Performance of the global minimum variance portfolio : a study on correlation and volatility estimation.

Abstract

This paper studies the returns of efficient portfolios based on different estimations of the covariance matrix. More specifically we derive the Global Minimum Variance Portfolio (GMVP) weights, using different estimation methodologies to derive

the covariance matrix. The out-of-sample performance of the GMVP was analyzed under varying assumptions in order to identify if there exists a set of methodology that

consistently outperform the others.

The results show that the use of GARCH(1,1) estimates for volatility would usually improve portfolio returns and Sharpe Ratio. This effect is observed regardless of the correlation estimation methodology used.

From the findings, the preferred methodology for estimating correlation estimates are Constant and Weight Positive Correlations as they tend to exhibit high Sharpe Ratios

with relatively low portfolio turnover compared to historical estimates of correlation.