Equity factor analysis via column subset selection

Modern finance has grown increasingly high-dimensional, with tens of thousands of stocks and bonds and other more complex instruments that are the basic units of strategies for hedging, risk management, and investment. The most popular way to understand this intimidating complexity has been through factor models, which decompose the whole universe of investment instruments into a few key drivers. The two main approaches to factor analysis are fundamental, where analysts hand-pick a set of key drivers, and statistical, where algorithmic techniques such as Principal Component Analysis (PCA) automatically determine what are the key drivers. The shortcoming of the fundamental approach is not being data-adaptive, while the statistical approach is not interpretable and does not lead to easy hedging strategies. We suggest an alternative approach to factor analysis, relying on column subset selection, which keeps the interpretability of the fundamental approach and the data-adaptivity of the statistical PCA-based approach.