We look at the performance of daily re-weighted GMV portfolios created using Projected- PCA, Robust PCA, Risk-Premium PCA and the POET thresholding method. The first three methods are estimators for latent factors in approximate factor models, and the last method is an estimator for the covariance matrix of the residuals in approximate factor models. We combine the methods to obtain the covariance matrices of high-dimensional sets of returns and use these matrices to construct GMV portfolios. The portfolios are compared based on their respective Sharpe ratios to discover which method or combination of methods provides the best performing portfolios. Both an empirical and a simulation study are carried out to investigate which method produces the best-performing portfolios. The studies both show that the portfolios with the highest Sharpe ratios are created using factors estimated by the Robust PCA method.

Additional Metadata
Keywords Approximate factor models, principal component analysis, optimal number of factors, eigenvalue decomposition, thresholding, global minimum variance portfolios
Thesis Advisor Grith, M.
Persistent URL hdl.handle.net/2105/52167
Series Econometrie
Citation
Hagen, R. (2020, May 28). Portfolio Allocation in High-Dimensional Data Sets using Approximate Factor Models. Econometrie. Retrieved from http://hdl.handle.net/2105/52167