A Robust Latent Factor Model for High-Dimensional Portfolio Selection

Portfolio selection is often faced with large noisy data sets of strongly correlated asset returns, and so is prone to unstable portfolio weights and serious estimation error. To attenuate these problems, this paper proposes a new latent factor model equipped with both a suitable robust estimator to deal with cellwise data contamination and a diagonally-dominant (DD) covariance structure to account for cross-sectional dependence among idiosyncratic components. The robust DD estimator is computationally efficient in high-dimensional portfolio formation. Our proposed model is found to compare favorably with a variety of state-of-the-art competitors in terms of its better out-of-sample performance across a wide range of real-world data sets