Factor Clustering with t-SNE

Abstract

We cluster asset pricing factors using the t-distributed Stochastic Neighborhood Embedding (t-SNE), one of the most empirically successful dimensionality reduction techniques. t-SNE endogenously separates the strategies into six distinct clusters. The first five clusters resemble the standard value, momentum, investment, profitability, and volatility strategies. The sixth cluster is new, and we denote it as the Firm cluster. We show that the first five clusters are low dimensional and are dominated by their corresponding first principal components, while the Firm cluster is intrinsically high dimensional.