Online Adaptive Machine Learning Based Algorithm for Implied Volatility Surface Modeling¶
Bibliographic record. Follow the original-source link for the publication.
| Field | Value |
|---|---|
| Primary domain | Volatility |
| Other domains | — |
| Methods | Financial Ml |
| Facets | — |
| Authors | Yaxiong Zeng, Diego Klabjan |
| Published | 2017-06-06 |
| Source | arXiv Quantitative Finance History |
| Identifiers | arxiv:1706.01833 |
| URL | Open original source |
Editorial synthesis¶
Why it matters¶
The study targets implied-volatility surfaces under pattern drift with an SVR that updates support vectors online and accelerates prediction on an FPGA. (abstract:S1, abstract:S2, abstract:S3, abstract:S4)
Main author claims¶
- The authors propose an online adaptive primal kernel SVR that uses local fitness and budget maintenance to update support vectors under pattern drift. (
abstract:S1,abstract:S2,abstract:S3) - The authors report a 132-fold FPGA speedup over CPU for the most computationally intensive parts of online prediction. (
abstract:S4) - The authors report: Using E-mini S&P 500 options tick data, the authors report that the Gaussian kernel better controls support-vector size than the linear kernel, that their method outperforms two online alternatives on complexity and regression error, and that results are better at the center than at the edges of the surface grid. (
abstract:S5,abstract:S6,abstract:S7)
Data, method, or discussion scope¶
The material covers the online SVR design, FPGA prediction acceleration, E-mini S&P 500 options intraday tick data, two online baselines, and hyperparameter sensitivity. (abstract:S1, abstract:S2, abstract:S3, abstract:S4, abstract:S5, abstract:S6, abstract:S7)
Main limitations¶
The 132-fold speedup applies only to intensive parts of online prediction; the empirical evidence names one options market, and weaker edge-of-grid results limit generalization to full-surface deployment. (abstract:S4, abstract:S5, abstract:S6)
Relationships¶
- None recorded.