I recently collaborated with Steve Ellner (Cornell) and got some analytical results that show when -- what forecast lengths and quasi-extinction thresholds --
extinction estimates are unavoidably bad versus most definitely good. This hopefully, although maybe that is naive, will help clarify that it isn't that extinction is or isn't predictable. It is that certain types of predictions are unavoidably bad and you should avoid that type of prediction.
Accepted in Ecology Letters: Ellner, S. P. and E. E. Holmes. 2008.
Commentary on Holmes et al. (2007): resolving the debate about when extinction is predictable.