This study examines the relationship between specifications for long-run output patterns and specifications for business cycle dynamics. In an application to US GDP, it is found that inferences about the nature of the trend in output are not robust to changes in the specification for short-run fluctuations. Similarly, the choice of which model best describes the transitory movements in output depends on the way in which the trend is specified. The empirical analysis makes use of Bayesian methods to compare time series models for US GDP. Inspection of the predictive densities for the individual data points suggests that the information contained in the data is largely limited to the observations associated with business cycle turning points.