This paper introduces a method for constructing impulse response functions that utilise historical datasets, where the confidence intervals are not explicitly based upon a measure of statistical uncertainty, but on differences in the past aggregate behaviour of economic agents.

This allows for a convenient interpretation of robust results and a great deal of flexibility in the way in which the method may be applied to a large suite of potential models. It also allows for several extensions, which include an intuitive framework for the construction of ensemble impulse response functions.

An illustration of this method is applied to a new dataset that includes more than ten years of quarterly data vintages that contain real-time measures of a large number of economic variables. A small-open-economy structural vector autoregressive model and a large structural dynamic factor model illustrate some of the advantages that are presented by this methodology.