Vasyl Golosnoy, Ruhr Universitat Bochum

will give a presentation on

A Simple Powerful Test for Global Minimum Variance Portfolio Weights

Abstract:

Realized global minimum variance portfolio (GMVP) weights are computed from inverted realized covariance matrices. Conventional statistical tests for GMVP weights are based on the vector difference between realized GMVP weights and those under the null hypothesis. However, inversion of realized covariance matrices for computing realized GMVP weights amplifies estimation errors which deteriorates both size and power of such tests, especially when the number of intraday returns is not much larger than the number of assets in the GMVP. To resolve this problem, we propose a novel approximation vector to replace the vector difference. Under the null hypothesis this approximation vector has the same limit distribution as the vector difference, but its computation is simple as it does not require matrix inversion. We analytically derive stochastic properties of the approximation vector and suggest the corresponding sta- tistical test. In Monte Carlo simulations we analyze the size and power of our novel test and illustrate its advantages. In the empirical application we show the usefulness of the new test on GMVP weights for practical purposes.