M-estimation for autoregression with infinite variance (Q1185791): Difference between revisions

Autoregressive processes with innovations which are in the domain of a law with tails of the type \(| x|^{-\alpha} L(| x|))\) where \(L(x)\) is a slowly varying function for the large values of variables, is considered and the performance of the \(M\)-estimators of the coefficients of the model is investigated. First, the asymptotic distribution of the estimators is derived for models without intercept and with known variance, later the result is generalized to allow for the intercept and an estimate of the scale. Secondly, LAD estimators are studied separately because the earlier results assumed the existence of a Lipschitz continuous derivative of the critical function \(\rho\). Moment conditions on the innovations implying the consistency of \(M\)-estimators (determined by a criterial function of the type \(\rho(x) = | x|^ \gamma\), \(\gamma > 0\)) are also presented.