Debunking the NAIRU myth

It’s important to try to estimate the unemployment rate that is equivalent to maximum employment because persistently operating below it pushes inflation higher, which brings me to our price stability mandate.

–Janet Yellen, January 18, 2017

A little more than half the income generated in America is paid to workers and most of the money spent in America goes to personal consumption. So it’s reasonable to think there is some relationship between the health of the job market and other important macro variables.

And, in fact, there is a robust connection between the change in the unemployment rate and the change in the real value of money spent on employee compensation per working-age American since the mid-1980s:

That chart shows the link between two real variables that have a logical connection to each other. The question for NAIRU believers is: why should a purely real variable (unemployment) have any bearing on a purely nominal one (inflation)?

In particular, is it reasonable to think there is an unemployment rate below which inflation necessarily gets faster and above which the pace of consumer price increases slows down? And even if there were such an unemployment rate at any point in time, would it be stable enough to be useful for policymakers concerned with smoothing the business cycle?

Many, including Federal Reserve boss Janet Yellen, seem to think the answer is “yes”, but the evidence points the other way, particularly since the mid-1980s.

First, some history. In 1926, Irving Fisher found a relationship between the level of unemployment and the rate of consumer price inflation in the US. In 1958, AW Phillips studied UK data from 1861-1957 and found a relationship between the jobless rate and the growth of nominal wages, although the relationship seems to have been an artifact of the gold standard given the vertical line he found in the postwar period:

Some people (wrongly) interpreted Phillips’s data to mean that there was a straightforward trade-off between the inflation rate and the unemployment rate. Policymakers could just pick any spot on “the Phillips Curve” they want. Among a certain set, the big debates in the 1960s were about whether the government should target an unemployment rate of 3 per cent or 5 per cent.

This worked out poorly, but the reaction took the form of an equally dubious idea: the Non-Accelerating Inflation Rate of Unemployment, or NAIRU. In this view, the change in the inflation rate should be related to the distance between the actual jobless rate and some theoretical level. If the unemployment rate were above this “neutral” level the inflation rate would slow down and potentially turn into outright deflation. If the jobless rate were “too low”, however, consumer prices would rise at an accelerating rate.

Suppose you believe NAIRU is a real thing. What would be the argument against pushing the unemployment rate as close to zero as possible? In theory, the cost of the policy would be ever-accelerating inflation, eventually perhaps leading to hyperinflation. But the reason to dislike excessive inflation is that it ultimately makes everyone poorer, which should, among other things, increase unemployment. (Just look at Venezuela, for a recent example.)

According to the wacky world of NAIRU, however, hyperinflation can coexist just fine with hyper-employment. Clearly there must be other mechanisms at work, or else we are leaving money on the table by allowing the jobless rate to ever rise above zero.

In case this argument seems strange, consider the following exchange the Fed had on this very topic back in July 1994 (emphasis added):

MR. LINDSEY. If we ran the model out, do we believe that if we applied some social rate of discount, the losses in output later on would be more than, less than, or equal to the gains in output in the short run [from letting inflation accelerate]?
MR. KOHN. The model itself doesn’t have, I don’t believe, losses in output from higher inflation rates.
MR. LINDSEY. Ever?…We never have a net loss in output resulting from a choice to go for inflation?
MR. PRELL. It does not take, in terms of a normal simple cost of capital calculation, a very big inflation differential to get you a net loss in the present value in the long run.
CHAIRMAN GREENSPAN. The argument as to why we get a net loss is “the Federal Reserve will react–do something.” But the question is, we are the Federal Reserve and why should we react if that’s true?
MR. LINDSEY. If we don’t believe that the present value of output in this economy will be lower by letting inflation alone, then we should let inflation go up. It’s as simple as that…Do we believe that printing money will increase the present value of output?
MR. BLINDER. Yes, I think we would. I believe that printing money will give the economy a temporary high that will not last and therefore in the integral sense that you said, yes, you get a larger integral of output over an historical period, if you never decided to end it–if you never said, when you got to 10 percent inflation, whoops, that wasn’t very good, and you went back to lower inflation.
CHAIRMAN GREENSPAN. Yes, but why would you conclude that at that point when, because as Ed Boehne says, 11 percent is still better?
MR. BLINDER. If 11 percent is better than 10 percent, if there’s no cost to inflation–I am a little bit surprised at the tenor of this conversation around here! [Laughter] There is some academic content that is–
CHAIRMAN GREENSPAN. In all seriousness, the question really gets to the models. Why would you believe that there is a cost of increased inflation from the models?

Greenspan never got a straight answer to his question but the consensus was that models based on NAIRU are basically wrong. Tellingly, it was none other than Janet Yellen who wrongly worried the unemployment rate was “too low” in the mid-to-late 1990s and would cause inflation to accelerate.

As it happens, the data don’t support the idea of NAIRU either, at least not since the mid-1980s. The test would be to compare changes in the unemployment rate against changes in the inflation rate. If NAIRU made sense, there should be a strong inverse relationship between the changes in the two series. And yet:

Regressing changes in core inflation against changes in the jobless rate gets you an r-squared of 0.11, which is basically meaningless. Moreover, that result is purely a product of the data points in the blue circle, which all occurred during the teeth of the financial crisis and could be blamed on the co-movement of employment and commodity prices. Take those out, and you end up with two perfectly unrelated series:

You get similar results if you use headline inflation rather than core inflation.

The intellectual confusion over the relationship between unemployment and inflation was especially salient during the Fed’s own policy debates in the aftermath of the crisis. The unemployment rate rose by 5 percentage points between mid-1979 and late 1982. It also rose by 5 percentage points between early 2008 and late 2009. Moreover, the jobless rate stayed above 9 per cent through first nine months of 2011. The Fed staff expected this would produce massive disinflation, or even deflation, yet it never happened.

By the January 2011 FOMC meeting, it should have been clear the old models weren’t sufficient. Instead of ditching the NAIRU concept, however, the Fed’s staff and many of the regional presidents tried to rehabilitate it by arguing the NAIRU had changed. (There were lots of reasons provided, including the extension of unemployment insurance benefits and skill mismatches.)

With the admirable exception of Richard Fisher at the Federal Reserve Bank of Dallas, the overwhelming consensus was that the crisis had raised the “non-accelerating inflation rate of unemployment” by about 1.5 percentage points:

Moreover, everyone except Fisher and the New York Fed’s Bill Dudley thought the crisis produced such long-lasting damage that the NAIRU would still be higher by 2015 (!) than it was before 2007. In reality, of course, the Fed has been forced to steadily revise its NAIRU estimates lower as the unemployment rate gradually normalises and inflation remains quiescent. The net effect was this rather ridiculous chart:

NAIRU isn’t just a useless concept, it’s a counterproductive one that encourages policymakers to focus on the jobless rate as a means to an end (price stability) even though there is zero connection between the two variables. The sooner NAIRU is buried and forgotten, the better.