In this talk we show that the issue of calibrating covariance matrices subject to spectral constraints arising from : a) noise filtering and conditioning requirements and b) prior views (or intervals) on volatilities and/or correlations, can be solved efficiently using semidefinite programming. We also explain how these spectral filtering techniques can be articulated with Ledoit-Wolf shrinkage approaches.

The resulting semidefinite program is solved using an efficient dual algorithm. A dynamic covariance-driven investment strategy over a multi-class investment universe is presented as a practical application.

François Oustry, CEO, RaisePartner (Paris-Grenoble, New-York, Dubai). Since incorporation of RaisePartner in France on July 4th 2001, François Oustry has been positioning RaisePartner as a pioneer company in Active Risk Management. In complement to traditional Risk Compliancy approaches, RaisePartner has been enabling hedge funds, asset management firms, family offices or sovereign funds to build “on the flight dedicated ex-ante risk models”. Rather than trying to obtain ex-post absolute risk figures, RaisePartner aims at detecting soon enough “firm-sensitive” market changes. Generic risk factors usually lead to strong similarities in ex-ante risk exposures between all market actors. Hence François Oustry and his team have bet on a new avenue: enabling market actors to build their dedicated and robust risk models thanks to user-friendly and fully interpretable risk modelling & optimization tools. Before founding RaisePartner, François Oustry was a researcher from 1999 to 2001 at the Fr e nch National Institute for Research in Computer Science and Control (INRIA). During his post-doctoral year in 1998 at the Courant Institute of Mathematical Sciences (New-York University), François has been advising several major Wall-Street firms and has paid a particular attention to the differentiation capabilities brought from Robust Control & Optimization techniques acquired at Stanford University (Information Systems Lab)."

Véronique Piolle is RaisePartner Head of Quant Consulting; she leads risk and optimization modeling specific work for RaisePartner clients: prop trading desks within investment banks, quantitative hedge funds or quant teams within pension funds in Europe. Véronique also collaborates with RaisePartner R&D Product Group on RaisePartner proprietary quant algorithms. Her last publication is "Semidefinite optimisation for global risk modeling, Journal of Asset Management 7, 142-153, 2006".

Paper: François Oustry, Véronique Piolle, Convex Filtering of Covariance Matrices

Additional Papers

UTS Presentation: François Oustry, Convex Optimization in Quantitative Finance
Q-Group Presentation: François Oustry, Véronique Piolle, Designing "Global Macro" Investment Strategies with Asymmetry Preferences