Covariance matrix estimation represents a challenging topic for many research fields. Sample covariance matrix might perform poorly in many circumstances, especially when the number of variables is approximately equal or greater then the number of observations. Moreover, when the precision matrix is the object of interest the sample covariance matrix might not be positive definite and more robust estimators must be used.
With this article I will try to give a brief (and non-comprehensive) overview of some of the topics in this research field. In particular, I will describe Stenian shrinkage, covariance matrix selection through penalized likelihood and graphical lasso implementing the description with some potential extensions of these methodologies… Read the full post by Davide Viviano ’17 on Barcelona GSE Data Scientists