Over the last year I have been working on a new working paper, which considers revisions in output gap estimates derived from statistical trend-cycle decompositions, just as Orphanides and van Norden (2002, ReStat) did. However, I consider models with stochastic volatility, that can adapt to the changing patterns in aggregate volatility like before (and after) the Great Moderation period. The model with time-varying volatility generate credible sets for the output gap that are tighter than in the constant-paramter case. Also, when comparing realtime estimates against "final" estimates (derived from the latest available data vintage) revisions are quite a bit smaller. 

Slides for a talk can be found here (or here)

A draft of the paper is here.