and Lewis methods are valuable — but it does warrant independent verification before the ghost-calibration assertion is relied upon. Hence the reader's CONDITIONAL verdict remains appropriate, with the condition being an independent short-maturity benchmark. The reader's weakest assumption about the unproven analyticity and the modified Adams accuracy for large |ξ| is the same underlying concern, and our test targets it directly.","tokens_in":41593,"tokens_out":5350,"duration_ms":49797,"concrete_test":"Implement an independent high-precision solver for the fractional Riccati equation (2.7) at parameters (1.1) that does not use the ansatz (2.13), for example a Chebyshev spectral collocation in t with step refinement per |ξ|, or a high-order exponential-integrator scheme, and verify convergence by requiring two successive refinements to agree to 1e-10. Use this solver to evaluate the characteristic function on the two sinh-acceleration grids used for T = 1/252 and T = 1/52, then compute ATM option prices and invert implied volatility. If the resulting ATM skew differs from the paper's SINH curves by more than 20% of the reported gap between those curves and the Euch-Rosenbaum curves, the ghost-calibration conclusion is not supported. A Monte Carlo simulation of the rough Heston process with a fine hybrid scheme and Richardson extrapolation would be a valuable second independent check.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central empirical claim is that the Euch-Rosenbaum parameters yield a much flatter and poorly fitting IV surface, with the ATM skew 'several times lower' than in [23]. This claim is decided by short-maturity skews, especially T = 1/52 and T = 1/252. In that regime, the paper's three modifications of the fractional Adams method all build on the same asymptotic ansatz h_as(ξ,t) = -0.5(ξ^2 + iξ) t^α / Γ(α+1), Eq. (2.13), and compute corrections with the same product-integration coefficients. The Conformal Bootstrap principle (Sec. 3.9) compares prices from different contours and from different modifications, but if the h_as ansatz is biased for large |ξ| and small t, all modifications inherit that bias and agreement between them does not detect it. The paper itself concedes that rigorous error bounds are lacking: Sec. 6 states 'any numerical procedure is a conditional one', and Sec. 3.8 calls the truncation bound (3.20) ad hoc and 'far from perfect'. The only external comparisons with the hybrid method of [16] are for moderate maturities; at T = 1/52 and T = 1/252, the hybrid prices in Tables 16-17 are far from the paper's benchmark and cannot validate the short-maturity ATM skew. Thus the headline 'ghost calibration' conclusion rests on a self-consistent numerical construction whose shared asymptotic approximation could, in principle, produce the same flat skew even if the true model skew is closer to [23].","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops fast and accurate pricing procedures for vanilla options in the rough Heston model, combining modified fractional Adams methods for the fractional Riccati equation with sinh-accelerated Fourier inversion. It presents numerical evidence that several popular Fourier inversion schemes (Carr-Madan, Lewis, COS, flat iFT) produce sizable pricing errors, and it uses the new method to compute implied volatility surfaces for the Euch-Rosenbaum calibrated parameters. The central empirical claim is that the correct implied volatility surface for those parameters is much flatter than reported in the original calibration, with an ATM skew several times lower, making the original calibration an example of 'ghost calibration'. The paper also proposes a Conformal Bootstrap principle for error assessment and outlines calibration procedures based on it.","tokens_in":41988,"tokens_out":3677,"duration_ms":34800,"significance":"If correct, the paper's central claim is significant: it challenges a widely cited rough Heston calibration and attributes the apparent fit to numerical error cancellation rather than model performance. The proposed pricing methodology is also practically valuable: it reports millisecond-scale pricing for moderate maturities, uses a mathematically motivated improvement of the standard Adams method, and includes detailed comparisons with the hybrid method of [16]. The paper is honest about the conditional nature of its error control, but this honesty cuts both ways: the headline claim rests on numerical error assessment that is explicitly heuristic. The absence of machine-checked proofs or independently certified short-maturity benchmarks means the current evidence is suggestive rather than conclusive.","major_comments":[{"comment":"The truncation parameter N in the pricing algorithm is selected from the ad-hoc bound (3.20), and the paper itself states that this bound is 'far from perfect' and that 'any numerical procedure is a conditional one' (Section 6). Because no rigorous proof of (3.18)-(3.20) is provided, the reported error levels, especially at T=1/52 and T=1/252 where decay is slowest, are not certified. This is load-bearing: the headline claim that the true ATM skew is several times lower than in [23] depends on accurate short-maturity prices produced with this unproven truncation control.","section":"Section 3.8, Eq. (3.20)"},{"comment":"Modifications I-III of the fractional Adams method all build on the same asymptotic ansatz (2.13), and the product-integration coefficients used in the corrections are shared. The Conformal Bootstrap principle compares prices evaluated on different contours and with different modifications, but if the ansatz (2.13) is biased for large |ξ| and small t, each modification inherits that bias and agreement between them does not detect it. The paper provides no independent test of (2.13) in exactly the regime where the ghost-calibration conclusion is decided.","section":"Section 2.2 and Section 3.9"},{"comment":"The external comparisons with the hybrid method of [16] agree well at moderate maturities (Tables 12-15), but at T=1/52 and T=1/252 the hybrid prices diverge strongly from the paper's benchmarks. For example, in Table 16 at K=1.05 the hybrid price is 4.113E-04 versus the benchmark 3.752E-04, and in Table 17 at K=1.05 the hybrid price is 6.39E-07 versus the benchmark 3.31E-08. These comparisons therefore do not validate the paper's benchmarks in the maturity range where the 'ghost calibration' claim is decided; an independent short-maturity benchmark is missing.","section":"Tables 16-17"},{"comment":"The benchmark prices are described as calculated using 'much finer and longer grids', but no error certificate for these benchmarks is provided beyond agreement between two sinh-deformation parameter sets. Because both parameter sets are generated by the same family of Adams modifications and the same unproved decay bound, the agreement is an internal consistency check rather than an accuracy certificate. Please provide reproducible code, exact or high-precision independent benchmarks, or a rigorous error bound for at least the short-maturity ATM skew that is central to the paper's main conclusion.","section":"Section 4.3 and Tables 12-17"}],"minor_comments":[{"comment":"The section title contains a typo: 'inifinite trapezoid rule' should read 'infinite trapezoid rule'.","section":"Section 3.6"},{"comment":"The sentence 'CM method caused problems in the financial industry, and has net been used by practitioners ever since' should read 'has not been used'.","section":"Section 3.3"},{"comment":"The header of Table 17 appears to contain a duplicated 'VH Err VH' column and lacks the 'Vfast' column that is described in the table notes; please correct the table formatting.","section":"Appendix B, Table 17"},{"comment":"The paper states that the ATM skew is 'several times lower' than in [23], but Figure 3(a) does not provide the numerical values of the skew at T=1/52 and T=1/252. Reporting these values explicitly would make the central comparison quantitative and easier to verify.","section":"Figure 3"},{"comment":"The caption of Figure 5 is grammatically garbled in the sentence beginning 'the true difference between the empirical implied volatilities...' and should be rewritten for clarity.","section":"Figure 5 caption"},{"comment":"There is a typographical error in the spelling of Levendorskiĭ in the abstract; the same issue appears in several other places in the text.","section":"Abstract"}],"recommendation":"major_revision","confidential_remarks":"The paper has a strong and provocative central claim, and the numerical methodology is clever and potentially useful. However, the short-maturity benchmarks that decide the ghost-calibration claim are not independently certified, and the Conformal Bootstrap principle cannot detect a shared bias in the asymptotic ansatz. I would advise the editor that acceptance should require either a rigorous error bound for the relevant short-maturity regime, or an independent verification (e.g., high-precision code, a different ODE solver, or a trusted external method) of the ATM skew at T=1/52 and T=1/252. Without such evidence, the headline claim remains an interesting but unproven numerical assertion."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know this paper is worth reading even if you do not buy its strongest claim. It does three genuinely useful things: it proposes three modifications of the fractional Adams method for the rough Heston Riccati equation, it applies sinh-acceleration to get very fast and accurate pricing at moderate maturities, and it documents large errors in the standard CM, COS, and Lewis implementations used in calibration studies. The numerical comparisons against the hybrid method of Callegaro–Grasselli–Pagès are fair and detailed, and the paper is unusually candid about the limits of its own error control. The internal consistency checks across different contour deformations and different Adams modifications are extensive and, for moderate maturities, convincing.\n\nThe soft spot is exactly where the paper makes its headline claim. The ghost-calibration conclusion—that the El Euch–Rosenbaum parameters give an ATM skew several times lower than reported—hinges on prices at T = 1/52 and T = 1/252. In that regime, the modified Adams methods all share the same asymptotic ansatz h_as(ξ,t) = -0.5(ξ^2+iξ)t^α/Γ(α+1), and the error control relies on the ad-hoc truncation bound (3.20) that is not proved and is admitted to be 'far from perfect.' The conformal bootstrap principle is a self-consistency check, not a rigorous error bound: agreement between contours and between modifications does not detect a common bias from that shared asymptotic approximation. The only independent comparison, the hybrid method, is far off at exactly these short maturities, so it cannot validate the skew. The paper itself concedes in Section 6 that 'any numerical procedure is a conditional one.' That is honest, but it means the paper's central claim is conditional on a heuristic the authors have not fully justified.\n\nWhat is solid: the critique of fixed-parameter Fourier inversion methods is well supported by examples, the new Adams modifications are genuinely useful, and the computational speed is impressive. The citation pattern is fine; the self-citations are to prior methods that are directly relevant.\n\nWho is this for? Anyone working on rough volatility calibration, Fourier pricing methods, or numerical fractional ODEs. It deserves a serious referee. I would accept it for review, but I would send it back demanding the code or reproducible scripts, plus an independent high-precision solution of the fractional Riccati equation (for example, a spectral or exponentially convergent method) for the short-maturity skew. Without that, the ghost-calibration claim remains credible but unproven.","headline":"A serious numerical-methods paper whose 'ghost calibration' claim about El Euch–Rosenbaum is plausible but not yet fully verified; it deserves peer review, with code and an independent short-maturity check required.","tokens_in":42448,"tokens_out":1744,"would_cite":true,"duration_ms":18441,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60-08","60E10","60G10","60G22","65C20","65D30","65G20","91G20"],"pacs":[],"model":"deepseek-v4-flash","headline":"Popular Fourier pricing can manufacture spurious smiles in rough Heston; a benchmark calibration is a ghost fit.","keywords":["rough Heston model","fractional Adams method","sinh-acceleration","ghost calibration","implied volatility surface","Fourier inversion","Conformal Bootstrap principle","fractional Riccati equation"],"falsifier":"Take the paper's calibrated parameter set (1.1), compute option prices with an independent high-accuracy method such as a very fine Monte Carlo simulation of the rough Heston process or a completely different quadrature with rigorous error bounds, for maturities from one week to one month and strikes within 20% of spot, and compare the implied volatilities to both the published surface from [23] and the paper's SINH surface. If the independent surface matches the published one rather than the flatter SINH one, the ghost-calibration claim is refuted; a cheaper check is to increase the Adams grid size $M$ and the Fourier truncation $N$ simultaneously and see whether the paper's prices move by more than the claimed $10^{-8}$.","tokens_in":2002,"feed_emoji":"📉","tokens_out":4981,"duration_ms":94463,"temperature":0.7,"pith_summary":"The paper argues that a widely cited calibration of the rough Heston model, the one in reference [23], is a numerical artifact rather than a genuine fit. Using its own corrected pricing machinery, a modified fractional Adams method with sinh-accelerated Fourier inversion, it constructs the implied volatility surface for those parameters and finds it markedly flatter and a poor fit to the original data. The mechanism is ghost calibration: model error and numerical error almost exactly cancel. The paper also proposes the Conformal Bootstrap principle, a practical agreement-based error check for models whose characteristic function lacks proven analyticity or decay. If correct, published empirical claims about the rough Heston smile and any calibrations built on standard Fourier methods need to be re-examined.","feed_headline":"Rough-Heston calibration result was a numerical ghost","feed_subtitle":"With accurate pricing the benchmark surface is much flatter and fits poorly; model and method errors had cancelled.","key_machinery":"Three components carry the argument. First, modifications of the fractional Adams method for the fractional Riccati equation replace the naive first predictor with the small-time asymptotics $-0.5(\\xi^2+i\\xi)t^\\alpha/\\Gamma(\\alpha+1)$, so that large spectral parameter $|\\xi|$ does not corrupt short-maturity prices. Second, $\\sinh$-acceleration for Fourier inversion deforms the integration contour into a hyperbolic curve $i\\omega_1+b\\sinh(i\\omega+y)$, making the oscillatory integrand decay exponentially and allowing a short trapezoid sum to achieve high accuracy; an ad-hoc bound, equation (3.20), is used to choose the truncation $N$. Third, the Conformal Bootstrap principle accepts a computed price only when two well-separated contour deformations agree to about $10^{-m}$, because agreement on distant grids makes a common large error essentially impossible. The paper also uses the identity $\\Phi=\\exp[\\int_0^\\tau(\\gamma\\theta h+vF)\\,ds]$, which avoids a separate fractional integration when assembling the log-characteristic function.","core_discovery":"The central claim is that the benchmark rough Heston calibration in reference [23] is an example of ghost calibration. For the same parameter set, the paper's numerical procedures produce an ATM skew several times lower than the published one, decaying quickly with maturity, and an implied volatility surface that is flatter and fits the market data poorly. It diagnoses the sources of the discrepancy: the standard fractional Adams method mishandles large Fourier frequencies in the predictor step for short maturities, and fixed-parameter Carr-Madan, COS, and Lewis inversions introduce systematic errors that can reshape a straight volatility slope into a smile. The paper asserts that its method evaluates thousands of vanilla prices in milliseconds with relative errors around $10^{-3}$ or better, and that the disagreement is large enough to invalidate calibration conclusions drawn from the standard methods.","pith_inferences":["The ghost-calibration mechanism likely extends beyond the rough Heston model to any pipeline coupling a numerically solved Riccati equation with fixed-parameter Fourier inversion, including affine models and deep-learning pricing surrogates trained on such prices.","A testable consequence of the paper's claim is that recalibrating the same dataset with a certified-accurate pricer should move the fitted parameters to a different region of parameter space and worsen the apparent fit, directly confirming error cancellation.","The Conformal Bootstrap principle could be automated as an online error certificate in production systems: reject any batch of prices unless two distinct contour deformations agree, then adaptively expand grids until they do.","A natural extension is to test the paper's short-maturity accuracy claims in higher-precision or interval arithmetic, especially where the explicitly ad-hoc decay bound (3.20) is the only support for the chosen truncation."],"forward_implications":["For the benchmark rough Heston parameters in [23], the published volatility smile and ATM skew are replaced by a flatter, poorly fitting surface, so any calibration conclusions drawn from that surface need to be rechecked.","Short-maturity out-of-the-money options, where rough-volatility effects are claimed to be strongest, are exactly where standard CM, COS, and Lewis schemes are least reliable, so empirical tests of rough-volatility skew behavior based on those schemes are suspect.","The modified Adams plus sinh-acceleration pipeline evaluates thousands of option prices in milliseconds with relative errors around $10^{-3}$ or better on a standard laptop, making reliable calibration computationally feasible.","Using a fixed set of Fourier-grid parameters across strikes and maturities, as in the common CM and COS practice, can manufacture smiles and skews that are not present in the true model prices; the Conformal Bootstrap principle provides a practical check against such artifacts.","Option prices outside the no-arbitrage bounds, which the paper reproduces for FFT-based and COS schemes, offer a simple diagnostic that a calibration is being driven by numerical error rather than model content."],"supporting_citations":[{"why":"Supplies the benchmark rough Heston model, the calibrated parameters, and the Lewis/Adams results that the paper re-evaluates and finds incorrect.","marker":"[23]"},{"why":"Hybrid method and published prices the paper compares against for speed and accuracy, especially for short-maturity options.","marker":"[16]"},{"why":"Introduces the sinh-acceleration Fourier inversion method that the paper adapts to the rough Heston model.","marker":"[11]"},{"why":"Defines ghost calibration, the error-cancellation phenomenon the paper uses to explain the spurious good fit.","marker":"[5]"},{"why":"The CM method whose fixed parameter choices the paper analyzes as a source of spurious smiles and no-arbitrage violations.","marker":"[17]"},{"why":"The COS method whose payoff modification and discretization errors the paper quantifies on rough Heston examples.","marker":"[24]"},{"why":"The Lewis Fourier inversion scheme used in later rough Heston studies; the paper shows its error growth and unreliability in the wings.","marker":"[41]"},{"why":"Empirical Heston calibration study showing that CM and COS produce incorrect volatility curves, which motivates the rough Heston analysis.","marker":"[19]"},{"why":"Earlier analysis of Fourier-method pitfalls in affine models that the paper extends to the numerically solved rough Heston characteristic function.","marker":"[40]"}],"fun_headline_variants":["Rough Heston benchmark was a numerical ghost","Ghost calibration exposed in rough Heston","Accurate pricing flattens rough Heston smiles","Rough Heston calibration errors cancel out","How to avoid ghost calibration in rough Heston"],"cache_read_input_tokens":44544,"weakest_assumption_plain":"The load-bearing premise is that the rough Heston characteristic function is analytic in a cone around the real axis and decays there fast enough; the paper's only support for this is an explicitly ad-hoc bound, so if that analyticity or decay fails, the quoted error floors and the ghost-calibration claim could change.","fun_headline_variants_meta":{"raw":{"variants":["Rough Heston benchmark was a numerical ghost","Ghost calibration exposed in rough Heston","Accurate pricing flattens rough Heston smiles","Rough Heston calibration errors cancel out","How to avoid ghost calibration in rough Heston"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000569,"raw_usage":{"total_tokens":2713,"prompt_tokens":987,"completion_tokens":1726,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":603,"completion_tokens_details":{"reasoning_tokens":1657}},"tokens_in":603,"tokens_out":1726,"duration_ms":10408,"temperature":1.0,"reasoning_tokens":1657,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T10:49:25.101213+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take the paper's calibrated parameter set (1.1), compute option prices with an independent high-accuracy method such as a very fine Monte Carlo simulation of the rough Heston process or a completely different quadrature with rigorous error bounds, for maturities from one week to one month and strikes within 20% of spot, and compare the implied volatilities to both the published surface from [23] and the paper's SINH surface. If the independent surface matches the published one rather than the flatter SINH one, the ghost-calibration claim is refuted; a cheaper check is to increase the Adams grid size $M$ and the Fourier truncation $N$ simultaneously and see whether the paper's prices move by more than the claimed $10^{-8}$.","supporting_citations":[{"cited_title":"Euch and M","cited_arxiv_id":null,"evidence_quote":"Supplies the benchmark rough Heston model, the calibrated parameters, and the Lewis/Adams results that the paper re-evaluates and finds incorrect."},{"cited_title":"Callegaro, M","cited_arxiv_id":null,"evidence_quote":"Hybrid method and published prices the paper compares against for speed and accuracy, especially for short-maturity options."},{"cited_title":"Carr and D.B","cited_arxiv_id":null,"evidence_quote":"The CM method whose fixed parameter choices the paper analyzes as a source of spurious smiles and no-arbitrage violations."},{"cited_title":"Fang and C.W","cited_arxiv_id":null,"evidence_quote":"The COS method whose payoff modification and discretization errors the paper quantifies on rough Heston examples."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"The Lewis Fourier inversion scheme used in later rough Heston studies; the paper shows its error growth and unreliability in the wings."},{"cited_title":"de Innocentis and S","cited_arxiv_id":null,"evidence_quote":"Empirical Heston calibration study showing that CM and COS produce incorrect volatility curves, which motivates the rough Heston analysis."},{"cited_title":"Levendorski ˘i","cited_arxiv_id":null,"evidence_quote":"Earlier analysis of Fourier-method pitfalls in affine models that the paper extends to the numerically solved rough Heston characteristic function."}],"review_version":1}