Abstract
The article considers a number of regression models with unknown coefficients. We describe some of these models and prove general propositions concerning their asymptotic behavior. The parametric set of unknown parameters is always assumed closed and in general unbounded. The case of an open set is simpler in our view, and at least the asymptotic distribution of the estimates in this case is generally normal under natural constraints, which is not always so with a compact constraint set. In what follows we consider criteria and estimates of a general form that hmlude many hitherto ignored cases of regression analysis. These are primarily M-estimates in P. Huber's terminology, which constitute the most natural class of robust regression estimates. We consider various classes of M-estimates for parametric and nonparametric regression models, which allow observation errors that are dependent at different time instants. We always assume discrete time; the continuous time case is not considered here, although many of our propositions are also valid for continuous time.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have