Robustness of Portfolio Selection Models---A Comparative Study

Markowitz』 mean-variance portfolio selection model has brought about many studies regarding portfolio selection. Many portfolio selection models have been proposed and investigated, which depend on historical data to estimate expected returns and risks for assets. Sampling method, sample size, and outliers mostly affect the accuracy of statistical estimation. Since all portfolio selection models depend on the estimation of asset returns and risks, they are surely affected by these factors. Sampling horizon is a major influence to the sampling of asset return. Meanwhile, research has showed that sampling horizon and sampling size have affection on portfolio risk estimation. However, no comparative study has ever investigated the affection of sampling horizon to the effectiveness of portfolio selection models. Studies have also found that most asset returns exhibit positively skewed or fat-tailed distributions. These low rates of return might be caused by occasional disadvantage events, which can then be reasonably regarded as outliers. However, these outliers are very likely to affect the estimation of asset expected returns and risks. It is almost sure that downside outliers are leading to overestimated risks, causing the investor to lose part of the return that originally can be earned. Different models that adopt different risk measures have different sensitivity to outliers, leading to the necessity of investigating the sensitivity of portfolio selection models to outliers Portfolios must be hold for a certain period before the dispersion of individual risks can be achieved. Therefore, portfolio selection should not be influenced by short-term events, and a robust portfolio selection model thus must be insensitive to outliers and sampling errors, so as to improve the quality of portfolio decision-making. To compare the robustness of portfolio selection models, this study plans to investigate the sensitivity of portfolio selection models to the distribution patterns of asset return, outliers, and sampling horizon.

Project ID:PF9706-1119
External Project ID:NSC96-2416-H182-009-MY2