Graph-Guided Time Series Generation with Applications to Financial Correlation Modeling

Howard Caulfield, James Gleeson

Keywords: Graph Neural Networks, Synthetic Data, Time Series, Finance

TL;DR: Graph-Guided time series generation offers dependency control to high dimensional and complex time series.

Abstract: Generating multivariate time series that maintain desired characteristics while controlling the dependency structure is an open challenge. We propose a graph-guided time series generation model that generates both desired node and hierarchical dependency structures. We highlight the effectiveness of the method through pricing a correlation trade for the US Presidential Election.

Submission Type: Extended abstract (max 4 main pages).

Submission Number: 33