Limiting spectral distribution of sample autocovariance matrices (Q396002)

The authors study the limiting empirical spectral distribution of the sample autocovariance matrix of a linear process with reasonable restriction on the coefficients. They prove that the limiting spectral distribution (as the size of the matrix increases) exists and the limit does not depend on the distribution of the underlying driving i.i.d.\ sequence. The support of the limiting spectral distribution is unbounded and the limit does not coincide with the spectral distribution of the theoretical autocovariance matrix. However, the authors show that it does so if one considers a suitably tapered version of the sample autocovariance matrix. For banded sample autocovariance matrices, the limit has unbounded support as long as the number of non-zero diagonals in proportion to the dimension of the matrix is bounded away from zero. If this ratio tends to zero, then the limit exists and again coincides with the spectral distribution of the theoretical autocovariance matrix. Finally, the authors also study the empirical spectral distribution of a naturally modified version of the autocovariance matrix which is not non-negative definite.