Résumé (EN)

What is the optimal number of uncorrelated strategies to include in a portfolio consisting of cross-asset strategies? Various criteria have been proposed for finding the optimal number of factors in a factor model, many of them in the framework of Principal Component Analysis. Using a refined information criterion, the authors estimate the number of factors associated with various asset classes. In the case of U.S. Treasury Bonds, the authors find the usual three factors, related to level, slope, and term spread, while credit spreads are affected by just one factor. Commodity prices are affected by two common factors, and currencies by only one. Bringing all these assets together, the authors find that a total of five factors are at work. The authors identify and interpret these factors, and investigate their stability over time by testing the significance of the correlations among the factors over a rolling window. During recessions, the number of uncorrelated strategies drops for U.S. Treasury Bond rates. For commodities, however, the concentration of correlation is weaker. This provides evidence to support their diversification potential, even during economic downturns.