Estimating time variation in measurement error from data revisions: an application to backcasting and forecasting in dynamic models
提出一种方法，利用统计机构的行为模型从数据修订的方差中估计测量误差及其随时间的变化，并基于英国总支出实时数据展示了该方法对预测性能的提升。
Abstract Over time, economic statistics are refined. This implies that data measuring recent economic events are typically less reliable than older data. Such time variation in measurement error affects optimal forecasts. Measurement error, and its time variation, are of course unobserved. Our contribution is to show how estimates of these can be recovered from the variance of revisions to data using a behavioural model of the statistics agency. We illustrate the gains in forecasting performance from exploiting these estimates using a real‐time dataset on UK aggregate expenditure data. Copyright © 2009 John Wiley & Sons, Ltd.