Robust Modeling for Financial Portfolio Selection

• Activity Number: 415
• Type: Contributed
• Date/Time: Wednesday, August 6, 2003 : 2:00 PM to 3:50 PM
• Sponsor: Business & Economics Statistics Section
• Author(s): Simona Costanzo*+
• Companies: London School of Economics
• Keywords: robust ; outliers ; portfolio selection models ; covariance matrices

Abstract

Despite the old age and the negative criticism concerning Mean-Variance efficient models for selecting financial portfolios, they are still widely used in practice. These methods require estimates for the asset return scatter matrix and location vector. These are commonly the sample mean and variance-covariance matrix, which are maximum likelihood estimates if the underlying distribution is normal. On the contrary the financial literature provides mounting evidence on the returns being long-tailed. It is also known that sample mean and covariance are sensitive to outliers. This paper discusses the construction of a portfolio selection model where the variance-covariance matrix and the expected returns are modeled robustly. Two robust alternatives are proposed. One assumes a heavy-tailed distribution of the returns, the other is obtained via high breakdown estimates of the location vector and scatter matrix. We show evidences in favour of the robust models. Empirical results compare the performances of the "classical" and robust methods and analyze the sensitivity of the estimates to changes in some parameters.