Robust portfolio construction using sorting signatures

Mean-variance analysis introduced by Harry Markowitz has been criticised in the past mainly due to the counter-intuitive and unstable nature of the resultant portfolios from the optimisation. These disappointing results have been linked to the presence of estimation error in the estimates of the expected returns and covariances which serve as input to the optimisation. Several attempts have been made to produce more reliable estimates, with a significant amount of effort and resources placed in estimation of expected returns, which is generally a more difficult task than estimation of covariances. Almgren and Chriss (2006) provide a methodology for portfolio selection in which the order of expected returns replaces the numerical values of the returns. This framework allows full use of the covariance matrix, in a method analogous to mean-variance optimisation. We adopt this framework in our analysis together with the robust optimisation technique introduced by Golts and Jones (2009) which improves the estimate of the covariance matrix by direct modification in the optimisation process. Golts and Jones (2009) argue that a reduction of the angle between the input return forecasts and the output portfolio positions results in more investment relevant portfolios, inline with the investment manager's insights. They relate this angle to the condition number of the covariance matrix and use robust optimisation to improve the conditioning of this matrix. Assuming perfect alpha foresight of an investment manager, we apply a combination of the techniques of Almgren and Chriss (2006) and Golts and Jones (2009) to South African equity data and show that the resultant robust portfolios, though conservative in their risk-adjusted return statistics, are more diversified and exhibit lower leverage than mean-variance portfolios. We further show that independent of the optimisation method, there is a marginal difference in the performance of portfolios created using ordering information and actual returns.

Grantor dc:publisher.institution

Division of Actuarial Science

Year dc:date.issued

2014

Author dc:creator

• Kasenene, Lillian

Advisors dc:contributor.advisor

• Bradfield, Dave
• Bailey, Geraldine

Language dc:language.iso

eng

Handle dc:identifier.uri

http://hdl.handle.net/11427/8527

OAI identifier oai:identifier

oai:open.uct.ac.za:11427/8527