Paper: arXiv 2412.16067

Authors: Svetlana Boyarchenko, Sergei Levendorskiǐ

Abstract

We use modifications of the Adams method and very fast and accurate sinh-acceleration method of the Fourier inversion (iFT) (S.Boyarchenko and Levendorskiĭ, IJTAF 2019, v.22) to evaluate prices of vanilla options; for options of moderate and long maturities and strikes not very far from the spot, thousands of prices can be calculated in several msec. with relative errors of the order of 0.5% and smaller running Matlab on a Mac with moderate characteristics. We demonstrate that for the calibrated set of parameters in Euch and Rosenbaum, Math. Finance 2019, v. 29, the correct implied volatility surface is significantly flatter and fits the data very poorly, hence, the calibration results in op.cit. is an example of the {"\em ghost calibration"} (M.Boyarchenko and Levendorkiĭ, Quantitative Finance 2015, v. 15): the errors of the model and numerical method almost cancel one another. We explain how calibration errors of this sort are generated by each of popular versions of numerical realizations of iFT (Carr-Madan, Lipton-Lewis and COS methods) with prefixed parameters of a numerical method, resulting in spurious volatility smiles and skews. We suggest a general {"\em Conformal Bootstrap principle"} which allows one to avoid ghost calibration errors. We outline schemes of application of Conformal Bootstrap principle and the method of the paper to the design of accurate and fast calibration procedures.

Complexity vs Empirical Score

  • Math Complexity: 8.5/10
  • Empirical Rigor: 7.0/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper employs advanced mathematical tools like fractional calculus, Fourier analysis, and conformal methods, indicating high complexity. It also presents extensive numerical comparisons, error analysis, and claims of accurate backtesting-ready pricing, demonstrating strong empirical rigor.

Research Flowchart

  flowchart TD
  Start["Research Goal: Correct implied volatility shapes & reliable pricing in rough Heston model"] --> Method["Methodology<br>Adams method + Fast iFT with Sinh-acceleration"]
  
  Method -->|"Input"| Data["Data<br>Calibrated parameters from Euch & Rosenbaum 2019"]
  
  Data --> Process["Computational Process<br>Calculate thousands of vanilla option prices"]
  
  Process -->|"High Speed"| Result1["Outcome: Prices in ms with <0.5% relative error"]
  Process -->|"Comparison"| Result2["Outcome: Correct IV surface is flatter than model output"]
  
  Result1 --> Conclusion["Key Findings<br>1. Ghost Calibration: Errors cancel spuriously<br>2. Conformal Bootstrap principle to avoid errors"]
  Result2 --> Conclusion

Evidence Card

What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card

Proposes

The paper proposes a Conformal Bootstrap principle and improved fractional Adams and sinh‑acceleration methods for fast, accurate pricing and calibration of the rough Heston model.

Source sentence
{“}Conformal Bootstrap principle{”} which allows one to avoid ghost calibration errors.

Assumes 5

  • The riskless rate r is constant and non‑negative. market
    Source sentence
    Let the riskless rate r ≥0 be constant,
  • The characteristic function Φ(ξ,T) admits analytic continuation to a strip around the real axis. model
    Source sentence
    Assume that there exist µ−(T) < −1 < 0 < µ+(T) s.t. for β ∈(−µ+(T), −µ−(T)), the exponential moments EQ[eβXT ] are finite. Equivalently, Φ(ξ, T) admits analytic continuation to a strip S(µ−(T),µ+(T))
  • Φ(ξ) can be evaluated with (almost) machine precision. model
    Source sentence
    Assume that a union U of a strip and cone of analyticity of the characteristic function Φ(ξ) is known, and Φ(ξ) can be calculated with an (almost) machine precision.
  • The integration contours Lj are not close and diverge at infinity. other
    Source sentence
    Construct at least two admissible conformal deformations of the line of integration L0 such that the contours Lj = χj(L0), j = 1 . . . , n, are not close and diverge at infinity
Show 1 more
  • The option data (K,T) plane can be partitioned into sub‑regions and numerical scheme parameters can be tuned per sub‑region. data
    Source sentence
    An accurate and not unnecessary time consuming pricing is possible only if the (K, T)-plane is divided into several subregions as well, and the parameters of the numerical scheme are chosen for each subregion separately.

Data

  • Calibrated parameter set from Euch and Rosenbaum (2019) options unknown
    Source sentence
    We demonstrate that for the calibrated set of parameters in Euch and Rosenbaum, Math. Finance 2019, v. 29, the correct implied volatility surface is significantly flatter and fits the data very poorly
  • Benchmark prices computed on finer and longer grids synthetic unknown
    Source sentence
    In the tables, we show the errors and relative errors w.r.t. benchmark prices calculated using much finer and longer grids

Costs

Not applicable The paper reports no trading or execution result.

Finds 5

  • The proposed method evaluates thousands of vanilla option prices in a few milliseconds with relative errors around 0.5% or smaller. pricing-calibration
    Source sentence
    We use modifications of the Adams method and very fast and accurate sinh-acceleration method of the Fourier inversion (iFT)
  • The sinh‑acceleration combined with the fractional Adams method achieves error tolerances below 1e‑6 and CPU costs thousands of times smaller than COS. performance
    Source sentence
    We also show that the relative errors of prices calculated using COS method are significantly larger than the errors of prices produces by the method of the paper at the CPU cost thousands of times smaller.
  • COS‑method prices exhibit relative errors between 5% and 22%, leading to spurious volatility smiles. pricing-calibration
    Source sentence
    We produced correct volatility curves for an example in [21] where COS method is used; the relative errors of COS-prices are in the range 5%-22%.
  • Ghost calibration arises when numerical errors cancel model errors, producing misleadingly good fits. theoretical
    Source sentence
    ... the errors of the model and numerical method almost cancel one another.
  • For short‑maturity options, grids must be significantly larger than for moderate maturities; using a single parameter set for all (K,T) leads to incorrect calibration results. negative
    Source sentence
    The common belief that the same set of parameters of the numerical scheme can be used for all pairs (K, T) of interest (as in CM and COS methods) leads to incorrect results, in calibration procedures especially.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 38 related papers). Coverage is partial, so this is neither replication nor refutation.

Cited by 1 paper in the archive (1 by independent authors): The Volterra Stein-Stein model with stochastic interest … (uses its method or data)

Code

No repository by the authors found in the paper.

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.