Partitioning the Cross-Section of Stocks: The Economic Value of Statistical Clusters

Title: Partitioning the Cross-Section of Stocks: The Economic Value of Statistical Clusters

Abstract: What is the optimal partition of the cross-section of stocks? I argue that this is an `ill-posed' problem as the ground truth is not known. However, given a benchmark (industries) and embedded into an economic context (return predictability), one can assess the economic value of a clustering. I use a simple, transparent and well-understood clustering method, k-means, and show that the induced partition has a large economic and statistically significant value that is not captured by industry membership. I term this signal the pure cluster signal, which is related to proxies for investor attention and limits to arbitrage.