Bernanke believes QE works, but having been caught off-guard once before, in 2007-2008, he doesn’t fully trust his own judgment. He fears some risk of bubbles, or excess private disintermediation, in either case resulting from low interest rates. He lets Tarullo and Stein carry related messages to the markets to signal possible fears without having to endorse them.
Let’s say he assigns these risky scenarios a fairly small p = 0.05. Still, another financial collapse would be a disaster, all the more for political economy reasons. Bernanke has spent down his own political capital and these days Republicans are more likely to be obstructionists. Fear of that disaster leads him to withdraw QE sooner than his “most likely to be true” opinion thinks prudent. Economists respond by defending the “most likely to be true” opinion, and by arguing moralistically rather than probabilistically. That doesn’t convince Bernanke, because said economists can only convince him that they are likely right, not that he should obliterate his p = 0.05 fear of being wrong. The current policy course continues, early withdrawal from QE is contemplated, and economists complain all the more. Outside observers find it hard to understand the disconnect.