English

Optimal cleaning for singular values of cross-covariance matrices

Statistics Theory 2021-11-19 v6 Probability Statistics Theory

Abstract

We give a new algorithm for the estimation of the cross-covariance matrix EXY\mathbb{E} XY' of two large dimensional signals XRnX\in\mathbb{R}^n, YRpY\in \mathbb{R}^p in the context where the number TT of observations of the pair (X,Y)(X,Y) is large but n/Tn/T and p/Tp/T are not supposed to be small. In the asymptotic regime where n,p,Tn,p,T are large, with high probability, this algorithm is optimal for the Frobenius norm among rotationally invariant estimators, i.e. estimators derived from the empirical estimator by cleaning the singular values, while letting singular vectors unchanged.

Keywords

Cite

@article{arxiv.1901.05543,
  title  = {Optimal cleaning for singular values of cross-covariance matrices},
  author = {Florent Benaych-Georges and Jean-Philippe Bouchaud and Marc Potters},
  journal= {arXiv preprint arXiv:1901.05543},
  year   = {2021}
}

Comments

38 pages, 8 figures, 2 tables. In v8: numerical simulations section widely extended, intro partly rephrased, section about overfitting extended, typos corrected, link to github repo for python codes added