| Abstrakt: |
In this paper we provide exact, explicit, formulae for the mean, variance and covariance of the sample autocovariances for general ARMA (pq) processes. We present approximate formulae for the mean and variance of the sample autocorrelations and compare them with simulated results. The bias of the sample autocorrelations for a wide range of ARMA (1, l) processes and for various lags is computed. The relationship between the structure of a time series model and the behaviour of the bias is discussed. Tables of the bias of the sample autocorrelations are produced and finally attention is drawn to the fact that the sample autocorrelations are seriously biased in a good many cases. |
