Optimal Factor Portfolio : A Dual-Sparse Method

This paper constructs a dual-sparse optimal mean-variance factor portfolio to improve the interpretation of the factor portfolio in a set of 187 anomaly portfolios. The dual-sparse method limits the nonzero elements of principal components (PCs) and number of factor invested in cross-section. We first impose substantive sparsity on the PCs with only around 20 elements, or 10% of underlying factors, and high degree of sparsity facilitates to interpret their economic meaning at little cost of the variance explained. We then shrink the cross-section of portfolio with at most 6 of sparse PCs and the portfolio can almost approximate the optimal mean-variance (MV) portfolio. The interpretable factor portfolio performs well in the out-of-sample period with an annualized Sharpe ratio 0.75 and is robust with different datasets