Lévy processes as weak limits of rough Heston models

Alessandro Bondi, Martin Forde · 2025

Read Lévy processes as weak limits of rough Heston models by Alessandro Bondi, Martin Forde (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2508-14835.txt.

Introduction

Alessandro Bondi and Martin Forde published a mathematics paper in 2025 on rough volatility models collapsing into Levy processes. The rough Heston model drives variance with a jagged memory kernel. Rescaling it, the authors prove the integrated variance at a fixed time converges weakly to an inverse Gaussian Levy process. Earlier routes to this limit forced special values of the roughness index.

Their proof needs no such forcing, which may matter more for real markets. With jumps added to the variance, the limit becomes a clock-changed barrier hitting-time process for a broad family of one-sided Levy processes. The convergence is then lifted to whole paths in a suitable topology.

Catalog record

CREATORAlessandro Bondi, Martin Forde
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE64 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2508-14835

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Lévy processes as weak limits of rough Heston models Alessandro Bondi, Martin Forde Abstract We show weak convergence of the marginals for a re-scaled rough Heston model to a Normal Inverse Gaussian (NIG) Lévy process. This shows we can obtain such a limit without having to impose that the true Hurst exponent H H for the model is 1 2 \frac{1}{2} as in [ AC24 ] , 1 1 1 The H H parameter in [ AC24 ] is not fixed to be 1 2 \frac{1}{2} , but the Hurst exponent for their family of V ε V^{\varepsilon} processes is of course 1 2 \frac{1}{2} since the models are all standard Heston for ε > 0 \varepsilon>0 . or that H ↘ − 1 2 H\searrow-\frac{1}{2} as in [ AAR25 ] , so the result potentially has increased financial relevance. We later extend to the case when V V has jumps, where we show weak convergence of the finite-dimensional distributions of the integrated variance to a deterministic time-change of the first passage time process to lower barriers for a more general class of spectrally positive Lévy processes. 2 2 2 Note that for citations to results in [ ALP19 ] , [ BLP24 ] , [ BPS24 ] , [ FGS21 ] and [ GK19 ] we are referring to Lemma or Theorem numbers in the published, not the arXiv versions. MSC2020: 60H20; 45D05; 60F05; 60G22 Keywords: Affine Volterra processes; Rough Heston model; Fast Mean reversion; Lévy processes hitting times; Volterra Integral equations 1 Introduction The Rough Heston stochastic volatility model was introduced in [ JR16 ] , and (using C- tightness arguments) they show that the model arises naturally as a weak large-time limit of a high-frequency market microstructure model driven by two nearly unstable Hawkes processes. [ ALP19 ] , [ ER19 ] and [ GK19 ] show that the characteristic function of the log stock price for Rough Heston-type models can be expressed in terms of the solution to a non-linear Volterra integral equation (VIE) (see also [ EFR18 ] , [ ER18 ] and [ BLP24 ] , [ BPS24 ] , [ CT20 ] for extensions to jumps in V ), which allows for accurate option pricing for H\ll 1 (and even H=0 ) using an Adams scheme to solve the VIE numerically. This avoids Monte Carlo (MC) techniques for which traditional MC schemes are notoriously inaccurate for H\ll 1 (both in terms of bias and sample variance). From the stochastic Fubini theorem, it is well known that A_{t}=\int_{0}^{t}V_{s}ds satisfies an equation of the form: A_{t}=G_{0}(t)\,+\,\int_{0}^{t}\kappa(t-s)W_{A_{s}}ds\, for some Brownian motion W (see e.g. Theorem 3.1 in [ JR20 ] ), which is a non-linear VIE for A in terms of the a.s. (\frac{1}{2}-\varepsilon)- Hölder continuous function W_{(\cdot)} , and is still well defined even if \kappa is only in L^{1}(0,T) ; this also allows us to consider the hyper-rough regime H\in(-\frac{1}{2},0] . If we discretize this SVE and re-write in terms of the final (discrete-time) increments of A and W , then by Eq. ( E-3 ) in Appendix E , we can use an independent sequence of Inverse Gaussian random variables to perform an approximate Monte Carlo simulation of A (see Algorithm 1 in [ AA25 ] for details) which is naturally suited to the regimes H\ll 1 and H\in(-\frac{1}{2},0] . The variance process V is (H-\varepsilon)- Hölder continuous like fBM (see e.g. Theorem 3.2 in [ JR16 ] ) and the spot and VIX smiles exhibit power-law skew for implied volatility in the small-time limit (see [ FGS21 ] , [ FSV21 ] , [ FGS22 ] for details). See also [ ALP19 ] , [ Cuch22 ] , [ FG24 ] , [ FGS21 ] , [ FS21 ] for further related results on rough Heston-type models. In related work, [ Jai15 ] considers two identical i.i.d. Hawkes processes N^{\pm} with kernel \phi\in L^{1}(0,\infty) and, assuming that the asset price P_{t}=\text{const}\times\lim_{u\to\infty}\mathbb{E}[N^{+}_{u}-N^{-}_{u}|\mathcal{F}_{t}] , i.e. a constant times the conditional expected value of all future order flow, then P is clearly a martingale, but can also be written in the propagator form P_{t}=\int_{0}^{t}\zeta(t-u)(dN^{+}_{u}-dN^{-}_{u}). If \zeta^{\prime}(t)=-\zeta(0)\phi(t)\, and if \zeta(\infty)>0 , \zeta(\infty) is the (non-transient) permanent price impact component of \zeta , and we can compute the expected market impact of an exogenous metaorder executed at constant rate v over duration \tau as MI(t)=v\int_{0}^{t\wedge\tau}\zeta(t-s)ds (see also [ JR20 ] for more on this). [ BL24 ] improve accuracy in estimating the tail part of the integral for Fourier inversion (for e.g. call options under the rough Heston model) using a sinh contour which goes outside the usual strip of analyticity for the mgf (but is admissible as long as it avoid poles of the characteristic function), and also discuss refinements to the usual Adams schemes for solving the rough Heston VIE. For articles on statistical estimation of H in more general setups, we refer the reader to [ CD24 ] , [ HS25 ] , [ BF18 ] (and follow-up papers on the LAN property and minimax theorems), and the asymptotic normality result for \sqrt{n}(\hat{H}_{n}-H),\frac{\sqrt{n}}{\log n}(\hat{\sigma}_{n}-\sigma) in Theorem 1 in [ Syz23 ] for \sigma B^{H} (where B^{H} is fBM with \sigma unknown) for the Whittle approximation (\hat{H}_{n},\hat{\sigma}_{n}) for the MLEs in terms of the Fisher information matrix in the high-frequency regime. [ AC24 ] show that a re-scaled standard Markov Heston model with fast mean-reversion and large vol-of-vol (via an H parameter which is not the Hurst exponent) tends weakly on path space to one of three different models (either Black-Scholes, a Normal Inverse Gaussian or a Normal Lévy model), depending on whether their H parameter is > , = , or < -\frac{1}{2} . [ AAR25 ] obtain a similar result without any \varepsilon parameter but instead letting H\searrow-\frac{1}{2} for the so-called hyper-rough Heston model (see also Section 5 in [ FGS21 ] and Section 7 in [ A21 ] for more on this model), and exploiting Dirac-type behaviour in their Lemma 2.4 (see Appendix E here for a short summary/formal derivation of their result). In this note, we fill in the gap between [ AC24 ] and [ AAR25 ] , by showing that a conceptually similar result is obtained for any H\in(0,\frac{1}{2}] (in particular our regime for H=\frac{1}{2} corresponds to the regime in Eq. (0.3) in [ AC24 ] with their H=-\frac{1}{2} ). In Section 2 , we extend the model to allow positive jumps in V^{\varepsilon} , and in this case (using the Laplace transform of the hitting time to a lower barrier for a spectrally positive Lévy process), we find that the limiting process for the integrated variance is a deterministic time-change of the first passage time process to lower barriers for a more general class of Lévy processes (the convergence is proved using a compactness argument with the Kolmogorov-Riesz-Fréchet theorem). 1.1 Asymptotics for the terminal log stock price for a re-scaled rough Heston model We work on a probability space (\Omega,\mathcal{F},\mathbb{Q}) throughout with filtration (\mathcal{F}_{t})_{t\geq 0} satisfying the usual conditions. Proposition 1.1 Consider a re-scaled rough Heston model for a log stock price process X^{\varepsilon}_{t} : \displaystyle dX^{\varepsilon}_{t} \displaystyle= \displaystyle-\frac{1}{2}V^{\varepsilon}_{t}dt\,+\,\sqrt{V^{\varepsilon}_{t}}dB_{t}, \displaystyle V^{\varepsilon}_{t} \displaystyle= \displaystyle V_{0}\,+\,\frac{1}{\Gamma(\alpha)}\int_{0}^{t}(t-s)^{\alpha-1}\Big(\frac{1}{\varepsilon}\lambda(\theta-V^{\varepsilon}_{s})ds+\frac{1}{\varepsilon}\nu\sqrt{V^{\varepsilon}_{s}}dW_{s}\Big), (1) where B and

Read the complete text