Master thesis project. The improved estimator of the covariance matrix of asset returns is employed to derive a new trading strategy based on a two-step procedure. First, it shrinks the asset universe via a subset selection, leaving only the most suitable assets. Then, it performs the mean-variance analysis. Back-testing is carried out in the U.S. stock market between 2018 and 2020. For comparison purposes, the code also implements also other strategies, such as the widely-used momentum strategy. The proposed technique is observed to deliver a very good and much more stable performance with respect to its competitors.