Tuesday, June 19, 2018

What Does "Serious" Mean?

A week ago I wrote this in a tweet:
I'm puzzled by the infatuation with NGDP targeting. We have good reasons to care about the path for the price level and the path for real GDP. Idea seems to be that if you smooth Py that you get optimal paths for P and y. That's hardly obvious, and doesn't fall out of any serious theory I'm aware of.
The proximate cause of that outburst was this blog post by David Beckworth who, as you might know, is a proponent of nominal GDP (NGDP) targeting.

I've written before about NGDP targeting - for example I found this post from six years ago. It turns out that not much has changed since then. My views haven't changed, and the promoters of NGDP targeting are more or less the same ones that were around six years ago - and I don't get the sense they've come up with anything new to tell us. In any case, though, I wanted to find out for sure, and that's what my tweet was about. It basically says: "I don't think you have a good argument, but in case you do, please tell me about it." I tried to provoke Beckworth a bit to get an answer out of him, but he just called me a troll.

But, at last, some degree of success, as George Selgin has gone to considerable trouble to research the economics literature to come up with what he thinks are "serious" arguments that would support NGDP targeting. The first order of business for George is to complain about my use of the word "serious:"
I'm not exactly sure what Stephen means by a "serious theory." But if he means coherent and thoughtful theoretical arguments by well-respected (and presumably "serious") economists, then there are all sorts of "serious theories" out there to which he might refer, with roots tracing back to the heyday of classical economics. Indeed, until the advent of the Keynesian revolution, a stable nominal GDP ideal, or something close to it, was at least as popular among highly-regarded economists as that of a stable output price level, its chief rival.
"Serious" means that a good case has been made, where "good" means a theory with conclusions that follow from reasonable assumptions, and some notion that this works empirically. Good science basically. Unfortunately, in economics there is plenty of bad science to wade through, and some of that bad science finds its way into reputable journals, and is written by economists with fancy titles working at reputable academic and non-academic institutions. That's why I qualified my statement with "serious." I know there's a mountain of economics writing arguing that NGDP targeting is the cat's meow. But anyone could find mountains of writing in economics about any number of crappy ideas. I want to know about the good stuff. And I'm not going to be convinced by the reputations of the people promoting the ideas or the popularity of the ideas. Everyone has to make their case, and we don't do science by opinion poll.

So, what's an example of a serious idea, related to monetary policy rules? Well, one of these is Friedman's constant money growth rule. Friedman had a theory to back that up in "The Role of Monetary Policy." That's a coherent theory, but it's all in words. Like many things that Friedman did, you can also formalize it, which is - more or less - what Lucas did. There was also a formidable empirical project, Friedman and Schwartz's Monetary History of the United States, which Friedman used to support the theory. With everything laid out like that, we're able to argue about the assumptions, the conclusions, and the evidence. Serious stuff. Friedman was convincing enough that central banks actually implemented his ideas. Some of that was a success (the Volcker disinflation) and some was not (ongoing inflation control).

We're now in a world in which most "serious" central banks are inflation-targeters. Typically, they're targeting inflation at 2%, and manipulating a short-run target for an overnight nominal interest rate to hit the inflation target. Some central banks - the Bank of Canada in particular - have have been successful inflation targeters. The Bank of Canada has been doing this since 1991, and they haven't deviated much from a 2% inflation path over the last 27 years. This chart shows just the last 20 years, as I was comparing Canada to Japan in a talk I gave:
In the inflation targeting game, that's pretty good. In Canada's case, inflation targeting has proved to be feasible, it's easily understood, it's conducted in a way that promotes central bank independence, and it appears to be compatible with good economic performance in other dimensions.

So what's to complain about? I think we can make a case that central banking should be conservative - this isn't a venue where we should treat experimentation lightly. To change procedure we have to think that something's going wrong with the current approach, and that switching gears is highly likely to give us an improvement. In Canada, the Bank reviews its agreement with the government of Canada every 5 years, and thinks about whether it's doing the right thing. The agreement won't be reconsidered until 2021, but the Bank is already mulling this over - a process which involves economic research and consultation with the public. Serious work, I think.

Is the work that George references in his blog post serious? George thinks so, but of course George's and my views of what is serious might differ. I may not be as ignorant of the work as George thinks - I've already seen much of it - so you might guess that I'm putting some or all of it in the non-serious bin. To take some examples:

1. McCallum 1987: McCallum was an early promoter of NGDP targeting. From the paper, this looks like some kind of modified monetarism. McCallum won't write a model down. He says we don't have agreement on what the model is, so apparently his policy rule is somehow agnostic. But he goes ahead anyway to run an "atheoretical" regression, and uses the estimated coefficients to simulate the performance of his rule. Yikes. Definitely not serious - a multitude of sins in that one.

2. Garin, Lester, Sims: This starts with a boilerplate New Keynesian (NK) model and ultimately expands this into a "medium scale" model. This is complicated enough that's it's not going to be amenable to analysis - pencil and paper techniques. The authors estimate and calibrate the parameters in the model and then evaluate the performance of alternative policy rules, according to the welfare loss relative to what would occur in an identical economy with flexible wages and flexible prices. So, this is serious work. There's a model that is formally laid out, there are references to the related literature, there is an attempt to measure the welfare effects of alternative policy rules given the model at hand. The authors report, in Table 7, that the welfare loss for NGDP targeting is .11% of GDP, from inflation targeting .86%, from output gap targeting .03%, and from a Taylor rule .24%. So, if we were living in this model NK world, then the best we could do is to target the output gap. This of course makes perfect sense, since the output gap is the deviation of output from output in the flexible wage/price equilibrium, so the welfare criterion is in fact the output gap. And NGDP targeting is second best, and certainly much better than pure inflation targeting or a Taylor rule.

But, once we're past the actual exercise at hand we can start to ask questions. What's the model leaving out that might be germane to the problem at hand? What else could we do with this model that might lead us to different conclusions? First, the model has sticky-price and sticky-wage frictions, but we know there are other frictions of importance for monetary policy - financial frictions in particular. Surely we don't think that this NK model will tell us anything about what to do in a financial crisis, for example. I might like to know what would happen under a naive policy. That's a little difficult here as we have to worry about determinacy issues, but indeterminacy isn't stopping most NK modelers. They typically ignore the global indeterminacy that's staring them in the face. The naive policy would be a useful benchmark, as this would give us a measure of how costly price and wage stickiness is. My guess is the welfare loss in these models is very small - not enough to justify the cost of running the Fed, I think.

The big difficulty here, I think, is indeterminacy. In simple NK models - Benhabib et al. for example - we know that a Taylor rule central banker can get stuck at the zero lower bound. The alternative policy rules considered in the paper are, as the authors argue, alternative Taylor rules - set the nominal interest rate with a particular target in mind, and alternative coefficients. For NGDP targeting, in the model the central bank sets the nominal interest rate each period to peg NGDP to a constant. I would like to know exactly what sort of rule that produces, and I would like to know about the global properties of the model - not just an approximation around the steady state. The indeterminacy issue is important, as "Taylor rule perils" can explain why some central banks have had trouble hitting inflation targets (the Bank of Japan in particular). NGDP targets aren't going to do any better if they are subject to the same perils - a problem not addressed in the paper. So, the jury is out on Garin-Lester-Sims.

3. Sheedy 2014: What I like about this paper is that it takes nominal intertemporal contracts seriously. My suspicion is that NK people are barking up the wrong tree in emphasizing potential inefficiencies in nominal spot market contracts (and it's not like labor contracts are typically spot contracts anyway). Debt contracts at, for example, 3-month, 1-year, 10-year, or 20-year horizons are subject to inflation risk for those horizons. So, if the central bank can make prices predictable over long horizons, that should reduce risk in credit markets substantially, and should make those markets work more efficiently. But in Sheedy's model, smoothing nominal income is apparently a good way to insure against this uninsured risk. And Sheedy's model is quite special. Indeed, it's addressing a monetary policy issue in a 3-period model - that's not a serious approach to monetary economics in my opinion. It's hard to model valued central bank liabilities in a sensible way in a finite horizon model. My guess is that, if you developed this idea in a serious monetary model, that an optimal policy rule might look like price level targeting. Someone would have to do the work though.

Conclusion? Well, some seriousness alright, but I still don't see NGDP targeting "falling out of any serious theory." I'm not yet ready to tell any central bankers (in case they're asking) that they should drop their inflation targets and adopt NGDP targeting.


  1. Thanks for this reply, Stephen. One note: I didn't at all intend to complain about your call for "serious" theory. I merely meant to make clear that we might not both define it in the same way. In fact, it doesn't seem that we define it all that differently, although your definition is clearly somewhat more strict that mine! Note, though, that I didn't include McCallum's article in my list, though I did include Bean's article assessing it. I consider Beans's paper more theoretical, if not more "serious."

    1. I guess I'm thinking in terms of what I would accept from a PhD student. Would I accept this as an answer on a prelim or as a chapter in a PhD dissertation? True or False: Is NGDP targeting a good idea?

      I liked the picture at the top of your blog post. I've seen "A Serious Man" twice, and would recommend it to anyone. It's a Coen brothers film, and the math professor in the picture is a stand-in for Ed Coen (father of the brothers) who was a professor in the econ department at the University of Minnesota. Neil Wallace knew him.

    2. By the way, McCallum was in your post. 4th paragraph, line 6. I just clicked on the link.

    3. I do indeed link to McCallum -- twice. But his isn't one of the works I particularly recommend. In the paragraph you mention, I am explaining why I don't include him. I'm a big fan of his work, but in this case I agree that his article doesn't get down to the brass tacks.

      Your dissertation criterion is, I think, reasonable but very strict (or would be, were we to insist that the committee be doing its job). I say so because IF the case could be made well but informally, or with a rather simple model, it might not be considered dissertation worthy, just because it isn't sufficiently impressive.

      I'm glad you like the picture. That's neat trivia about Ed Coen!

    4. McCallum is mentioned in George Selgin's post, but not as one of the articles that he lists as 'serious' NGDP targeting work.

      In fact, he specifically uses McCallum as an example of an article that shouldn't be thought of as convincing, because it misses 'the crucial point that output and price level fluctuations are themselves sometimes optimal, and that NGDP targeting is in turn desirable precisely because it allows such optimal fluctuations in y and P to occur.'

      It would be somewhat misleading to not acknowledge that George did not promote this as 'serious' work.

    5. I see now. Looks like we agree on McCallum at least.

  2. Sorry, Stephen: Instead of "Bean" I should have said Bradley and Jansen.

    I've also added a brief postscript to my post acknowledging and responding to your reply.

  3. It's fairly easy to get NGDP targeting as something close to the best policy rule if the only friction in the model is sticky nominal wages. I derive this in a blog post using some kind of Calvo wage setting - with a Cobb-Douglas production function what the central bank should do in that case is to do level targeting of nominal labor income per capita. The basic idea is that if there's a negative TFP shock, then if real wages remain where they are that will cause a fall in employment and output. To avoid that, the central bank engineers some inflation to push real wages down. This is the "labor market stability" argument used by NGDP targeting proponents as I understand it. I don't buy their "financial stability" argument - for this purpose I'm convinced that price level targeting is the best policy.

    The problem, of course, is that different frictions require different monetary policy regimes to deal with. If the opportunity cost of holding money is causing an inefficiency, the optimal policy is to set the nominal interest rate on some non-scarce nominally risk-free asset at zero. If prices are sticky, the optimal policy is a price level target. If wages are sticky, the optimal policy is a labor income level target. Then there are all of the people who think if you just take interest rates negative enough that will create a recovery in a recession, so they favor a high inflation target instead. The choice of policy regime should in the end depend on actual data about the significance of each friction we expect monetary policy to mitigate, not on stylized models which are "calibrated" and then used to run "policy experiments". NGDP target proponents really think that sticky wages are the most significant friction we have to deal with, and they think if we could just fix that problem we wouldn't have to worry about another big recession. Whether that's true or not is something that should ideally be settled by empirical work - things like micro-level evidence on how sticky nominal wages really are, which choice of production function fits the data best, so on.

    The problem with the usual evidence cited by "market monetarists" is that if a central bank tries to stabilize inflation, it will look like *every* recession is caused by "tight money" (i.e falling NGDP growth) and every stable growth period caused by "good monetary policy" (i.e stable NGDP growth), even though what's really going on could be something that's independent of monetary policy entirely. Unless central banks explicitly target NGDP there is no way to directly test the market monetarist business cycle theory, something I don't think they realize, or at least if they do realize it they don't act like they do.

  4. The chart is instructive. In the 20 year period illustrated in the chart, the average Canadian's purchasing power of a dollar set aside in 1998 has fallen from 1 to 0.71 in real terms. The Bank of Canada's policy is effectively a tax on savings. How can this be considered to be good performance?

    If the Fisher relationship between real and nominal interest rates holds, then it follows that the Bank of Canada's policy of targeting 2% annual inflation has keep the rate of interest too high and held back economic growth. The only persuasive argument for a target inflation rate higher than zero is that it promotes devaluation of the national currency relative to states that pursue lower inflation rate targets and thereby improves the nominal competitiveness of the nation's exports. Perhaps.

    From a practical point of view, public sector unions argue for 2% to 5% annual wage and benefit increases based on expectations of inflation and living cost increases. Local governments tend to agree, in part because exempt employee wages and benefits increases are 1:1 with those of the non-exempt employees employed by the local government. In contrast, private sector employee wages and benefits have not kept up with inflation because of competitive pressures, principally arising from imports; and, retirees' pensions' cost of living allowance increases are nominally pegged to the level of 1/2 of the inflation rate (e.g., UBC retirees). From the point of view of the taxpayer who is not a local government employee, the Bank of Japan's inflation rate performance looks very enticing compared to that of the Bank of Canada.

    On the issues raised in the text of the article, your observation on the relative ranking of central bank targets in terms of welfare maximization is an important one: targeting the capacity gap has much to recommend it over other approaches. Would that others were as objective and forthright as you are in this article.

    1. The choice of 2% as an inflation target has nothing to recommend it relative to, say, 0% or 4%. There's some benefit of course to committing to an inflation rate. But, note that, for the average Canadian, the loss from inflation is trivial given the average holding period for cash. If you had held your wealth from 1988 to now in currency, I'd say you either having something to hide, or you're not too bright. Even just holding it as Treasury bills wouldn't be so bad.

    2. In reply, I'd have to say that if 2% has nothing to recommend it over 0%, then choose 0% as the target. If your musings on neo-Keynesian models have merit, and they give every appearance that they do have merit, then a target of 0% beneficially minimizes the output gap over time with less variability.

      The holding period for cash (e.g., scrip or paper money and coin) is indeed short, but the average holding period for deposits at commercial banks in the form of savings accounts and chequing accounts and in guaranteed investment certificates (a form of savings deposit) is significantly longer and is therefore more exposed to the loss of purchasing power, especially in older generations for whom preparation for future anticipated expenses that arise with age becomes a more pressing concern even as their incomes stagnate in nominal terms.

      It is facile in those for whom money is easy to come by to consider an individual who holds large liquid balances as either an absconder or a naïf. Reality, however, is different. One needs to consider that life is to a greater or lesser degree uncertain, even in this day and age. Holding liquid credit balances is a form of insurance; the greater the perceived likelihood of an unexpected change in circumstances for the worse, the greater the propensity to hedge by holding easily convertible liquid resources. Not all do though, to their peril. Some others maintain excessive balances--amongst these are municipal governments and school districts, in Canada.

      The central bank governor does not deign to consider those aspects but devotes his attention to the financial markets and the government of the day, if he or she wishes to maintain their trust. To that extent, a 0% target is superior to a 2% or a 4% target, except for the debtor.

  5. N-gDp targeting has already been, historically, widely debated. It should be a dead issue.

    Economic theories and properties are described by math equations. Output trajectories can be measured and interpreted using statistical outcomes. The calculations performed are generally expressed by using mathematical functions: by using rates-of-change, in the flow-of-funds, per unit of time. The functions depend upon how both the domain and codomain are defined in economic terms.

    Data dependency depends upon accurate definitions. Accurate prognostications are contingent upon "conforming statistics" (and a little grunt work). There is a lot of room for improvement in the government’s statistics. Data dependency requires, as William Barnett of “Divisia Monetary Aggregates”, an Oswald Distinguished Professor of Macroeconomics at KU (an actual former NASA “rocket scientist”), advocated, that the Fed should establish a “Bureau of Financial Statistics”.

    The figures used for determining economic flows are non-conforming, as determined by the limitations on all analyses based upon broad statistical aggregates, namely, data is not currently being compiled accurately, or in a manner which conforms to rigid theoretical concepts.

    Even so, using surrogates, anyone can predict the trajectory of both real-output and inflation. The upshot is that the math which explains gDp, demonstrates that N-gDp LPT is apocryphal period.

  6. Actually, IS there a good theoretical reason to prefer inflation rate stability (in practice an inflation rate ceiling) as opposed to stability in the price level trend?

    And isn't NGDP targeting just one special case of a dual mandate with equal weights on deviations from trend levels of Real GDP and the Price Level + willingness to use other instruments if the nominal ST interest is constrained?

    I agree that I'm not aware of any "serious" model that examines these questions.

  7. " The basic idea is that if there's a negative TFP shock, then if real wages remain where they are that will cause a fall in employment and output. To avoid that, the central bank engineers some inflation to push real wages down." Well, you could put it that way. But I think it better to say that, under NGDP targeting, the central bank doesn't "engineer" anything in response to any TFP shocks. In this case, it lets P rise, as it will tend to do in the absence of any monetary response. Thus real wages can fall without any decline in nominal wages. The alternative of keeping P from rising is to have the Fed actively respond to the TFP shock by contracting M.

    As for Friedman's rule, it is not obviously less consistent with a nominal NGDP target than with an inflation target. On the contrary: if TFP innovations translate into like innovations to the real neutral rate, the inflation rate should fall if TFP rises (thus achieving a similar increase in the return on non-interest earning money balances) and vice-versa.