Efficient inference in multivariate fractionally integrated time series models

From MaRDI portal





The author proposes maximum likelihood estimators for multivariate fractionally integrated time series models and develops a number of tests based on these estimators. The results generalise the previous work on univariate models. One of the main features of the author's approach is that it leads to inference based on normal and \(\chi^2\)-distributions which is in striking contrast with the related work on unit roots and nonstationary models. More specifically, maximum likelihood estimators based on the Gaussian likelihood are developed. These are used to derive Lagrange multipliers, likelihood ratio (LR), and Wald test statistics under local alternatives. The cases of fractional integration of the same order and different orders are considered separately. The error process may be white noise or a multivariate autoregression. The distributional results do not require normality but when that assumption is made the tests become asymptotically efficient. The results may be used to test jointly various hypotheses about the integration orders of the components of the vector process and to construct joint confidence regions for them. Especially valuable is the possibility to test for common integration order without specifying a particular value for it. Extensive simulation evaluation of the finite sample performance of the LR statistic is presented.





Describes a project that uses

Uses Software






This page was built for publication: Efficient inference in multivariate fractionally integrated time series models

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3156187)