Dynamic portfolio selection with sector-specific regularization

This paper proposes a new algorithm for dynamic portfolio selection that takes a sector structure into account. We consider regularization with respect to within and between sector variation of portfolio weights, additional to sparsity and trans- action cost controls. Our model includes two special cases as benchmarks: a dy- namic conditional correlation model with shrinkage estimation of the unconditional covariance matrix, and the equally weighted portfolio. We propose an algorithm for estimation of the model parameters and calibration of the penalty terms based on cross-validation. In an empirical study, we find that the within-sector penalty has by far the highest contribution to the reduction of out-of-sample volatility of portfolio returns. Our model improves both the pure DCC with shrinkage and the equally-weighted portfolio out-of-sample.