A simulation study on vector arma processes with nonstationary innovation:a new approach to identification

Abstract

This paper considers the problem of identification of vector autoregressive-moving average (VARMA) processes with nonstationary innovations and suggests a new approach to the problem. This approach has successfully been applied recently by Singh and Tampubolon (1995) to the recordings of the Kobe earthquake which occurred on 16 January, 1995, in the univariate case. The same techniques have been extended in this paper for the identification of vector ARMA processes with nonstationary white-noise in the sense that its covariance structure is assumed to be time-dependent; a few simulated examples are discussed for illustration. Furthermore, they have been applied to the identification of the vector ARMA models that were fitted to the Kobe earthquake vibrations recorded at two stations, viz. (i) Charters Towers, Queensland and (ii) Hobart, Tasmania, both in Australia. The theoretical results have been supported by simulation studies.