Local quadratic spectral and covariance matrix estimation

Abstract

The problem of estimating the spectral density matrix $f\left(w\right)$ of a multi-variate time series is revisited with special focus on the frequencies $w=0$ and $w=\pi$. Recognizing that the entries of the spectral density matrix at these two boundary points are real-valued, we propose a new estimator constructed from a local polynomial regression of the real portion of the multi-variate periodogram. The case $w=0$ is of particular importance, since $f\left(0\right)$ is associated with the large-sample covariance matrix of the sample mean; hence, estimating $f\left(0\right)$ is crucial to conduct any sort of statistical inference on the mean. We explore the properties of the local polynomial estimator through theory and simulations, and discuss an application to inflation and unemployment.