Covariance matrix filtering with bootstrapped hierarchies

Cleaning covariance matrices is a highly non-trivial problem, yet of central importance in the statistical inference dependence between objects. We propose here probabilistic hierarchical clustering method, named Bootstrapped Average Hierarchical Clustering (BAHC), that particularly effective high-dimensional case, i.e., when there are more objects than features. When applied to DNA microarray, our method yields distinct structures cannot be accounted for by usual clustering. then use global minimum-variance risk management test and find BAHC leads significantly smaller realized compared state-of-the-art linear nonlinear filtering methods case. Spectral decomposition shows better captures persistence structure asset price returns calibration periods.