Machine learning for option pricing: an empirical investigation of network architectures¶
Bibliographic record. Follow the original-source link for the publication.
| Field | Value |
|---|---|
| Primary domain | Volatility |
| Other domains | — |
| Methods | Financial Ml |
| Facets | — |
| Authors | Serena Della Corte, Laurens Van Mieghem, Antonis Papapantoleon, Jonas Papazoglou-Hennig |
| Published | 2026-01-29 |
| Source | arXiv Quantitative Finance History |
| Identifiers | arxiv:2307.07657 |
| URL | Open original source |
Editorial synthesis¶
Why it matters¶
This preprint studies how architecture choice affects accuracy and training time in option-pricing tasks, informing model-family selection while leaving parameter-budget fairness and training stability unspecified. (abstract:S1, abstract:S2, abstract:S3, abstract:S4, abstract:S5, abstract:S6, abstract:S7)
Main author claims¶
- The authors claim that much of the literature uses plain feedforward networks, and they investigate whether architecture choice affects accuracy and training time. (
abstract:S1,abstract:S2,abstract:S3) - They claim generalized highway networks perform best under MSE and training-time criteria for Black-Scholes/Heston pricing within their budgets, while simplified DGM gives lowest error for transformed implied-volatility tasks. (
abstract:S4,abstract:S5)
Data, method, or discussion scope¶
The scope includes architecture comparison claims, metrics, parameter budgets, and a mention of real-market-data experiments for implied-volatility, without explicit hyperparameters or significance metrics. (abstract:S1, abstract:S2, abstract:S3, abstract:S4, abstract:S5, abstract:S6, abstract:S7)
Main limitations¶
Best-performance claims depend on unspecified budget and task definitions, and no seeding/stability or cross-asset transfer procedures are reported. (abstract:S4, abstract:S6, abstract:S7, abstract:S3)
Relationships¶
- None recorded.