Steve writes wide-rangingly, taking issue with several aspects of empirical macroeconomics.
Steve goes for the modern VAR literature, which seeks to identify monetary policy shocks, and measures that these shocks have effects on real variables, and that their effects on all variables take time to their full effects.
He rightly points out that we might be concerned how VAR researchers identify those shocks. But, can we dismiss all of them?
The methods I know about comprise; short-run timing restrictions embedded in the VAR [eg Christiano-Eichembaum-Evans]; long-run restrictions [Blanchard-Quah]; sign restricitions [Uhlig and many others]; narrative methods of identifying monetary policy shocks [Romer and Romer]; monetary policy surprise measures [Rudebusch; heavily criticised by Sims himself, but still] constructed from the gap between rate expectations and outturns; external instruments [Stock and Watson, Mertens and Ravn…].
With the exception of the Uhlig work on sign restrictions, [contradicted by other, later work in a similar vein] I thought it was the case that this work found non-neutrality of monetary policy shocks. And that there was a reasonable consensus about the lags.
We can quibble with probably all of the papers in this large literature individually, but I find the overall conclusions persuasive.
Moreover, all of the work tries to use theory to some extent to identify those shocks, and that ought to reassure someone like Steve. For example, the early Blanchard-Quah work uses the insight that satisfactory models have the property that ‘nominal’ shocks should be neutral on real output in the long run. Narrative measures are explicitly motivated by the theoretical distinction between expected, systematic changes in policy, and the rest. Even timing restrictions are theoretically motivated – though perhaps most contentious. And sign restrictions are restrictions that come from theoretical models. Rarely are data allowed simply to ‘speak for themselves’.
Steve also seems to question whether the shocks that VAR researchers crave are interesting: the proper concern for us being the consequences of a particular systematic rule for monetary policy.
In reply to this: the unsystematic shocks, which if central banks had been doing their jobs properly in history would simply not be there, since they have no reason to be there [caveat: save for reasons of experimentation], are the unfortunate accidents that allow us to identify the parameters of the underlying structural model [including those in and out of the policy rules in place in the past]. Once this is done, we can then evaluate what the effects of alternative systematic policies would be, and try to work out what the best one to adopt is.
Another part of Steve’s pessimism about empirical macro connects the irrelevance of ‘long and variable lags’ in a world where there is open and communicated forward-planning by policymakers, and private agents that may form expectations about future central bank actions.
I don’t see a fundamental problem here.
For a start, provided we accept that history is nonetheless full of monetary policy shocks, [with randomness in early vintages of data, why would it not?] we can characterise all this forward-lookingness in a model [for the sake of argument, say the NK model], and see what a shock does in that model, and compare it to the counterpart in the data [the VAR].
Second, if we think think that there are unsystematic announcements pertaining to future events – what the literature terms ‘news shocks’, we can try to identify those too. And several have.