Asset Allocation by Variance Sensitivity Analysis

This article provides a solution to the curse of dimensionality associated to multivariate generalized autoregressive conditionally heteroskedastic (GARCH) estimation. We work with univariate portfolio GARCH models and show how the multivariate dimension of the portfolio allocation problem may be recovered from the univariate approach. The main tool we use is "variance sensitivity analysis," the change in the portfolio variance induced by an infinitesimal change in the portfolio allocation. We suggest a computationally feasible method to find minimum variance portfolios and estimate full variance-covariance matrices. An application to real data portfolios implements our methodology and compares its performance against that of selected popular alternatives. Copyright 2004, Oxford University Press.