Title: Asymptotic theory of M-estimators for linear regression in time series
Authors: Bent Nielsen - University of Oxford (United Kingdom) [presenting]
Soren Johansen - University of Copenhagen (Denmark)
Abstract: An asymptotic theory is provided for a class of regression M-estimators. The objective function must be continuous, but it can be non-convex and non-differentiable. The regression equation has innovations that can have a continuous or discrete distribution. The regressors must satisfy an assumption on the frequency of small regressors. This is met by a variety of deterministic or stochastic regressors, including stationary and random walk regressors. In a previous paper we have shown that condition along with some mild conditions on the criterion function is sufficient to ensure boundedness, or tightness, of their non-standardized distributions.