Nominal GDP Growth Targeting

In 1997, Ben Bernanke and Frederic Mishkin explained why they do not think that targeting nominal GDP growth is better than targeting inflation:

Is Inflation the Right Goal Variable for Monetary Policy? The consensus that monetary policy is neutral in the long run restricts the set of feasible long-run goal variables for monetary policy, but inflation is not the only possibility. Notably, a number of economists have proposed that central banks should target the growth rate of nominal GDP rather than inflation (Taylor (1985); Hall and Mankiw (1994)). Nominal GDP growth, which can be thought of as "velocity-corrected" money growth (that is, if velocity were constant, nominal GDP growth and money growth would be equal, by definition), has the advantage that it does put some weight on output as well as prices. Under a nominal GDP target, a decline in projected real output growth would automatically imply an increase in the central bank's inflation target, which would tend to be stabilizing.[14] Also, Cecchetti (1995) has presented simulations that suggest that policies directed to stabilizing nominal GDP growth may be more likely to produce good economic outcomes, given the difficulty of predicting and controlling inflation:

Nominal GDP targeting is a reasonable alternative to inflation targeting, and one that is generally consistent with the overall strategy for monetary policy discussed in this article. However, we have three reasons for mildly preferring inflation targets to nominal GDP targets. First, information on prices is more timely and frequently received than data on nominal GDP (and could be made even more so), a practical consideration which offsets some of the theoretical appeal of the nominal GDP target. Although 20 collection of data on nominal GDP could also be improved, measurement of nominal GDP involves data on current quantities as well as current prices and thus is probably intrinsically more difficult to accomplish in a timely fashion. Second, given the various escape clauses and provisions for short-run flexibility built into the inflation-targeting approach, we doubt that there is much practical difference in the degree to which inflation targeting and nominal GDP targeting would allow accommodation of short-run stabilization objectives. Finally, and perhaps most important, it seems likely that the concept of inflation is better understood by the public than the concept of nominal GDP, which could easily be confused with real GDP. If this is so, the objectives of communication and transparency would be better served by the use of an inflation target. As a matter of revealed preference, all central banks which have thus far adopted this general framework have chosen to target inflation rather than nominal GDP.

Here's the question I have, and a shot at the answer. If we link the growth of nominal GDP to the federal funds rate (as opposed to the money supply), then it shares many characteristics of a Taylor rule. That is, the growth rate of nominal GDP is equal to the growth rate of output plus the growth rate of prices, i.e. output growth plus inflation. Thus, an interest rate rule connected to the growth rate of nominal GDP would be of form i = f(output growth, inflation) which is very close to a standard Taylor rule formulation.

My intuition was that nominal GDP targeting based upon a rule of this type would exhibit indeterminacy. However, surprisingly to me, this is not the case. In 2003, Kaushik Mitra showed that a rule where the interest rate is adjusted so as to keep nominal GDP growth as close as possible to a constant value does not suffer from this problem. The rational expectations equilibrium is locally unique under nominal GDP growth targeting (essentially, the rule satisfies the Taylor principle of moving the nominal interest rate more than one-to-one with expected inflation). Furthermore, the equilibrium is stable under learning. This is important because, as Mitra notes:

[Howitt] explicitly warned that, in general, any RE analysis of monetary policy should be supplemented with an investigation of its stability under learning. He emphasized that the assumption of RE can be quite misleading in the context of a fixed monetary regime; if the regime is not conducive to learnability, then the consequences can be quite different from those predicted under RE

Note, however, an important exception to the encouraging results for nominal GDP growth targeting using interest rate rules. Some people define nominal GDP targeting as setting expected nominal GDP growth one period ahead equal to a fixed value. This version of nominal interest rate targeting does not satisfy the Taylor principle (i.e. indeterminacy is a problem) and it is not stable under learning. Thus, this shows that the form of the nominal GDP targeting rule matters, and some rules can perform very badly.

I have not said much about the Taylor rule versus nominal GDP growth targeting debate, in part because if GDP growth is linked to money, i.e. if we use a money rule rather than an interest rate rule, then, while that is fine theoretically, there are big problems with defining and measuring the appropriate monetary aggregate to target (such an aggregate may very well be dynamic as well as difficult to measure, i.e. it changes over time, further complicating this approach). As I've discussed here in the past, and others have discussed recently as well, the relationship between money and nominal GDP appears to break down in the early 1990s (you can see this in a graph of M2 velocity). Thus, a rule linked to money runs into the problem of how to define money, and it's not a problem we have solved. So there didn't seem to be much reason to think hard about these kinds of rules.

The other reason I have not embraced nominal GDP targeting is the one discussed above. If an interest rate rule is used instead of a money rule, it seemed like a version of a Taylor rule with different coefficients (and a different measure of output -- the Taylor rule uses percentage deviation from full employment, while under nominal GDP targeting the variable is output growth), so it wasn't clear that it could resolve the known problems with these kinds of rules. In fact, I thought it would make problems such as indeterminacy even worse. But that turns out not to be the case, nominal GDP targeting seems to do better in terms of determinacy and stability under learning. (The results are, of course, model dependent so it's as much a debate about the proper model of the macroeconomy as it is a debate about monetary policy rules. Some rules seem to be robust across a wide variety of models, and due to our uncertainty over the correct macro model to use, my evaluation of alternative policy rules gives large weight to the robustness feature).

My bias has been toward Taylor rules rather than nominal GDP targeting, mostly because I think of GDP targeting as linked to the money supply rather than the interest rate. But given results such as Mitra's on interest rate rules, I think we need to keep an open mind on the optimal form of the monetary policy rule.

So, for those of you who are advocates of nominal GDP targeting and have studied nominal GDP targeting in depth, (a) what important results concerning nominal GDP targeting have I left out or gotten wrong? (b) Why should I prefer one rule over the other? In particular, for proponents of nominal GDP targeting, what are the main arguments for this approach? Why is targeting nominal GDP better than a Taylor rule? I don't think I've seen a simple, bullet-point type summary of the pros and cons of nominal GDP targeting versus a Taylor rule (while you're at it, setting aside issues of measurement, under what conditions is nominal GDP a better nominal anchor than money?). Also, I didn't talk about targeting the level of nominal GDP at all, just the growth rate, but some of you advocate level targeting rather than growth targeting. This comes up with Taylor rules as well, and whether to target growth rates or levels, or both, depends upon how persistence is modeled and the type of rigidity that is imposed on the model. Is the same true for nominal GDP targeting? When is level targeting better than growth targeting, or vice-versa?  The main question here is the relative merits of nominal GDP targeting versus a Taylor rule, but how the results vary with level versus growth targeting (and other factors such as the presence of forward or backward looking elements in the rule) is also of interest.

[I'm very interested in this question, so I'll post (or link to) all reasonable responses that discuss of the merits of alternative policy rules.]

Update: Andy Harless in comments:

I like NGDP level (or "path") targeting largely because it separates the Fed from the politically contentious issue of what the inflation rate should be. If the inflation rate turns out to be too high or too low for someone's taste, the Fed can say, "Hey, we did our job. Complain to the real economy about that." Of course, it's no accident that I've come to this position recently, at a time when the Fed could really use some political cover for aggressive unconventional policies designed to stimulate recovery from a particularly deep recession. All this nonsense about how QE2 (and more importantly QE3 and QE4 -- since QE2 is not aggressive enough) is potentially inflationary: the answer should be, "Yeah, so what. It's not our job to set the inflation rate. If the real economy responds, then great. If inflation responds instead, then the real economy is behaving badly. Somebody else needs to fix that, maybe. Not us." I believe, however, that NGDP growth rate targeting is a very bad idea. While some shocks are indeed persistent, this persistence (a) is not always present and (b) often ought to be resisted when it is. A big increase in the unemployment rate may result in a permanently lower real output path as workers lose their skills, but monetary policy ought to lean against this tendency, even at the expense of temporarily raising the inflation rate. With NGDP growth rate targeting, you basically try to avoid recovering from a recession unless the inflation rate goes down. That's clearly bad policy. Right now we have some people arguing that QE2 was unnecessary because NGDP already appears to be rising at a normal rate. That's clearly a silly way to think. NGDP needs to rise at a higher than normal rate now, or the unemployment rate will remain unnecessarily high. I don't see how anyone can advocate NGDP growth rate targeting after looking at the situation we're in today.

...To clarify..., my preference over 4 possible policy regimes would be: best: NGDP level targeting second best: Price level targeting second worst: Inflation rate targeting worst: NGDP growth rate targeting Let me say, also, that NGDP level targeting has an advantage over a Taylor rule in that it never breaks. When the Taylor rule implies an interest rate below zero, the appropriate course of action is unclear, and there is a lot of room for dispute. With NGDP targeting, there is less room for dispute: just look at your forecast; if it says NGDP will be below target, then you need a more aggressive policy.

Update: Scott Sumner's reply, and my response.