Algorithms for Learning Graphs in Financial Markets

  • 2020-12-31 02:48:35
  • José Vinícius de Miranda Cardoso, Jiaxi Ying, Daniel Perez Palomar
  • 24

Abstract

In the past two decades, the field of applied finance has tremendouslybenefited from graph theory. As a result, novel methods ranging from assetnetwork estimation to hierarchical asset selection and portfolio allocation arenow part of practitioners' toolboxes. In this paper, we investigate thefundamental problem of learning undirected graphical models under Laplacianstructural constraints from the point of view of financial market times seriesdata. In particular, we present natural justifications, supported by empiricalevidence, for the usage of the Laplacian matrix as a model for the precisionmatrix of financial assets, while also establishing a direct link that revealshow Laplacian constraints are coupled to meaningful physical interpretationsrelated to the market index factor and to conditional correlations betweenstocks. Those interpretations lead to a set of guidelines that practitionersshould be aware of when estimating graphs in financial markets. In addition, wedesign numerical algorithms based on the alternating direction method ofmultipliers to learn undirected, weighted graphs that take into accountstylized facts that are intrinsic to financial data such as heavy tails andmodularity. We illustrate how to leverage the learned graphs into practicalscenarios such as stock time series clustering and foreign exchange networkestimation. The proposed graph learning algorithms outperform thestate-of-the-art methods in an extensive set of practical experiments.Furthermore, we obtain theoretical and empirical convergence results for theproposed algorithms. Along with the developed methodologies for graph learningin financial markets, we release an R package, called fingraph, accommodatingthe code and data to obtain all the experimental results.

 

Quick Read (beta)

loading the full paper ...