Nonstandard estimation methods in one-dimensional and spatial first-order autoregression models

Abstract

The article proposes alternative parameter estimators for autoregression that differ from the least squares estimates. In unstable (critical) cases, where the characteristic equation's roots lie on the unit circle, least squares estimators generally exhibit a complex asymptotic distribution. In contrast, the proposed nonstandard estimators tend to have a simpler asymptotic distribution in most critical cases.