Deep Learning Option Pricing with Market Implied Volatility Surfaces¶
Bibliographic record. Follow the original-source link for the publication.
| Field | Value |
|---|---|
| Primary domain | Volatility |
| Other domains | — |
| Methods | Financial Ml |
| Facets | Instrument Index Options |
| Authors | Lijie Ding, Egang Lu, Kin Cheung |
| Published | 2025-09-07 |
| Source | arXiv Quantitative Finance History |
| Identifiers | arxiv:2509.05911 |
| URL | Open original source |
Editorial synthesis¶
Why it matters¶
This preprint compresses implied-volatility surfaces with a VAE and combines latent factors with an MLP pricer, potentially changing computational cost and valuation throughput for American and Asian options, though the claims remain abstract-level. (abstract:S1, abstract:S3, abstract:S4, abstract:S5, abstract:S7, abstract:S8)
Main author claims¶
- The authors claim to construct arbitrage-free volatility surfaces from EOD S&P 500 options and compress cross-maturity/strike surfaces into a 10-dimensional VAE latent space. (
abstract:S2,abstract:S3) - They claim staged training with final end-to-end fine-tuning yields an efficient one-pass pricer and achieves high accuracy for American and Asian options, with errors concentrated in long maturities and ATM strikes. (
abstract:S4,abstract:S5,abstract:S6,abstract:S7)
Data, method, or discussion scope¶
The scope covers stated architecture and training flow only; no explicit loss metric definitions, baseline definitions, compute budget, or data-cleaning rules are provided. (abstract:S1, abstract:S2, abstract:S3, abstract:S4, abstract:S5, abstract:S6, abstract:S7, abstract:S8)
Main limitations¶
Accuracy is reported without error distributions or significance thresholds; failure modes in input drift, reconstruction anomalies, and production monitoring thresholds are not defined. (abstract:S6, abstract:S7, abstract:S5, abstract:S1)
Relationships¶
- None recorded.