arXiv Science⌕ Search

arXiv · 2611.20746

European derivatives pricing via constrained Gaussian process regressions

Abstract

This article develops a Gaussian process method for pricing European derivatives when the underlyings, and the risk factors driving them, follow a multivariate jump-diffusion. A Gaussian process prior on the pricing function is conditioned on the Feynman-Kac partial integro-differential equation at a set of collocation points. The price is returned as a closed-form and differentiable function, allowing the Greeks to be obtained without simulation or finite differences. The jump terms are handled iteratively: the price is computed as the limit of a sequence of boundary value problems. Our main theoretical contribution is a convergence result for this iteration: using Weissinger's fixed-point theorem in the supremum norm, we show that it converges from any starting point, and provide a worst-case number of iterations for a given precision. The method is illustrated with derivatives of two underlyings under the Heston model with Kou double-exponential jumps, a five-dimensional problem, and benchmarked against Monte Carlo simulation.

Explore related subjects

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Arnaud Servais, Donatien Hainaut. 2026-11-16. European derivatives pricing via constrained Gaussian process regressions. https://arxiv.science/abs/2611.20746

Citations retain the original authors and DOI and include this site’s record URL. If this discovery service contributed to your research, you can also cite arXiv Science by Indiciae: BibTeX · RIS.

KEEP EXPLORING

Related papers

Quantifying Repair Risk in Crypto Option Surfaces

Option desks adjust crypto option quotes into price surfaces that remain within bid-ask intervals and satisfy standard no-arbitrage restrictions across strikes before computing Greeks, hedge quantities, or risk measures. For the same quote panel, however, these constraints may admit multiple such surfaces, which we call admissible repairs, with different downstream outputs. We define the resulting variation as repair risk and propose repair-risk validation, a matched framework for quantifying dispersion in Greeks, hedge quantities, and risk measures. The mechanism is finite-difference amplification: two admissible repairs within price distance $\varepsilon$ can differ in local normalized-strike curvature by $4\varepsilon/h^2$, where $h$ is local normalized-strike spacing. We evaluate repair risk on Deribit BTC and ETH option chains, using $L_1$ and $L_2$ projections of midpoint prices onto the same admissible set as matched controls. Across 108 matched evaluations spanning 18 sampled calendar months, the median 90th percentile of the scaled spot-gamma gap for $L_2$ versus the $ρ=10$ smoothness-penalized repair is nearly twenty times the matched $L_1$--$L_2$ control. We also instantiate NN-Proj, a neural-projection repair, to test whether learned repairs exhibit repair risk. Despite good held-out fit, its downstream outputs remain distinct from those of classical repairs. Across five seeds with the same training window and architecture, its curvature dispersion exceeds the matched $L_1$--$L_2$ control in 17 of 18 months ($p=1.45\times10^{-4}$). For a USD 1 million risk reversal, switching from $L_2$ to the $ρ=10$ repair shifts the median 90th-percentile hedge notional by 18.62 basis points of gross spot notional. VaR-style estimates also exhibit greater dispersion across repairs. The repair rule is therefore part of the risk model, not merely a feasibility check.

q-fin.PR↗

Supply Chain Propagation of Textual Signals: LLM Embeddings and Cross-Sectional Return Predictability

This paper proposes a novel asset pricing framework that augments large language model (LLM) embeddings of annual report disclosures with supply chain knowledge graph (KG) propagation. Using FinBERT embeddings of 10-K MD&A sections for 255 S&P 500 firms over 2011-2025, two sets of return predictors are constructed: direct LLM embeddings and network-augmented embeddings, where firm-level signals propagate through inter-firm linkages. Fama-MacBeth cross-sectional regressions reveal that the network-augmented factor (net_pc_5) carries significant return predictability with a Newey-West t-statistic of -2.64, even after controlling for momentum, volatility, and firm size. A long-short portfolio sorted on net_pc_5 achieves an annualized Sharpe ratio of 0.86 and a Fama-French five-factor alpha of 7.27% per year (t = 2.30). The predictive power survives out-of-sample tests, placebo experiments, sector-neutralization, and subsample analysis. The findings suggest that inter-firm network structure contains pricing-relevant information beyond firm-level textual disclosures.

q-fin.PR↗

Reaction-boundary variance and adjoint-consistent local-volatility projection

We derive an operational-time variance kernel for a latent-order-book reaction boundary and use it to separate three objects usually collapsed in calendar-time volatility models: a structural boundary cumulant, a clock projection, and a pricing-measure choice. The reaction boundary is the zero of a bid--ask imbalance field. For a locally linear book, signed order-flow perturbations displace this zero through a damped Abel response kernel, so the variance of boundary increments is obtained as a finite-scale Green-function cumulant rather than introduced as a primitive diffusion coefficient. For long-memory forcing with exponent $0<γ<1$, the operational variance has a closed asymptotic form involving effective signed-forcing intensity, liquidity slope, resilience, memory, and operational coarse-graining scale. A deterministic activity clock gives the benchmark local-volatility projection. More general, non-unique clocks generate candidate calendar-time pricing systems. We argue that such projections are admissible only when the induced forward density operator and backward valuation operator remain adjoint on the same state space. Adjoint consistency is therefore a reality constraint on operational-to-calendar time projection: it disciplines non-unique time and identifies where incompleteness enters.

q-fin.PR↗