Title: Identification robust predictive ability testing
Authors: Andrea Naghi - Erasmus University (Netherlands) [presenting]
Abstract: The aim is to analyze the predictive accuracy evaluation of models that are strongly identified in some part of the parameter space but non-identified or weakly identified in another part of the parameter space. We show that when comparing the predictive ability of models that might be affected by identification deficiencies, when the parameter estimation error is negligible, the null distribution of out-of-sample predictive ability tests is not well approximated by the standard normal distribution. As a result, employing a standard (strong) identification critical value can lead to misleading inference. We propose methods to make the out-of-sample predictive ability tests robust to identification loss. These methods use a different critical value than the standard one and include: a least-favorable critical value and a data dependent critical value. In settings where the parameter error is non-negligible, it is shown that the asymptotic distribution of the usual predictive ability test is standard, even when one allows for the model(s)to be only semi-strongly identified.