High-dimensional Portfolio Choice using Graphical Lasso

Ruochen Li
This paper studies a high-dimensional portfolio choice problem using a machine learning method Graphical Lasso. It considers a 60-asset portfolio with 49 equities and 11 bonds. It compares the proposed method Graphical Lasso to four other popular alternative methods, Equal-Weight portfolio, Sample-Covariance portfolio, Linear and Non-linear-Shrinkage portfolio. We produce five out-of-sample predictions to compare the performance of the five methods, with a variation of time horizons, and
more » ... cing frequencies. The results show that Graphical Lasso outperforms all other methods except Equal-Weight portfolio consistently and outperforms all methods in most of the cases.
doi:10.17615/0gy3-xa70 fatcat:632j24qt5nfohos7hygutzhyg4