On the Empirical Spectral Distribution of Lag-Covariance Matrix in Singular Spectrum Analysis

Abstract

Singular Spectrum Analysis (SSA) is a non-parametric and rapidly developing method of time series analysis. Recently, this technique receives much attention in a variety of fields. In SSA, a special matrix that is called lag-covariance matrix plays a pivotal role in analyzing stationary time series. The objective of this paper is to examine whether the Empirical Spectral Distribution (ESD) of lag-covariance matrix converges to Marˇcenko{Pastur distribution or not. Such limiting distribution can help us to provide more reliable statistical inference when encountering with high-dimensional data. Moreover, a simulation study is performed and some tools of Random Matrix Theory (RMT) are used