CDAR Co-Director Lisa R. Goldberg to present a seminar titled “Identifying Financial Risk Factor with Sparse and Low-Rank Decompositions” (link to page).
Abstract: We show how to use sparse and low-rank (SLR) matrix decompositions based on convex optimization to extract financial risk factors from a sample return covariance matrix. We provide an example that highlights the difference between this approach and the academic standard for financial factor identification, principal component analysis (PCA), which makes systematic errors. Using finance-oriented metrics, we analyze the accuracy of SLR and PCA on equally weighted portfolios and minimum variance portfolios in a simulated global equity market. Finally, we discuss non-convex programming formulations that show promise in identifying numerous sparse factors (industries, counties, etc) at various scales. A preprint that gives some background on what we’re up to is linked here: https://papers.ssrn.com/sol3/papers2.cfm?abstract_id=2800237 and more information can be found ion this page: http://cdar.berkeley.edu/research/risk-factors-and-low-rank-sparse-decompositions/Download the slides from this presentation: slr2017