Great article, Joseph.

I am wondering if the marginal benefits of the Size factor is due to its conditional efficacy for the other factors. E.g. allowing a portfolio to have a small Size tilt normally can improve signal efficacy (and Sharpe ratio) for major style factors.

An easy test would be running 2 additional optimizations: 1. maximizing Sharpe using the set of factors in your chart except for Size, and do not specify a Size constraint. 2. maximizing Sharpe using the set of factors in your chart except for Size, but include a Size neutral constraint. If there's a significant difference between 1 and 2, and the Size exposures (measured in Zscore rather than weights) are similar between 1 and your original analysis, then it's more likely to be the conditional efficacy effect.

Thanks,

Patrick