This article fi?rst reviews the issues involved in specifying the trend component in an unobserved components (UC) model. The conditions under which the maximum likelihood (ML) estimators in a misspecifi?ed model converge to fi?xed values are obtained and these values are computed for a range of models, with attention focussed on the case where the true trend is a random walk plus drift but a model with a stochastic slope is fi?tted. The robustness of various speci?cations is assessed by comparing MSEs of ?filtered and smoothed estimators of the trend using the algorithm developed by Harvey and Delle Monache (JEDC, 2009). Cycles are then brought into the analysis and efficiency and robustness with respect to smoothing and fi?ltering are assessed. The article concludes by examining various measures of the output gap and the experience of recent years.

VIEW PAPER