Methods for decomposing aggregate time series into trend and cycle components are frequently used in macroeconomics. When choosing which method to employ for this purpose, revision properties are often considered. In this article, the trend/cycle decomposition methods suggested by Hodrick and Prescott (1997) and by Hamilton (2018) are compared with respect to their revision properties at different time-series sample positions. Considering revisions for different sample positions individually, without aggregating into a summary statistic for the entire sample, can nuance results. It is shown that the filters have distinctly differing real-time revision properties that vary considerably across sample positions. While the investigated revision properties are worse for the HP filter during an initial period when more data become available, the Hamilton filter’s properties degrade slowly over time and tend to eventually become worse than those of the HP filter when more observations have become available. These nuances are not possible to trace out when investigating revision properties using methods that do not account for properties at specific sample positions. The results imply that which decomposition method is preferable in terms of real-time uncertainty depends on what profile of instability is deemed least problematic for the empirical situation under consideration. As a consequence, choosing between the two filters becomes an important trade-off with respect to the filters’ revision properties in the light of the specific task at hand and the preferences of the practitioner.
