Searcharxiv⌕ Search
Home/Archive/arXiv

arXiv · 2610.28847

DSINC beyond Heston: the 3/2 and other non-affine stochastic-volatility models

Andrey Itkin
View original source ↗Read source document ↗Export metadata ↓

Abstract

This paper extends the damped-sinc (DSINC) method for American options beyond the affine Heston class to general one-factor stochastic-volatility models. We show that the collapse identity, which reduces the early-exercise integral to a single spectral integral, remains valid for these models. Damping between zero and one is admissible whenever the discounted spot is a nonnegative local martingale. An Itô transformation of the volatility factorizes the conditional characteristic function and identifies a natural coordinate for approximating the exercise boundary. The 3/2 model admits closed-form joint transforms in terms of Kummer functions, while uncorrelated SABR admits a Bessel representation. When a practical closed-form transform is unavailable, the required volatility integral is computed directly by solving one backward equation per frequency, without a grid in the log-price. Through a known change of volatility coordinate, this approach also accommodates correlated SABR. Tests on the 3/2, SABR, GARCH and lognormal-volatility models show agreement with fine finite-difference benchmarks to within one basis point of the strike. Computational cost is comparable to or lower than that of our finite-difference implementation, except for the slower Hankel implementation of uncorrelated SABR.

Explore related subjects

q-fin.CPcs.NAmath.NAq-fin.MFq-fin.PR

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Andrey Itkin. 2026-10-20. DSINC beyond Heston: the 3/2 and other non-affine stochastic-volatility models. https://arxiv.org/abs/2610.28847

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

Rethinking Synthetic Scenario Realism: Compatibility, Not Fidelity, Drives Hedging Performance

Deep hedging has emerged as a data-driven approach to optimizing hedging strategies. Since real market data are often limited, synthetic price paths play a crucial role in training these strategies. However, existing approaches primarily evaluate synthetic data generators based on statistical realism, while its relationship with hedging performance remains unclear. In this work, we introduce a decision-centric perspective on synthetic data for deep hedging through the notion of compatibility, which captures discrepancies in hedging risk between synthetic and real market environments. Theoretically, we establish an upper bound on hedging performance degradation in terms of learning error and a compatibility gap, and show that statistical realism and compatibility can structurally diverge. In controlled experiments, we demonstrate that the compatibility gap captures variations in hedging degradation across tasks even when generator realism is held fixed. Empirically, using real market data, we show that statistical realism does not consistently predict hedging performance, and that generator rankings depend on both hedging strategy classes and task structure.

q-fin.CP↗
arXiv2026

Lineage-Aware Search Control for LLM-Based Evolutionary Factor Mining

Formulaic alpha factor mining searches for symbolic expressions that transform market data into predictive signals. Recent LLM-based evolutionary factor mining methods use large language models as semantic variation operators. Under the conventional LLM-EA paradigm, stochastic variation and fitness-based selection can leave the search weakly directed and concentrated around a narrow set of structurally similar factors. We argue that these limitations arise because current methods mainly organize search at the individual and population levels, leaving lineage-level control underused. A lineage connects successive factor variants through evolutionary descent. We propose LineageEvo, which uses lineage in two coupled forms of search control: (1) lineage-conditioned LLM variation, which distills Lineage-Scoped Search Experience into operator-specific priors and fuses them with global operator priors, providing persistent, lineage-relevant guidance for LLM generation; and (2) lineage-stratified search allocation, which balances factor quality against lineage concentration by distributing the search budget across lineages during evolutionary search. Together, these mechanisms make semantic variation more direction aware and prevent branches from dominating the evolution, thereby promoting the discovery of predictive and diverse factors. We evaluate LineageEvo on CSI 500 and S&P 500 in terms of factor predictiveness, portfolio performance, search efficiency, and lineage- and expression-level diversity. LineageEvo achieves the strongest predictive performance on both markets, while LSPS produces a more balanced lineage distribution and a structurally broader factor pool. The results support lineage as a useful intermediate unit for controlling LLM-based evolutionary factor search.

q-fin.CP↗
arXiv2026

Kernel-based collocation methods for Heath-Jarrow-Morton models with Musiela parametrization

We study kernel-based collocation methods for numerical solutions of Heath-Jarrow-Morton models in Musiela parametrization. The methods are Euler-Maruyama approximations of finite-dimensional stochastic differential equations and can be used with standard Monte Carlo procedures for derivative pricing. We give a conditional mean-square error estimate under explicit stability assumptions on the differentiated cardinal basis over the full evaluation region required by the volatility functional. The estimate uses the spatial-radius exponent dictated by the squared-error calculation. Two finite-resolution examples illustrate the method for a Vasicek model and for caplet pricing.

q-fin.CP↗