Sequential fixed accuracy estimation for nonstationary autoregressive processes (Q2304245)

The paper deals with two problems of sequential parameters estimation in autoregressive processes. The first one is related to the \(\text{AR}(p)\) model with Gaussian disturbances. A new construction of sequential estimates on the basis of the least square (LS) estimates is proposed. An important property of these estimates is that they have a non-asymptotic normal joint distribution for any values of unknown parameters \((\theta_1,\ldots,\theta_p)^{'}\in\Re^p\) [\textit{T. L. Lai} and \textit{C. Z. Wei}, in: Multivariate analysis, Proc. 6th Int. Symp., Pittsburgh/Pa. 1983, Multivariate Anal. 6, 375--393 (1985; Zbl 0607.62120); \textit{T. L. Lai} and \textit{C. Z. Wei}, J. Multivariate Anal. 13, 1--23 (1983; Zbl 0509.62081); \textit{T. L. Lai} and \textit{D. Siegmund}, Ann. Stat. 11, 478--485 (1983; Zbl 0519.62076)]. The second issue is related to the unstable \(\text{AR}(p)\) process when the noise distribution is unspecified. Under general condition on the admissible parameter set, the uniform asymptotic normality of the proposed estimates is proved. There are adequate interesting examples related to the considered problem.