Mathematical Finance Seminar
March 7, 2002 , 5:30 PM to 7:00 PM
Francois Oustry, Raise Partner
Spectral Covariance Matrix Correction and Applications in Finance
Existing quantitative portfolio management softwares, based on
traditional Mean-Variance approaches, often lead to various numerical
instabilities which are not acceptable for an investment manager.
Using a Von Neumann-Kato covariance matrix correction step as well as
an efficient dual semidefinite programming scheme, we show that
stable and well-conditioned risk estimators can be obtained.
Some practical results will also be presented to illustrate the impact
of the correction on the efficiency as well as the sensivity of
constructed portfolios. A particular focus is proposed to see how to use
the correction step to get a robust implementation of a Black-Litterman
type global asset allocation model.