High-dimensional Dynamic Portfolio Selection with Machine Learning

Work in progress

This paper extends the static Markowitz portfolio choice to a dynamic and conditional problem by modeling portfolio weights as a function of characteristics, and incorporates a LASSO-type penalty to select the high-dimensional function coefficients. It outperforms the dynamic portfolio choice problem without a LASSO penalty and the benchmark of static and equally weighted portfolios in terms of out-of-sample Sharpe ratios.