A Linear Weight Estimator for Dynamic Global Minimum Variance Portfolio Allocation
40 Pages Posted: 11 Mar 2022 Last revised: 30 Nov 2023
Date Written: October 25, 2023
Abstract
This paper introduces a linear weight estimation (LWE) framework as a novel semi-parametric method for dynamic global minimum variance portfolio (GVMP) allocation. The LWE method assumes a dynamic linear model for the \textit{ex ante} optimal GMVP weights. Based on a time series of daily realized covariance estimates, the LWE model parameters can easily be estimated in closed form using the method of moments. Importantly, we prove that the estimated LWE portfolio weights directly and uniquely minimize a finite sample estimate of the unconditional portfolio variance, which is not achieved by most of the existing methods in the literature. Empirical results demonstrate that LWE outperforms competing estimators in terms of out-of-sample portfolio variance measures. LWE can also be extended to incorporate controls for investment constraints, which further balances its economic performance and transaction costs.
Keywords: portfolio allocation, high-frequency finance, realized measures, forecasting
JEL Classification: C51, C58, G11, C32
Suggested Citation: Suggested Citation