The aim of this project is to show empirically the importance of covariance filtering in the prediction of the risk of a portfolio. To this extent, we compare the out-of-sample risk of the Global Minimum Variance Portfolio. We first compute it naively and then with each of the following methods: eigen values clipping, Rotationally Invariant Estimators (RIE), Average Oracle (AO) and Bootstrap Average linkage Hierarchical Clustering (BAHC). We analyze also the Optimal Mean-Variance portfolio performance for each of the mentioned techniques.

The dataset that we use for the analysis consists of daily close-to-close return observation of the US equity markets data for 42 years: from 01-01-1980 to 31-03-2022. We select the 1135 stocks with the largest market capitalization using WRDS. Therefore, in total, we have 5598 rows and 1135 columns.