ORDER DETERMINATION OF MULTIVARIATE AUTOREGRESSIVE TIME SERIES WITH UNIT ROOTS (Q3219619)
From MaRDI portal
!
This is the item page for this Wikibase entity, intended for internal use and editing purposes. Please use the normal view instead:
scientific article; zbMATH DE number 3886924
| Language | Label | Description | Also known as |
|---|---|---|---|
| default for all languages | No label defined |
||
| English | ORDER DETERMINATION OF MULTIVARIATE AUTOREGRESSIVE TIME SERIES WITH UNIT ROOTS |
scientific article; zbMATH DE number 3886924 |
Statements
ORDER DETERMINATION OF MULTIVARIATE AUTOREGRESSIVE TIME SERIES WITH UNIT ROOTS (English)
0 references
1984
0 references
autoregressive time series
0 references
nonstationarity
0 references
estimation
0 references
order determination
0 references
weak consistency
0 references
finite fourth-order moments
0 references
least squares
0 references
0 references
0.9149417
0 references
0.9054574
0 references
0.8886306
0 references
0.8868351
0 references
The d dimensional autoregressive (AR) process of order p is defined by \(X(t)=\sum^{p}_{i=1}A_ iX(t-i)+c+e(t),\) where the e(t) are independent (0,G), det \(G\neq 0\), and have finite fourth-order moments. The \(A_ i\) are such that the associated characteristic equation has some roots equal to 1. If the \(A_ i\) are fitted by least squares to n observations, let \(\hat G_ k\) denote the mean residual sum of squares matrix, when a kth order autoregression is fitted, \(k=0,1,...,L\). The order p is estimated by that k which minimizes log det \(\hat G_ k+kf(n)/n.\) It is proved that such a k is a weakly consistent estimate of p if and only if \(L\geq p\) and f(n) is an increasing function so that f(n)\(\to \infty\) and f(n)/n\(\to 0\) and \(n\to \infty\). This generalizes results by the same author (unpublished dissertation) for the case of all roots less than 1 in absolute value. In a later section of the paper the case of a scalar AR of order 2 is studied in detail, and it is found that the existence of roots has a favourable impact on the probability of underestimating the order. For this same model simulations are presented in a final section.
0 references