Forecasting with measurement errors in dynamic models

Abstract

In this paper, we explore the consequences for forecasting of the following two facts: first, that over time statistics agencies revise and improve published data, so that observations on more recent events are those that are least well measured. Second, that economies are such that observations on the most recent events contain the largest signal about the future. We discuss a variety of forecasting problems in this environment, and present an application using a univariate model of the quarterly growth of UK private consumption expenditure.