Recent Posts

Saturday, February 18, 2017

NAIRU Should Be Bashed, Smashed, And Trashed

Professor Simon Wren-Lewis recently wrote an article "NAIRU Bashing," in which he attempts to salvage some value out of the concept. As observed by Ramanan, his defense of NAIRU can be summarised as: There Is No Alternative (TINA), A lot of what Professor Wren-Lewis wrote might appear similar to what I have written on the topic. (I will publish the relevant excerpt from Interest Rate Cycles: An Introduction  shortly, in case readers would wish to contrast and compare.) However, the belief that there is no alternative to NAIRU is silly. If economics were scholarly (as I discuss here), knowledge would be additive, and we would not have such debates.

What is NAIRU?

NAIRU stands for the "nonaccelerating inflation rate of unemployment." As noted by Ramanan and others, this phrase does not actually make sense; people just liked the cool acronym. I discuss it in the upcoming book excerpt, but I have some already published articles on the concept (example).

The quick summary of the NAIRU story is that if the unemployment rate is below the NAIRU level, inflation will rise (the time series of the price level will accelerate); inflation falls if the unemployment rate is above NAIRU.

A Vaguely-Defined NAIRU is Meaningless

Professor Wren-Lewis accepts that the NAIRU is hard to measure, and so it is a easy target for ridicule,
The NAIRU is the level of unemployment at which inflation is stable. Ever since economists invented the concept people have poked fun at how difficult to measure and elusive the NAIRU appears to be, and these articles often end with the proclamation that it is time we ditched the concept. Even good journalists can do it. But few of these attempts to trash the NAIRU answer a very simple and obvious question - how else do we link the real economy to inflation?
I believe it is safe to say that most post-Keynesians accept the link between labour market tightness and (wage) inflation, and so one might interpret this as being qualitatively similar to "NAIRU." However, this is perhaps too generous to NAIRU.

NAIRU implies that there is a well-defined number which can be compared to the unemployment rate that allows us to give quantitative estimates for the changes in the inflation rate. If that well-defined number cannot be inferred from the data, it contradicts the entire premise.

If we want to create a mathematical model of the economy, it looks like the best that we can hope for is some form of a "generalised output gap" that does not have a direct correspondence to any particular economic time series. The unemployment rate probably offers useful information about capacity pressures within the economy, but its usefulness depends highly upon the institutional context (as discussed in the next section).

[Update] I have had a few comments, as well as being in a multi-way argument on Twitter. I am not particularly good at following Twitter arguments when they involve 4+ people, so I did not have a lot to say there. I just wanted to add a few points here.

  • I have written similar things to Professor Wren-Lewis. I just wanted to try to explain what the standard post-Keynesian objections (as well as mine) to his remarks. The complaints are theoretical, and do not imply that post-Keynesians believe there is no relationship whatsoever between labour market "capacity pressures" and inflation. In fact, I am sure there are some teaching models somewhere used by post-Keynesians that are functionally equivalent to NAIRU. The point is that "teaching models" are not viewed as a gospel truth.
  • Revising NAIRU to be a range of values for the unemployment rate where there is no effect on inflation is a change of definition; the "R" stands for "rate", not "range." One can make this revision, but it makes the concept even more useless. We now need to estimate a range for this deadzone. Since inflation has not really moved over the past few decades, that estimated deadzone could easily be between 3%-10%, which means that the model is almost entirely non-falsifiable; we just increase the deadzone to match the increased volatility of the unemployment rate.
  • Furthermore, NAIRU estimates in the euro periphery are implausibly high (basically just the moving average of the realised unemployment rate). As such, this is an example of the NAIRU concept breaking down at high unemployment rates.
  • NAIRU estimates are based on looking at observed changes in inflation and the unemployment rate (plus whatever variables the modeller wants to add, such as inflation expectations). The level of NAIRU is often moving at a pace comparable to the measured economic series, and so the model is just telling is that trends in the inflation rate are persistent, and so the model cannot distinguish between the case where the unemployment rate affects inflation, and the case where it does not.

For an analytical example of the failure of NAIRU, we need not look any further than the behaviour of the U.S. economy over the past cycle. Despite the massive spike in the unemployment rate after the crisis, deflationary pressures were mild. Meanwhile, the steady drop in the unemployment rate during the expansion has had no observable effects on inflation. We have seen hawks warn for almost a decade that the unemployment rate was "near NAIRU," and that inflation would take off any minute now. However, the classic example of the failure of NAIRU was during the 1990s, which I discussed in Interest Rate Cycles.

Dubious Policy Implications

Simon Wren-Lewis:
I could go on and on, and write my own NAIRU bashing piece. But here is the rub. If we really think there is no relationship between unemployment and inflation, why on earth are we not trying to get unemployment below 4%? We know that the government could, by spending more, raise demand and reduce unemployment. And why would we ever raise interest rates above their lower bound?
I’ve been there, done that. While we should not be obsessed by the 1970s, we should not wipe it from our minds either. Then policy makers did in effect ditch the NAIRU, and we got uncomfortably high inflation.
This is an example of the questionable policy advice that belief in the NAIRU brings. By attaching magical significance to a single time series, we apparently do not need to think about economic institutions.

I am not going to hold myself out as an expert on the 1970s inflation. That said, I would point out that observers pointed out the inflationary bias of various government policies in real time. (I have Hyman Minsky in mind, but he was hardly alone in making such claims.) If your policies have an inflationary bias, you should not be surprised by high inflation. Meanwhile, policymakers were surprised that high inflation coexisted with high unemployment, so Professor Wren-Lewis' comment that they "ditched NAIRU" does not appear to fit the historical record of what policymakers were thinking.

Take the MMT Job Guarantee as an example of the importance of institutions. If the Job Guarantee were implemented with no indexation of the guaranteed wage rate, there is no plausible reason to expect any form of inflation to result (looking through any one-time changes that result to the change in the structure in the labour market). If government policy were aimed at preserving price level stability, it would be entirely reasonable to expect that 0% inflation would coexist with 0% involuntary unemployment. (There would presumably be people not in work, yet not in a Job Guarantee job, but that would represent a voluntary choice. For example, professionals would likely prefer to remain unemployed and undertake a job search while living off of savings.) Such an institutional framework would make the NAIRU concept utterly useless for forecasting inflation.

Bonus Point: Contradicts Rational Expectations

One of the side effects of accepting that NAIRU is hard to measure is that it invalidate rational expectations. If economists who specialise in monetary policy cannot come up with a method of predicting inflation in a quantitative fashion, it is clear that the households -- even the magical "representative household" -- cannot do so either.

Realism and Inflation Forecasting

The post-Keynesian literature on inflation is broad and deep. My flippant summary: inflation is complicated.

Although that sounds useless, it is unfortunately realistic. I was involved in the inflation-linked bond market. This is a huge market, with returns directly linked to inflation (as the name suggests...). The ability to forecast said returns using a reduced form model and one time series would be of utmost interest to market participants.

The reality is that NAIRU-based models do not work. And it is not just a question of adding a few wrinkles to the model. At this point in time, tens of thousands of people have had access to time series databases and analysis software, and they have been hammering away at all of the possible models like the proverbial infinite monkeys on typewriters. (I write from the experience of being one of those monkeys earlier in my career.) If a reduced form model was even close to being able to provide useful forecasts, that model would have been discovered by trial and error alone.

As a result, inflation strategists have been forced to dig around the ugly details of the construction of the CPI, and employ a lot of ad hoc guesswork. The post-Keynesian approach helps explain why the reduced form modelling approaches are generally failures.

(c) Brian Romanchuk 2017


  1. "I believe it is safe to say that most post-Keynesians accept the link between labour market tightness and (wage) inflation"

    Only in the private sector market. So there is a simple solution to that. Don't go anywhere near the limit. You hold policy slightly tight to ensure that you don't get the problems. 'Hard' employment is then inflation constrained as it is now.

    You then pick up the slack with 'soft' employment - a Job Guarantee - which has the advantage of providing an additional anchor to the private market because Job Guarantee workers are a credible threat to those at the low end in the private sector.

    Policy only gets into trouble because it is trying to run the private sector over hot to get as much private sector employment as possible.

    Once you get away from needing that it is far easier to run a stable private sector.

  2. “As noted by Ramanan and others, this phrase does not actually make sense; people just liked the cool acronym.” Ridiculous remark. There are loads of other acronyms in economics and the English language. I don’t know of anyone else attacking an acronym because advocates of the acronym have an exaggerated idea of their own “coolness”. Reason no one else does that is because to do so is, as I say, ridiculous.

    “NAIRU implies that there is a well-defined number…”. Well if you want to claim that, please cite some NAIRU advocate who claims NAIRU can be accurately measured. I’ve never come across one. Certainly I never use the acronym with the intention to suggest it can be accurately measured.

    “Despite the massive spike in the unemployment rate after the crisis, deflationary pressures were mild.” The word deflation has two quite different sense. You should differentiate between the two, but I assume you’re referring to the “falling prices” sense.

    Well the reason prices do not fall all that rapidly in a recession was pointed out about 80 years ago by Keynes: it’s the “wages are sticky downwards” effect. That is, it is difficult to cut wages, particularly in heavily unionised sectors. And if wages do not fall, then prices won’t fall much either.

    “Meanwhile, the steady drop in the unemployment rate during the expansion has had no observable effects on inflation.” Well that’s because the NAIRU idea primarily refers to the big RISE in inflation that occurs when unemployment falls. I.e. NAIRU advocates agree with one of the basic characteristics of the Phillips curve, namely that it is more or less horizontal at high unemployment levels. Certainly I do.

    But that point is probably too subtle for the NAIRU bashers.

    “…we apparently do not need to think about economic institutions.” It is complete and total nonsense to suggest that NAIRU advocates are not interested in possible improvements to the labour market, like JG, improved training, improved employment agencies and so on.

    1. The phrase that NAIRU stands for does not make sense. It almost looks like English, but it isn't.

      The definition of NAIRU is that it is a single number, and the difference between that number and the undemployment rate is supposed to be directly tied to movements in the inflation rate. If there is no single number, nor can the relationship between that number and the inflation rate cannot be measured, the concept makes no sense.

      NAIRU implies a very specific model about inflation. If that model is invalid, it makes no sense to speak about it. If you want to call a vague relationship between the labour market and inflation "NAIRU", you are effectively unaware of the definition of the term that you are using.

      The recent example is only one of the areas where NAIRU failed horribly in practice. The 1990s was when most sensible people gave up on the concept.

      NAIRU is not a policy position, other than as an argument for complacency. (Wren-Lewis' statement is a fairly blatent example of policy complacency.) A person can advocate any number of labour market reforms, and still use a defective analytical tool (NAIRU). The fact that the NAIRU depends upon institutions is another reason why it is utterly meaningless.

    2. “The phrase that NAIRU stands for does not make sense. It almost looks like English, but it isn't.” What – so all acronyms that “look like English” should be banned? I’ll bet there are literally HUNDREDS of those! What about IT (information technology)? Let me know when your campaign to ban the latter acronym starts: I’ll be the first to make a mockery of it.

      Next, the fact that human beings as yet cannot ACCURATELY measure a relationship is not a reason to reject the relationship if it is obvious to anyone with some common sense that the relationship exists. It is blindingly obvious that when unemployment falls below some level, inflation rises. Ergo the relationship exists.

      Prior to pictures being take of the far side of the Moon, astronomers did not suggest there were no craters there just because they couldn’t accurately count them. Medics do not reject what looks like a relationship between a substance suspected of being carcinogenic and cancer just because the relationship can’t be measured with extreme accuracy.

      “The fact that the NAIRU depends upon institutions is another reason why it is utterly meaningless.” Whaat? Yes: NAIRU declines if the quality of education and training improves. What of it? The advocates of NAIRU are well aware of that.

      The number of murders and thefts probably also depends on the quality of education. Ergo all suggestions that there might be a relationship between crime and other variables (e.g. alcohol intake) are invalid? I never heard such nonsense.

    3. Umm, "nonaccelerating inflation..." is the phrase that makes no sense.

      The whole point of NAIRU is that there is a simple relationship between the unemployment rate and inflation. If we cannot actually find that relationship in the data, "NAIRU" does not exist. It's a very simple argument.

      If you want to make up your own definition of NAIRU, feel free to do so. However, I am not referring to your private definition here.

    4. If you cannot demonstrate a way to calculate NAIRU in real time, and this methodology does not give good out-of-sample inflation predictions, it's a mythical concept. You could just as well use Father Christmas (FC) to explain inflation.

  3. "...inflation falls if the unemployment rate is above NAIRU."

    This is an interesting point. NAIRU is usually discussed in terms of what happens when unemployment goes below it; there is often no discussion of what might happen in the other direction.

    It is quite conceivable that there should be a level of unemployment below which inflation accelerates, but above which it is stable - at least for a significant range. That level would seem to meet the normal definition of the NAIRU.

    In fact, it seems quite plausible to me that at very low levels of unemployment, workers' bargaining power becomes such that inflation would go out of control. However, on the whole, I'd say such a level is well below normal unemployment levels and as such is rarely if ever relevant. Destabilisation of inflation is more likely to come from shocks external to the labour market.

    1. I guess there is the price stickiness argument, so that the effect is not always symmetric. However, if the effect only worked one way (only when below), the inflation rate would have to be monotonically increasing. That obviously does not fit real world post-1980 data very well. During most of that era, we were well away from the zero bound, and so that story cannot be invoked.

    2. Yeah - when I wrote "at least for a significant range" I meant that beyond that range inflation might be decelerating.

      But also I think that whether inflation is decelerating, stable or accelerating may not be a question just of where you are on the unemployment scale, but also where you presently are on the inflation scale. The higher inflation, the more likley you are to be in an accelerating inflation situation.

    3. Correct. Your point ties up with the Phillips curve which is horizontal at high unemployment levels.

    4. Nick,

      Adding a deadzone really makes identificatio near impossible. In practice, this means that NAIRU starts kicking in at the late point in the cycle, when inflation is accelerating. We can only identify that point by waiting for inflation to rise. Since NAIRU magically moves around to "explain" the data each cycle, we can only backcast when we are in the effective region.

      If inflation trends have momentum (which a glance at a chart says it does), the NAIRU-with a deadzone is effectively non-falsifiable.

    5. I'd been inclined to agree. For various reasons, I would expect great measurement difficulties here, as per my closing comment in my own post. I think my point here was really that it may be that there is some kind of NAIRU in its most basic sense, but that doesn't make it a useful concept.

  4. "In fact, it seems quite plausible to me that at very low levels of unemployment, workers' bargaining power becomes such that inflation would go out of control."

    That can happen at very high levels of unemployment - because the employment market is non-fungible.

    It's the usual problem of over abstraction.

    1. Sure, the labour market is segmented and that is very relevant. My own recent post on this is based exactly on that point.

      Whilst some labour markets will be tight even when unemployment is high, in order to get accelerating wage driven inflation, you need labour shortages in enough markets to drive the average up. This would represent a quite severe mismatch.

  5. Are there good examples of marked wage inflation in the absence of other policy or situational considerations? Where the only change has been a tight labor market?

    I ask because what I see on the ground just doesn't support it. My perspective on this may be limited, though.

    1. If we look at the developed countries, there's generally one period of rising inflation: the 1960s-1970s. The 1980s saw disinflation, and then "flatflation" from 1990 on. (Places like Japan had slightly different experiences.)

      On top of those trends, there were cyclical movements, but those cycles were small relative to the big trend movements.

      Not a whole lot of data points to work with, once we take into account the tendency for inflation to follow trends. It's pretty easy to backcast the 1970s, the question is whether we can say anything useful on a forward-looking basis. NAIRU failed horribly when put into practice in the 1990s. My impression is that no competent market economist takes the original idea seriously. (There are new versions, which are probably just as bad in practical terms, but do not fit the original definition of NAIRU. NAIRU itself was a replacement for the failed "natural rate of unemployment" concept.)

  6. A point that may be obvious yet unstated: the aggregate Wage Bill is reduced by increasing aggregate unemployment levels even if wages are sticky for the employed.

    Currently the link below opens a 22 page paper by Ball and Mankiw:

    The NAIRU in Theory and Practice

    The paper develops some basic equations to discuss the theory of NAIRU, econometric problems, and policy. Starting on pdf page 8 there is an estimate of the NAIRU where U* = 6.1% for the US from 1960-2000 based on a simplified model. There is a discussion trying to explain why the so-called NAIRU peaked around 1980. I find it hard to believe in the concept of NAIRU as specified in such models as anything other than a particular expression of a curve fitting exercise.

  7. Neil (first comment above) seems think that since it's what he calls the "private sector market" that is responsible for the inflation that occurs near full employment, the solution is public sector employment in the form of JG.

    Just one slight problem there: there's been an astronomic expansion in the public sector over the last 100 years, but no obvious improvement in unemployment levels as a result.

    Incidentally I worked out the solution to that little conundrum in the papers I was writing 10-20 years ago.

    It's also good to see Neil now accepting that inflation rears its ugly head near full employment, so presumably he now accepts the basic idea behind NAIRU.


Note: Posts may be moderated, and there may be a considerable delay before they appear.

Although I welcome people who disagree with me, please be civil.

Please note that my spam comment filter appears to dislike long "anonymous" posts. I get no warning about this, and only go through my "spambox" infrequently. The best bet it to keep comments short, and if you think the spam filter struck, let me know with a short comment.