Limited access Past event
Brown Bag Seminar with Nicky Grant, University of St Andrews Efficient GMM estimation with (almost) singular variance
Abstract: Asymptotic properties of the 2-step efficient GMM estimator are studied when moment variance is (asymptotically) singular at the true parameter. The non-normal asymptotic distribution is derived, where super consistency is established under a certain condition. The intuition for this new result is discussed, along with simulation experiments with (almost) singular moments, including a linearised DSGE model with stochastic singularity. The paper argues against common approaches to deal with (almost) singular moments, such as regularisation, showing it lead to a slower rate of convergence and inefficiency in some instances.