Traditional mean variance optimization assumes that future returns and covariances of all the assets in the universe are known exactly. In practice, these input parameters are subject to estimation errors that may render the output of the optimization algorithm essentially useless. Here we...