Speaker:

David Morton (Swissquote - EPFL)

Title:

Weighted covariance and correlation matrices: old friends and power-estimators

Abstract:

We analyze in the same work the dynamical and spectral properties of time-weighted covariance and correlation matrices. The results of this approach lead us to introduce a class of covariance and correlation estimators with power-law weighting. We investigate theoretically and by simulation the spectral and temporal characteristics of the so-called power-estimators and compare them with current standards, such as the rectangular and exponentially weighted (EWMA) matrices. Power-estimators are shown to capture the long memory of the return volatility, and feature better conditioning and a narrower spectral noise band than exponentially weighted matrices. These features, along with a small number of fitting parameters, makes this class of estimators particularily suited for financial risk assessment and portfolio allocation.