Skip to content

Multivariate Realized Volatility Forecasting with Graph Neural Network

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

Field Value
Primary domain Volatility
Other domains
Methods Financial Ml
Facets
Authors Qinkai Chen, Christian-Yann Robert
Published 2021-12-17
Source arXiv Quantitative Finance History
Identifiers arxiv:2112.09015
URL Open original source

Editorial synthesis

Why it matters

Integrating relational structure with limit-order-book data may improve multivariate short-term volatility forecasts; the abstract does not show that these forecasts translate into tradable signals. (abstract:S1, abstract:S2, abstract:S3, abstract:S4, abstract:S5, abstract:S6)

Main author claims

  • They introduce a Graph Transformer Network for short-term realized-volatility forecasting in a multivariate, order-book-based framework. (abstract:S3, abstract:S4)
  • The authors report: The model combines limit order book features with temporal and cross-sectional relations from multiple sources. (abstract:S5)
  • The authors report: Experiments on about 500 stocks from the S&P 500 report better performance than benchmarks. (abstract:S6)

Data, method, or discussion scope

Scope is multivariate short-horizon realized-volatility forecasting over roughly 500 S&P 500 stocks using graph-transformer models and LO B inputs. (abstract:S3, abstract:S5, abstract:S6)

Main limitations

Graph construction, benchmark definitions, and computational constraints are not specified, and external generalization checks are absent. (abstract:S4, abstract:S5, abstract:S6)

Relationships

  • None recorded.