Title: Multiple forecast comparisons in unstable environments and high dimensions
Authors: Ekaterina Smetanina - University of Chicago (United States) [presenting]
Abstract: A methodology is developed for the forecast evaluation of various models in potentially unstable environments. We first propose a new measure of forecast performance in unstable environments and then develop a methodology for selecting the best forecasting model from a pool of models while allowing for general types in potential instabilities (e.g. best forecasting model that changes over time). At a given point in time, the methodology allows practitioners to construct a set of best models - models that are indistinguishable from each other given our new metric.