Peer Group Identification in Factor Portfolios: A Data-Driven Approach

Are factor characteristics more informative when compared with the entire investment universe or a relevant subset of peers? Motivated by a belief that the answer is dependent on the identity of the peer groups used, this article provides a novel perspective on this longstanding question by using clusters derived from stock returns in place of the industrial and geographical peer groups typically used by investors. The author presents empirical results in support of the use of return-derived clusters, with a key finding being that the optimal set of peer groups varies by investment universe and period and that standard classification taxonomies that fail to account for these nuances are, on average, inferior to a simple data-driven approach that does take them into account.