Asset allocation with factor-based covariance matrices
研究了基于因子模型的协方差矩阵能否提升最小方差投资组合的表现，发现因子模型预测更准，因子组合在标准差和夏普比率上显著优于多个基准，考虑交易成本后静态协方差矩阵的风险调整收益更高。
• A factor-based framework improves minimum-variance portfolio performance. • Latent factors based on machine learning methods enhance covariance matrix estimates. • Portfolios based on dimensionality reduction methods outperform several benchmarks. • Type of structure imposed raises trade-offs key to optimal portfolio performance. We examine whether a factor-based framework to construct the covariance matrix can enhance the performance of minimum-variance portfolios. We conduct a comprehensive comparative analysis of a wide range of factor models, which can differ based on the machine learning dimensionality reduction approach used to construct the latent factors and whether the covariance matrix is static or dynamic. The results indicate that factor models exhibit superior predictive accuracy compared to several covariance benchmarks, which can be attributed to the reduced degree of over predictions. Factor-based portfolios generate statistically significant outperformance with respect to standard deviation and Sharpe ratio relative to multiple portfolio benchmarks. After accounting for transaction costs strategies based on static covariance matrices outperform dynamic specifications in terms of risk-adjusted returns.