Tony Yates had recently written a couple of posts (here, and here, but see also the discussion with Andy Harless on the second) slamming the idea of NGDP targets. (From now on I assume this refers to targeting the level of NGDP.) Now you might think that NGDP targets do not need any support from lukewarm advocates like me, given all the supporters in the econ blogging world. That would be wrong, because - as Tony rightly says - most advocates of NGDP targets tend to argue in a model free way. Both he and I want to stay close to the academic literature, at least as a starting point.
I think Tony is wrong when he says that “the case for levels based targets – including NGDP levels targets – is, both practically and analytically, extremely weak”. In making such a claim, Tony should be very worried that one of the supporters of NGDP targets is Michael Woodford, who literally wrote the book on modern monetary theory.
He rightly focuses on the big plus for any levels based target, which is that it can mimic the optimal but time inconsistent policy. After a negative shock to inflation, the central bank commits to future above target inflation which reduces the impact of the negative shock. This is useful in any situation, but particularly at the Zero Lower Bound. He has two arguments against this: one analytical and one practical.
The analytical argument is that in more complex models discretionary policy can yield multiple equilibria. I do not see why that is relevant. We are comparing policy that follows a rule (NGDP targets) with an optimal policy that assumes commitment.
His practical argument against NGDP is that it would be difficult to communicate. I think that kind of follows from its time inconsistent nature. So is Tony saying that the central bank can never follow a time inconsistent optimal policy? The whole point (in my view) of NGDP targets is that they try and directly address the communication problem. The welfare gains from following the optimal time inconsistent policy are large (see my discussion of Iván Werning’s paper for example), so to wave those away as ‘difficult to communicate’ seems - if I may say so - terribly old school central banking.
In a similar manner, Tony argues - again correctly - that these commitment gains depend on forward looking expectations. Under adaptive expectations, levels targets are costly. However central banks, alongside the academic literature, do routinely assume rational expectations in most of the stuff they do. Furthermore, a key argument for inflation targets is that agents are forward looking, and it is an argument that central banks frequently invoke. We need some consistency here.
One final argument Tony uses that really should not be made is that simple models show that inflation variability is about twenty times more important than output variability in assessing welfare, and therefore NGDP targets give too great a weight to output. (The link is a bit more complicated than that - I tried to sketch the maths here.) While these models may be a useful approximation to how the economy works (which is what they were designed to do), they are not a useful approximation to weighing the welfare costs of inflation variability against output variability (which they were never designed for). We know that from survey data and empirical work that looks at the cost of unemployment: I argue this case in detail here. These considerations suggest to me that targeting NGDP will give too little weight to output relative to inflation following shocks.
Having said all this, it is great that Tony is opening up the discussion on the correct level, so we can get away from what often seems like faith based arguments for NGDP targets. I think the framework that he seems to have in mind is also the correct one: the ultimate policy target would be inflation (and the output gap: I would want a dual mandate), and NGDP would be an intermediate target to achieving welfare maximising paths. So I hope this discussion continues. My one last plea is that arguments make clear whether a NGDP targeting regime is being compared to some form of optimal policy, or policy as currently practiced: as I suggest here these are (unfortunately) different things.