Incorporating prior financial domain knowledge into neural networks for implied volatility surface prediction

Bibliographic record. Follow the original-source link for the publication.

Field	Value
Primary domain	Volatility
Other domains	—
Methods	Financial Ml
Facets	—
Authors	Yu Zheng, Yongxin Yang, Bowei Chen
Published	2021-05-28
Source	arXiv Quantitative Finance History
Identifiers	arxiv:1904.12834
URL	Open original source

Editorial synthesis

Why it matters

The study embeds smile shape, no-arbitrage boundaries, and asymptotic-slope priors in an IV-surface network, enabling comparison with unconstrained data-driven models; loss penalties do not amount to a strict theoretical guarantee. (abstract:S5, abstract:S6, abstract:S7)

Main author claims

• The authors report: A novel neural network model is proposed that incorporates prior domain knowledge through a volatility-smile activation function and embedding arbitrage-free conditions in the loss function. (abstract:S3, abstract:S4)
• The authors report: The proposed model outperforms benchmark models on 20 years of S&P 500 index option data. (abstract:S6)
• The authors report: The model empirically satisfies the embedded domain knowledge conditions, showing consistency with existing financial theories. (abstract:S7)

Data, method, or discussion scope

Uses 20 years of S&P 500 index option data to predict the implied volatility surface. Compares against benchmark models (not named). The focus is on architectural innovation with domain knowledge embedding and ex-post validation of theoretical consistency. (abstract:S6, abstract:S7)

Main limitations

The abstract does not specify the types of benchmark models, the error metrics used, or the incremental contribution of constraints to out-of-sample prediction. The study focuses on the S&P 500 index, with unknown generalizability to other asset classes. Adding domain knowledge may increase model complexity and implementation difficulty. (abstract:S6, abstract:S4)

Relationships

• None recorded.