Local asymptotic normality for multivariate linear processes (Q1316604): Difference between revisions

Let \(\{x_ t\}\) be i.i.d. \(d\)-dimensional random vectors and \(\{\gamma_ j\}\) \(d \times d\) nonrandom matrices. The author considers a linear process \(y_ t=\sum^ \infty_{j=0} \gamma_ j x_{t-j}\). Local asymptotic normality (LAN) is established for the likelihood ratios for the distribution of \((x_ j:j \leq 0\), \(y_ 1,\dots,y_ n)\). From this general result LAN for \(\text{AR} (\infty)\) and LAN for \(\text{MA} (\infty)\) are easily derived. Formulas for average Fisher information matrices are presented including the case of finite-order multivariate ARMA models. The paper is a part of the author's PhD dissertation written under guidance of D. Brillinger.