In economic policy, new findings can sometimes break through the noise based on their timing, relevance, and importantly, who finds the findings. For example, when the President’s Council of Economic Advisors (CEA) publishes highly relevant results on a very hot topic in the annual Economic Report of the President (ERP), it matters.
This year’s ERP, out today, has a series of figures in it that may well become important. At least, I hope so, because the CEA is presenting vital information about the evolving constraints on our ability to track a relationship at the core of macroeconomics: that between unemployment and inflation. And that, in turn, suggests the natural rate of unemployment is both lower than commonly thought and a lot harder to accurately pin down.
Why is it so important? The “Phillips curve” represents the negative correlation between inflation and unemployment, and the Fed works off of this correlation to calibrate monetary policy designed to balance the competing goals of full employment and stable inflation. The larger the negative correlation, the greater the competition. If a mere tick down in unemployment led to a sharp tick up in price growth, the Fed would have to strike quickly and firmly by raising interest rates to stop unemployment from falling and prices from spiraling upwards.
Instead, these three figures from the new ERP, out this morning, tell the opposite story, showing a much diminished correlation over time.
The first figure shows the strength of the correlation between unemployment and prices over time (technically, it plots a statistic called “R-squared,” the percent of the variance in price changes explained by the unemployment rate over subsequent 20-year time periods). Peak correlation occurred in the early 1990s, but the relationship has weakened since then, and the end of the figure shows that unemployment explains almost none of the variation in inflation in recent years.
The next figure shows what this means in terms of the responsiveness, or “elasticity,” of price changes to the unemployment rate. Over the full period, the elasticity was significant: a one point increase in unemployment led inflation to fall by -0.4 percentage point. But, as you see, the elasticity is far from constant, and by the end of the period, it too is just about zero.
Economists have long used the equation you see in those graphs to back out the so-called NAIRU—the non-accelerating inflation rate of unemployment—or the lowest unemployment rate consistent with stable prices. You can see where knowing this lower limit to the jobless rate would be extremely important information to the Fed.
But the CEAs findings (see last figure below) corroborate what some of us have long maintained: the natural rate a) has been falling for a while, and b) is hard to pin down to a reliable point estimate. For those of us who were always suspicious that a “natural rate” could be identified accurately enough to guide policy, the mid-1990s were highly instructive. Back then, as you see in the chart, economists thought the lowest you could go on unemployment was 6 percent. But Fed chair Alan Greenspan recognized other moving parts in the economy, most importantly faster productivity growth, that meant unemployment call fall below the supposed natural rate without juicing inflation. He was right: unemployment was 4 percent in 2000, and inflation was pretty well-behaved.
And, of course, if you’re following this in real time today, you know that despite the fact that unemployment is down to what the Fed thinks is full employment (4.9 percent), inflation has been so low that the Fed’s long been missing their 2 percent target. Which, by the way, is perfectly consistent with the CEA charts above showing such a flat Phillips curve.
What you’re left with is a point estimate for the natural rate of around 4.5 percent surrounded by a 50-percent confidence band that in 2014 ranges from –4.3 to 6.1 (the figure shows the confidence band to stop at zero because even full-employment freak like me isn’t pulling for negative rates of unemployment). In other words, we cannot, with confidence estimate a natural rate right now.
Does all this mean the relationship between slack and inflation is dead forever? That the Fed can stop worrying? Certainly “no” on the first point. These lines move around and someday they may revert to earlier patterns. I guarantee you that if the job market keeps tightening, workers will have more bargaining power and that will lead to faster wage, if not price, growth. We may already be seeing some of that. I also worry that today’s low productivity growth, the opposite of what Greenspan faced, poses a constraint on these dynamics.
But these figures should lead to a major rethink by those, including some Fed governors, who think transient factors like cheap oil and the strong dollar are temporarily jamming the signal from the Phillips curve to the natural rate. The findings at the end of each series above are based on the last 20 years of data. None of us know the future, but when models fail like this, we must look under new rocks.
Most importantly, the Fed must be, as Chair Yellen often stresses, data driven, not model driven. They can’t know the natural rate with any confidence right now. They must know the weakness of the slack/inflation correlation. Add to those facts how critical it is for working people that we get to and stay full employment, and the bar to pre-emptive rate hikes should be extremely high.
The CEA, much like the CBO across town, does not congenitally go out on limbs. The fact that they’re publishing this work suggests it’s closer to the mainstream than when Dean Baker and I were making noises about this back in the Greenspan years. That is a real advance in economics and anyone drawing a paycheck should thank these wonks and their staffs for their honest and hopefully highly-influential work.
As powerless as labor is now, just think of the public beating public employees are taking in Wisconsin, picture a world where labor has some leverage. What is to stop big business from just raising prices? It happened in the 1970s, the very last time labor still could throw their weight around. It happened even in the face of spiraling inflation caused by commodities, especially oil. It happened even though it was a classic prisoners dilemma. Why would businesses that use consolidation, monopoly, monopsony, oligopoly, globalization, exploited immigrant labor, movement to lower wage states which actually cost more than higher wages, suddenly decide to give up profits to workers. No, not a penny. They will pass on every real wage increase to the consumer until the Fed brakes the economy, restores the surplus army of labor. Power is more important than growth. Relative wage differentials is more powerful than absolute levels. Better to be a medieval king than a modern moderately compensated babbit.
> Most importantly, the Fed must be, as Chair Yellen often stresses, data driven, not model driven.
I expect that taking that statement out of context is part of the problem I have with it but how can one be “data-driven” in the absence of a model? (Against my better judgment at the moment) I’ll quote Krugman, “It’s not the reliance on data; numbers can be good, and can even be revelatory. But data never tell a story on their own. They need to be viewed through the lens of some kind of model, and it’s very important to do your best to get a good model.” What’s Chain Yellen’s distinction between data-driven and model-driven?
> Most importantly, the Fed must be, as Chair Yellen often stresses, data driven, not model driven.
There is no “data” without a model. The decision of what statistics to collect and publish is driven by a model. It takes a model to infer that different people’s employment statuses can be tabulated on the basis of one person = one unit of employment or unemployment. GDP is a model driven number, as is the GDP deflators. Does Yellen not know this or is she making a distinction between some models that have become GOD models and other mere mortal models?
“None of us know the future, but when models fail like this, we must look under new rocks.”
I would submit that this data already is the start of a new model which must take into account that under certain circumstances the Phillips curve is unreliable.
Maybe the formalization of the model isn’t complete, but the data shows it’s there and policy recommendations can be made utilizing this knowledge, while acknowledging that additional work still needs to be done. (I was going to say refinements, but that would be premature.)
Model-driven means “My model says X, so it is X, and if data says othervise, data is wrong”. Data-driven, in this sense, should mean “My model says X, data says Y, I should change my model to be in line with data”.
> Model-driven means “My model says X, so it is X, and if data says othervise, data is wrong”.
That’s recipe for trouble. Anyone claiming the data is wrong had better have a thorough understanding of measurement uncertainty and sources of measurement error. They’d also better have validated the model and established the limits of its applicability using data which they know (“know” = very very confident) are valid.
> Data-driven, in this sense, should mean “My model says X, data says Y, I should change my model to be in line with data”.
To that I say: All models should be data-driven. (That and I’d define “model-driven” and “data-driven” a bit differently.)
Is it possible that u4, u5, and u6 are and have been relatively more important than normal in recent years?
There’s a lot of data here and a lot can be said. Let me keep it to one simple point for now:
The first graph is not that interesting to me. All it really shows to me is that there is a higher negative correlation between inflation and unemployment at times when one or the other is driving a change. In other words, the more dynamic one of these is, the higher the correlation.
The recent nearness to zero can easily be explained because there was very little change in either over that period. Unemployment has officially dropped but with inflation so close to zero, it just won’t show because there’s slack. Slack means no correlation.
Perhaps the graph makes a good way to illustrate what employment slack looks like mathematically, but I haven’t learned much from it.
I’d be interested in what you have to say about this in the context of Daniel Alpert’s “oversupply” theory: price mechanisms can’t adjust when there are hundreds of millions of third-world workers waiting to urbanize; and monetary policy barely works when it’s cheaper to hire temporary workers than to invest in expansion (because companies still have to repay the principal!).
This is an actual question, not an assertion masquerading as one. 🙂
Good ?. I’ll try to get to it. I’ve written a bit about a version of this problem in the context of global trade dynamics around the topic “savings glut.” See chap 5 in Reconnection Agenda.
Totally with you on the savings glut being a symptom of the same phenomenon.
For the correct/testable Phillips Curve see Sec.7 ‘Keynes’ Employment Function and the Gratuitous Phillips Curve Disaster’