Recently the new president of the Minneapolis Fed, Narayana Kotcherlakota, also at the U. of Minnesota, and a longtime advocate of DSGE modeling, "apologized" for the poor performance of "modern" macro models in predicting/explaining the crash and recession of 2008, http://www.minneapolisfed.org/publications_papers/pub_display.cfm?id=4428. He said that the models need much more complicated modeling of financial markets and dynamics, duh.
This led to James Morley of Washington University and Macroeconomic Advisers, who last year was forecasting a V-bounce recovery from the recession (which did not work, unfortunately), to write a critique of this paper, with which I largely agree, http://macroadvisers.blogspot.com/2020/06/emperor-has-no-clothes-ma-on-state-of.html. He points out that it is silly of Kotcherlakota to bemoan policymakers having fallen back on "long-discarded models" when considering such things as the fiscal stimulus, asking "Who is this 'we' who discarded those models?" He accurately points out that the best of these models, such as the one by Smet and Wouters, continue to rely on AR modeling of supposed exogenous shocks that have nothing to do with policy or anything else. In the end, he says that improved models should incorporate the ideas of Hyman Minsky, with which I strongly agree.
Others have commented on this, including Brad DeLong and also the estimable Mark Thoma at economists view, where there are extensive comments after his extensive excepts from Morley's paper, http://economistsview.typepad.com/economistsview/2020/06/the-state-of-modern-macro.html#comments.