Graph Learning in Finance

Graphs of cryptos and fiat currencies

Financial markets are inherently interconnected systems where assets, institutions, and market participants form complex networks of relationships that evolve over time. Graph learning has emerged as a powerful framework for understanding and modeling these financial networks, enabling the discovery of meaningful relationships among stocks, cryptocurrencies, or other financial instruments through their return time series, trading volumes, or other market indicators. By representing financial data as nodes in a graph and learning the edge weights that capture dependencies and spillover effects, graph learning techniques can reveal market sectors, risk propagation channels, and trading communities that may not be apparent from traditional financial analysis. This network perspective is particularly valuable for portfolio optimization, systemic risk assessment, and market surveillance, where understanding the underlying structure of financial relationships can lead to more robust investment strategies and better risk management practices.

Software

GitHub software webpage

  • spectralGraphTopology: Structured graph learning via Laplacian spectral constraints (NeurIPS 2019) [CRAN]
  • sparseGraph: Nonconvex Sparse Graph Learning under Laplacian-structured Graphical Model (NeurIPS 2020)
  • fingraph: Graphical Models in Heavy-Tailed Markets (NeurIPS 2021) [CRAN]
  • bipartite: Learning Bipartite Graphs: Heavy Tails and Multiple Components (NeurIPS 2022) [CRAN]

Book chapters

Papers