Finance-Informed Neural Network: Learning the Geometry of Option Pricing¶
Bibliographic record. Follow the original-source link for the publication.
| Field | Value |
|---|---|
| Primary domain | Volatility |
| Other domains | Hedging Exposure Risk |
| Methods | Financial Ml, Research Methods |
| Facets | — |
| Authors | Amine M. Aboussalah, Xuanze Li, Cheng Chi, Raj Patel |
| Published | 2026-03-11 |
| Source | arXiv Quantitative Finance History |
| Identifiers | arxiv:2412.12213 |
| URL | Open original source |
Editorial synthesis¶
Why it matters¶
FINN reframes pricing and hedging as learning a replication-consistent pricing operator, which is significant for reducing parametric misspecification risk and extending to illiquid markets, if trainability and calibration are sound. (abstract:S1, abstract:S2, abstract:S3, abstract:S4, abstract:S6, abstract:S7)
Main author claims¶
- The authors claim FINN is learned through a self-supervised replication objective based on dynamic hedging, integrating financial structure into the neural network. (
abstract:S1,abstract:S2) - They claim minimizing replication error recovers the arbitrage-free pricing operator, yields economically meaningful sensitivities, and remains stable in settings without reliable closed-form solutions. (
abstract:S3,abstract:S5,abstract:S4) - The authors claim FINN reconstructs implied-volatility surfaces closer to market values than Heston calibrations and can use historical spot prices to construct coherent option prices and Greeks for assets without listed options. (
abstract:S6,abstract:S7)
Data, method, or discussion scope¶
Scope includes theoretical objectives, replication consistency claims, recovery of Black-Scholes behavior, Heston robustness, implied-volatility surface performance versus Heston calibrations, and training on historical spot for illiquid assets. Hyperparameters and cost model integration are not provided. (abstract:S1, abstract:S2, abstract:S3, abstract:S4, abstract:S6, abstract:S7)
Main limitations¶
The claim of improved IV surface reconstruction lacks explicit metric definitions and windows, and abstract-level reporting does not specify transaction-cost treatment, liquidity thresholds, or walk-forward drawdown checks. (abstract:S4, abstract:S6, abstract:S9)
Relationships¶
- None recorded.