Financial Networks with the Graphical-2-Lasso

• Activity Number: 179
• Type: Contributed
• Date/Time: Monday, August 1, 2011 : 10:30 AM to 12:20 PM
• Sponsor: ENAR
• Author(s): Alan Burton Lenarcic*+
• Companies: The University of North Carolina at Chapel Hill
• Keywords: Bayes ; networks ; lasso ; finance ; correlation ; algorithms

Abstract

Advances in network and model selection motivated by L1 convex algorithms now provide penalized likelihood estimators to study the realm of partial correlation and conditional independence in complex networks. The 2Lasso method was motivated to improve Lasso false positive rate with negligible additional computational complexity, giving Lasso an informative Bayesian latent data model that solves the question of noise coefficient thresholding. Thus 2Lasso is also applicable to graphical-lasso network inspection, and provides structure to avoid issues like correlation bias, that could hurt glasso in an arena like portfolio allocation, where accurate correlation estimates are essential. Incidences of financial contagion, or spikes in asset correlation, suggest moments where investors might need improved regularized covariance estimation. Anomalies like the May 5th 2010 crash, driven in part through algorithmic trading, show that these decisions need to be made quickly, often in milliseconds. Lasso's robustness, speed, well documented k < n < p performance, make it a promising candidate; oddly enough it can serve up Bayes confidence measures too!