Recent Posts

Friday, February 24, 2017

NAIRU (Again)

Professor Simon Wren-Lewis responded to critics of his earlier article on NAIRU. (He did not like the title of my article, which was perhaps over the top relative to the strength of my views on the matter.) Interestingly enough, I came across another paper this week which shows how this area should be approached.

TL;DR NAIRU is falsifiable, and it was falsified -- twenty years ago. Replaced by non-falsifiable output gap models.

The criticism of my article:
Accepting the concept of the NAIRU does not mean you have to agree with their judgements. But if you want to argue that they could be doing something better, you need to use the language of macroeconomics. You can say, as many besides myself have done, that the NAIRU is either a lot lower than central bank estimates, or is currently so uncertain that these estimates should not influence policy. But if you [that's me!] say that the NAIRU has to be Bashed, Smashed, And Trashed, you will not get anywhere.
I realise that when I wrote the article, I assumed that the reader was aware of the following points.
  1. NAIRU was a cutting edge mainstream macroeconomic concept in the early 1990s.
  2. When attempted to be applied in the mid-1990s by policymakers it failed horribly. Alan Greenspan's positive reputation as a central banker almost solely relies upon his ignoring the people relying on NAIRU estimates.
  3. It's 2017.
(I want to emphasise that these points are not some wacky heterodox economics mumbo-jumbo; this pretty much was the consensus mainstream view by the late-1990s, in North America at least.)

In my view, discussing NAIRU is akin to saying: "Hey, let's control inflation by having the central bank grow the monetary base by 5% a year!" or "This new band Nirvana is the future of popular music!"

The NAIRU concept is reasonable within a toy model.  I have no objection to toy models; I am currently writing a book about implementing toy models in Python. I imagine that a model with a NAIRU (or the equivalent) could be one of the first inflation models I will build within my framework. However, when introducing a toy model, we need to:
  1. discuss its predictions; and
  2. explain why it is wrong.
The problem with the NAIRU toy model is that we cannot relate the variables in the toy model to real world data.

If one wants to insist on using economist jargon, NAIRU is meaningless because it assumes that we can map a not directly measurable concept -- a generalised output gap -- essentially onto a single measured time series. We cannot say that "NAIRU either is a lower than central bank estimates [or uncertain]" if it does not exist.

[Update: Modified this section so that it is an even clearer response to a comment.]

The whole point of the standard NAIRU definition is that it is easy to observe: you just need to back out the acceleration of inflation (keeping in mind there may be other variables whose influence needs to be isolated). If inflation expectations are constant -- which they have been since the 1990s in the United States --  then:
  1. inflation [price level] will accelerate higher if UR(t) < NAIRU(t); while
  2. inflation [price level] will decelerate lower if UR(t) > NAIRU(t).
This is exactly what NAIRU stands for:  "nonaccelerating inflation rate of unemployment."

Since the early 1990s, the United States has had three multi-year expansions (each nearly a decade in length), and the unemployment rate moved significantly during those cycles. If the time series NAIRU(t) existed, it would have easily been picked up by statistical tests -- we have years of data with clean, roughly straight, trends.

However, there was no observed tendency for inflation to accelerate in any direction. This can be easily explained if we accept an alternative hypothesis: no such time series NAIRU(t) exists.

This is why most mainstream macro has retreated to discussing output gaps of various types. Output gaps have to be inferred via various statistical techniques, and they are inherently fuzzier. It may be that Professor Wren-Lewis has some of these more recent models in mind when he is referring to NAIRU; but that is a complete misnomer. If we are referring to an output gap concept, write output gap, and not NAIRU, which has a straightforward technical definition.

If you want to use standard academic terms, NAIRU is falsifiable, and was in fact falsified. The generalised output gaps that popped up to replace NAIRU are pretty much unfalsifiable.

Post-Keynesian inflation theories also rely upon accepting fuzziness (although there are similar toy models floating around). My view is that we need to embrace fuzziness and qualitative analysis.

Constructive Analysis

Finally, Flavia Dantas and L. Randall Wray published "Full Employment: Are We There Yet?" I am in the middle of various other projects, so I have not had a chance to do anything other than skim it yet. It looks interesting, but readers will have to make up their own mind.

However, the key point in the current context is that they are digging through the ensemble of labour market data, and trying to discern capacity pressures. Jan Kregel notes in the foreward:

There are indications that the unemployment rate is overstating the health of the labor market.
In other words, the unemployment rate is misleading as a result of institutional/structural factors; hence making NAIRU models useless.

(c) Brian Romanchuk 2017


  1. Apologies if I've mentioned this here before.

    Why isn't there a Non-accelerating inflation rate of competition (NAIRC)? Why does labour have to be underemployed rather than capital?

    Whenever I hear about wage price spirals I always ask how the price goes up and stays up. Because if one firm responds to pressure with a volume response it will always outcompete anybody who tries to put up prices. That's why fast food restaurants struggle to put up their prices in response to minimum wage increases. Instead there is a tight loop with the employees (fast food employees are often fast food fans) and that drives volume Henry Ford style.

    I read elsewhere today that if tariffs are put on foreign steel imports the price of steel will go up. It only will if nobody responds with a volume response, and anybody that does will out compete their price response rivals.

    So there is a system for government intervention there. If it sees a price response then clearly there is insufficient innovation and investment in that private sector market segment to justify the notion of profit and it is time to nationalise it. The threat of that should be enough to discourage price responses.

    1. I admit that I don't have strong feelings about this; just amazement that the NAIRU concept has been exhumed.

      The post-Keynesian argument that "inflation is complicated" makes a lot of sense. I wrote an article awhile ago (which I was supposed to link in with) in which I run through various service sector drivers of core inflation in the United States: rent, medical, and education. These have been major drivers of the inflation rate, and their trends have had very little to so with capacity pressure trends.

      Something like steel is driven entirely by global industry trends (and trade policy); but then again, only has limited effect on developed country consumer price inflation.

      In any event, I will be writing a more advanced text on inflation (sooner or later). My feeling is that I want to say as little as possible before I write that book, so that I don't end up publicly stuck to positions before looking again at the subject. (I did a lot of research years ago, but I want to keep as open a mind as possible.)

    2. "Why does labour have to be underemployed rather than capital?"

      This is an interesting question. I'd say the answer lies in the fact that the concept of the NAIRU doesn't necessarily involve any casual link from unemployment to inflation.

      If we increase aggregate demand, we would expect:
      a) unemployment to fall; and
      b) inflation pressures to build up.

      The principle behind NAIRU is: first, that there is a level of demand at which those inflationary pressures will cause accelerating inflation; and secondly, that unemployment is non-zero at that level of demand. That level of unemployment being the NAIRU, of course.

      The point here is that it need not actually be pressures in the labour market that cause that inflation - it could be caused by shortages in other areas - what you might think of as a shortage of capital. All that matters is higher demand causes both lower unemployment and greater inflationary pressure.

    3. First, there is a NAIRC: the rate of plant utilisation is regularly measured. Didn't you know that?

      Second, what an economy primarily runs short of as it approaches full employment of labour is . . . . . wait for it . . . . labour!! But obviously, as you suggest, shortages of other items can contribute.

    4. Worth mentioning as well that the reason we focus on unemployment rather than plant utilisation rates is that we care more about unemployment.

  2. Thankyou for pointing out (at great length) that NAIRU cannot accurately measured. I worked that out YEARS ago, and I assume most people with half a brain did so as well.

    But that does not destroy the basic NAIRU concept. The fact that you do not ACCURATELY know the relationship between two variables, does not prove the relationship does not exist.

    1. Sigh. By *definition*, NAIRU is easily measured - *if* it exists.

      NAIRU = "nonaccelerating inflation rate of unemployment." In an environment where inflation expectations are stable (which they have been since the early 1990s, at least in the United States), that means that we could easily infer the level of NAIRU based on observed changes in inflation. However, this estimation procedure fails out of sample, which means that we actually cannot estimate it.

      Since it cannot be measured, it does not exist.

      You are using some random output gap concept, and calling it NAIRU. I guess that is what Professor Wren-Lewis is doing as well. I don't care about your concept; I am discussing the standard definition.

    2. Sigh. It is nonsense to suggest that just because "inflation expectations" are stable, ergo NAIRU can be accurately measured. If the general population thinks or "expects" there will be good weather for the next week, does that prove professional weather forecasters can accurately predict the weather.

      Pure false logic.

      "Since it cannot be measured, it does not exist"

      OK, since the weather cannot be accurately predicted a week in advance, ergo there is no such thing as weather forecasting - have I got that right? Or are you going to argue that because the weather cannot be accurately predicted, ergo all weather forecasts are useless? I can't wait to find out.

    3. The definition of NAIRU implies an easily measured response. That response is not observable. NAIRU does not exist, by its own definition.

      You are using a different definition of NAIRU than the technical definition. Why not accept that and move on?

    4. Ralph Musgrave, maybe you should come up with a new name for what you are describing, because what you describe is not NAIRU and I don't see why you bother to defend NAIRU when you obviously don't agree with it either.

      As Brian has demonstrated, NAIRU refers to a NUMBER. The estimates of that number have been wrong. Without a specified number the NAIRU is no longer what proponents claimed it to be. Estimates of NAIRU that are inaccurate, but lead to policy changes anyway, can result in hundreds of thousands of unemployed people for no reason what so ever- not even the rationale that the unemployed serve to keep inflation low.

      Maybe your interpretation should be called something else. A relationship between full employment and excess demand and inflation is not NAIRU. Come up with a new word. (SNAFU and FUBAR are already taken)

  3. Regarding pricing power in macroeconomics I would look at industries where a scarce natural resource is extracted via capital and labor such as oil, gas, and large fisheries. Could prices fall even on the downward slope of peak oil or gas extraction? Yes if the costs of capital investment have already been recovered in the distant past and if firms face sufficiently low operating costs. If capital investment outpaces aggregate demand the investments will not be validated by profit and some level of bankruptcy will ensue. Firms gain pricing power without collusion when it is possible for dominant firm(s) to raise prices, as the price leader, and other firms also raise prices, as price followers, then the whole industry takes more revenue without competition for market share. If the playing field is level then all firms are assumed to be price takers. Industries with large and long term capital investments would not exist without any expectation of pricing power, and when this expectation does not manifest, the social solution is bankruptcy to get the assets free from invalid debts.

  4. According to our generous host:
    1. “NAIRU = "nonaccelerating inflation rate of unemployment."”
    2. “The definition of NAIRU implies an easily measured response.”
    Point 2 is a non-sequitur.
    Jerry Brown correctly says: "estimates of [NAURU] have been wrong". This is an argument for better estimates, if possible. It is not a reason why NAIRU estimates "should be called something else".

  5. Kong, NAIRU estimates should be called something else- something like BS. Ralph Musgrave's argument is not NAIRU and he should call it something else. It is not a bad argument, it just is not NAIRU.

    NAIRU is the nonaccelerating inflation rate of unemployment. For that concept to be useful as far as economic policy it requires an ability to measure that rate accurately in advance. The concept implies measurability. That has not happened. The measurements have been wrong, therefore NAIRU theory has been wrong. I don't see why this is so difficult for people to understand. If the estimates had been correct it would lend credence to the idea (but not prove it), but if the proponents own measurements turn out to be wrong then either the idea is wrong or it is entirely useless in that it cant be measured.

    1. KongKing,

      Given the advances in digital computing, you could put a grid over the range of UR values, and calculate all the potential NAIRU time series, and reject all of them.

      According to the definition, deviations from NAIRU has an easily inferred effect on inflation, which is allegedly only affected by "random shocks". The probability of those shocks lining up to actual outcomes is about 0%.

      Mainstream macro abolished uncertainty, and replaced it with randomness. Complaining about an "uncertain" NAIRU without a theory that allows uncertainty is ridiculous.

      Meanwhile, if you cannot come up with a NAIRU estimate, how does the representative household?

  6. A PK model need not be NAIRU per se. Rather it could be conflict inflation with the unemployment rate being one variable that influences worker bargaining power. That is how I think about inflation.

    And getting a robust reading on worker bargaining power? No way! I've constructed some indices which I think sensible using principle components but there is no way I would use them to make finely-tuned predictions.


Note: Posts are manually moderated, with a varying delay. Some disappear.

The comment section here is largely dead. My Substack or Twitter are better places to have a conversation.

Given that this is largely a backup way to reach me, I am going to reject posts that annoy me. Please post lengthy essays elsewhere.