Soumendra Lahiri - Higher order asymptotic properties of the Bootstrap in post model selection inference
I note on slide 11, true parameter depends on n. How necessary is this added complexity?
In referece to the McKeague and Qian JASA discussion paper from 2015, `An adaptive resampling test for detecting the presence of significant predictors'
I've posted it here (with the discussion and rejoinder):
Several of this workshop's participants gave comments in the discussion (Richard Samworth, Soumendra Lahiri, Larry Brown).
It was pointed out, in particular by Belloni and Chernozhukov, that the conditions needed for pointwise model or variable selection consistency and other optimality properties of the selection procedure are causing problems for the construction of inference procedures that are uniformly valid in a region of the parameter space.
In your talk, you went from procedures having strong oracle (plus VSC), to oracle (plus VSC) to just VSC. In light of the discussion mentioned above, do you see any role of `weakening' of the oracle requirements, or weakening the selection optimality requirements, in achieving higher-order accuracy?