Sunday, June 30, 2019

The Phillips curve is still dead

Greg Mankiw posted a clever graph a month ago, which he titled "The Phillips Curve is Alive and Well."

No, Greg, the Phillips curve is still as dead as Generalissimo Franco.

The lines, in case you can't see them are the employment-population ratio 25-54, and the average hourly earnings of production and nonsupervisory employees. Wait a minute, the Phillips curve, as it appears in contemporary macroeconomics, is a relation between inflation, a coordinated rise in prices and wages,  not real wages or hourly earnings, and unemployment or the output gap, not the employment-population ratio. How does the traditional Phillips curve look? Here is unemployment vs. CPI inflation

and here is inflation vs. the GDP gap:

Here is "core" (less food and energy) inflation vs. unemployment:

Except for one little blip in the depths of the 2009 recession. The Phillips curve is dead. (Long live the Phillips curve, the crowd sings nonetheless.) Inflation trundles along, ignoring unemployment or the output gap.

What's going on? Primarily, I think Greg goes deeply wrong in looking at average hourly earnings, or wages for short. The whole art and magic of the Phillips curve is about inflation, the rise in both prices and wages.  Greg's graph is perfectly sensible microeconomics. The labor market is tight, demand for labor is high, you have to pay people more to get them to work. The rise in wages is a rise in real wages, a rise in wages relative to prices.

Similarly, one might imagine tight product markets, with strong demand, as a time that output prices and measured inflation would rise relative to wages.

The puzzle and promise of the Phillips curve is the idea that tighter labor markets, traditionally measured by the unemployment rate, correlate with higher wages and prices. That takes more doing. Typically, you have to think that workers are fooled into working for what they think are higher real wages, and only later discover that prices have gone up too. And you have to think that firms rather mechanically raise prices passing on higher labor costs, and keep selling things when they do. Despite the intuitive appeal of tight markets leading to rising prices and wages, that simple intuition is wrong to describe a correlation between tight markets and both prices and wages, which is what the Phillips curve is and was.

The employment-population ratio is a little bit curious but less so. Much modern labor economics doesn't focus on unemployment.

What is happening should be cause for celebration by the way -- real wages are rising. From growth to inequality to the hand-wringing about declining labor share, it's hard to find anything bad to say about that!

Greg's "Phillips curve" also does not extend backwards. Here's what happens if. you push the data slider to the left on Greg's graph, going back to the 1960s rather than start in 1990:

Greg's correlation is absent in the heyday of the Phillips curve. Greg's alive Phillips curve was born in 1990.  (What you're seeing is, of course, the rise in labor force participation, particularly among women, until 1990.) That's why the traditional (ex ante!) Phillips curve really was about gap measures

The conventional inflation-unemployment Phillips curve also died just about contemporaneously with the Generalissimo:

The negative correlation which Phillips noted around 1960 turned to a positive, or stagflationary correlation in the 1970s. One nice negatively correlated data point in the disinflation of 1982 is it.

The policy world, including the Fed, ECB, and related institutions, continues to believe in the Phillips Curve, and as causation not just correlation: tight labor markets cause inflation. But its evident death is causing some unsettled feelings for sure.


Catching up on Greg's blog, I also found a lovely and sage quip:
Washington Post columnist Robert Samuelson argues "It’s time we tear up our economics textbooks and start over." He uses my book as a prime example. Perhaps not surprisingly, I disagree. My summary of Samuelson's article: Economics textbooks should be more like economics journalism, says an economics journalist.
There is so much "starting over" in the air -- modern monetary magic on the left, neo-mercantilism on the right -- that understanding long settled questions is indeed what education should be about. (And not just the sharing of untutored opinions.)
Textbook writers, on the other hand, emphasize those things that are true, important, and unknown to the typical reader (an 18 year old college freshman). Newness has little relevance. The lessons of Adam Smith do not apply only to the 18th century, the lessons of David Ricardo do not apply only to the 19th century, and the lessons of John Maynard Keynes do not apply only to the 20th century. They are timeless ideas that may not make good news stories but should be central to introductory economics. Just as Newtonian mechanics should remain central to introductory physics.
Well, I think Keynes will go the way of phlogiston, but I agree with the point, and anyway a good 19th century scientist should know what phlogiston is.



Or maybe we should call it the Phillips Cloud. Here is the traditional inflation vs. unemployment graph, for the 1990-today sample and then the whole postwar period

Some economists run a regression line here, and proclaim the Phillips curve to be flat. They conclude, unemployment is incredibly sensitive to inflation -- just a bit more inflation would make a lot of jobs. I conclude it's just mush.


  1. I had to look up "phlogiston".

    I am perhaps among the untutored.

    Still, the way many pundits and academics discuss the outlook for prices, one would think an inflationary phlogiston is embedded in every fiber and crevice of the modern economy.

    1. Yeah me too I had to look up that word.

      i'm skeptical of the philip's curve as a reliable macro economic indicator.. i feel the scope is too big for it to be reliably accurate as there can be cyclical issues on the economy like the midwest flooding affecting prices for an indefinite time frame.

  2. A small point: Phillips's Phillips curve related to wages, not general price inflation.

  3. An exogenous increase in the money supply leads in the long run to an equal increase in the price level. That's what some people call money neutrality or a vertical long-run Phillips curve. Some economists argue (forcefully, e.g. Golosov and Lucas 2007) that approximately the same thing happens also in the very short run -- just a quarter or two after the shock. That's a short-term vertical PC for those who prefer to put inflation on the left-hand side, a flat one for those (like Golosov and Lucas) who put inflation on the right hand side. Note that a close-to-vertical short term PC (in the traditional sense) is "super-alive" in that a small increase in output goes along with a big inflation spike. And a dead PC is one which is so flat that you need a huge movement in output to produce only a small (close to none) inflation response. John seems to refer to the latter case when talking about a dead PC. But I find it somewhat bizarre when people appeal at the same time to flexible prices (and hence Golosov-Lucas!), and also talk about the dead PC!! Is it dead or is it super alive? You see, after the monetary shock either inflation, or real variables (or both) should move. Something must give, otherwise somebody just burnt that extra money...

  4. When a fellow economics major at UChicago told me in the early 1970s his instructor in the econ class was teaching the Phillips Curve, which had not been included in my own experience with the class in the early 1960s. I told him I thought the idea was nonsense upon first learning it, and I am pleased to see you agree.
    It does seem to be based on a logically fallacious leap from a clear micro phenomenon in the labor market to some general statement about the price level. I find too much of macro to be built on those fallacies of composition.
    As F.A. Hayek sagely observed: "Neither averages nor aggregates directly act upon each other, because choices are made by individuals."

    1. When I first encountered the Phillips Curve in the mid 1970s it went along with statements that the unemployment vs inflation curve seems to have shifted (because that was the start of high unemployment and high inflation together). It also went with statements that various conditions (e.g. high inflation) were now permanent.

      Over the 45 years since my first economics class I've continued to hear about "permanent" changes to the economy or markets. Most if not all have instead proved to be transient.

  5. I too had to google "phlogiston." Interesting that its debunking was ultimately an empirical exercise.

    Additionally insightful as Wikipedia points out that:
    "Many people tried to remodel their theories on phlogiston in order to have the theory work with what Lavoisier was doing in his experiments. Pierre Macquer reworded his theory many times, and even though he is said to have thought the theory of phlogiston was doomed, he stood by phlogiston and tried to make the theory work."

    The historical roadmap for an easy exit ramp from Phillips curve theory is not optimistic.

  6. The Phillips curve has to be a myth. Why? Instead of looking at "unemployment", just think of the total amount spent on wages. If people spend more money on wages (employment x averages_wages), there should be less to spend on other things, that means that there should actually be a negative correlation between the total spent on wages and the total spent on consumption. That would have to mean that after accounting for the effects of inflation, price changes and wage prices have to be negatively correlated.

    Of course, the other reason the Phillips curve is a myth is that the only things that permanently affect inflation are technology and the money supply. Notions of increased spending, saving or borrowing are functions that by construction revert. Saving now means more spending later. Borrowing now means spending more now, but spending less later.

    When wages increase, this might correspond with a temporary boost in spending and demand, or a sudden boost in the money supply that inflates away wealth meaning people have to work more.

    But it's just as likely to correspond with increased utility of labor. That increased utility of labor is a technological innovation, and will correspond with a decrease in prices.

    Here's a simple test that we could actually use to disentangle the two:

    1. When wages revert to some equilibrium, this should correlate positively with prices.

    2. When wages experience permanent innovations, this should correlate negatively with prices.

    Simply run an AR model, where changes in wages are a function of past changes. We should see that the expected changes correlate positively with price changes, and the unexpected changes correlate negatively with price changes.

  7. Don't we expect the Phillips Curve to be absent in the data if the Fed is successfully controlling inflation? That aside, it looks like in the first graph that in each recession, unemployment jumps up and inflation then drops. I would argue that in normal non-recessionary times, the Fed is keeping inflation under control, so no PC would be evident. In a recession, the Fed loses control, so inflation drops and unemployment jumps.

  8. I'll put out here that government policy can reverse the Phillips curve. If a government borrows and spends along with unemployment, prices will go up with unemployment. Similarly, if unemployment is due to regulations that make it more costly to hire someone at a given wage, we'll see a negative correlation between prices and unemployment.

    That said, in a market where a government does not react to unemployment or fiddle with regulations, a shock to the quantity of labor supply, a shock to technology that lowers the demand for labor as an input, a shock increase in spending from savings would all find Phillips curve results.

    Given a successful government policy to correct for price changes as a function of employment by expanding or contracting the money supply, we should expect the disappearance of the Phillips curve.

    The real way to assess the curve would be to control for government policy.

  9. Whereas, there is no single entity called "the price level," and
    whereas a rise in the CPI is merely a symptom of inflation, and
    whereas the amount of money being created is inflation,
    Therefore we economists need to readjust our theories to more-closely comport with causal factors.

  10. From the FRED data, we see almost no correlation between levels of unemployment and changes in CPI. However, almost every way you look at it, you see negative contemporaneous correlation between changes in unemployment and changes in CPI.

    I'll posit that contemporaneous changes are just a function of less people working, and more people saving. That means that people's utility from wealth changes, so that prices for consumption goods fall.

    That means that what lowers prices is a change from employment to unemployment or a change from consuming to saving. But once that change is over, no continuing effect on prices can be found.

    You can check this out by measuring the correlations of changes in the FRED data, or by running a simple VARMA model to disentangle surprises from expected changes.

  11. It's tough talking about a Phillips Curve without actually drawing one! :-)

    I plotted annual data from 1948 to 2018 and I see the usual Phillips Loops, including for the most recent period. A decent guess at the natural unemployment rate is still ca. five per cent, perhaps a tad more.

    Would gladly make my picture available here, but I don't know how to.

  12. I've always felt pretty uncomfortable with the hand-waving required to explain the phillips curve. Tighter labour markets result in higher wages (fine), but that translates into higher prices (really?). Why is it that higher input costs for labour are passed on? What proportion of businesses costs are actually labour, and what is capital? And if labour costs are high, why not substitute capital instead? Why is it that we're assuming that higher labour costs will end up with an economy like Zimbabwe, instead of an economy which optimises out the demand for labour like Japan?

    Oh, and I'm pretty sure that a regression with a flat line of best fit means that the coefficient is zero (or at a minimum the R squared is very low). So a flat phillips curve is a curve with very little confidence in a relationship which is effectively non existent.

  13. I agree that the scatter-plot is a cloud, but No, that doesn't prove that a PC does not exist. Take any model with a Phillips curve (e.g. Golosov-Lucas 2007 or if you prefer Calvo or really anything in-between) and run it with both demand (e.g money or gov't spending) and supply shocks (e.g. oil price or wage markup), and you will get a cloud like the one shown. Lack of unconditional correlation is no proof of non-existence of a relationship.

    But *conditional* on a demand (e.g. money increase) shock, something must happen. Either prices will go up, or output, or a little of both. Surely John is not arguing that absolutely nothing happens? Check Zimbabwe or Argentina... I'd say they have close to vertical PCs. But other countries certainly have flatter PCs than that.

    What does the slope of the PC depend on? One factor is long-run inflation. At high inflation, firms reprice faster and workers demand higher wages more often. The curve is steeper in that money impulses are transmitted faster to the price level, as in Golosov-Lucas. At low steady-state inflation, e.g. close to zero, firms and workers don't have as much incentives to change their prices or wages so often and so the economy is more Calvo-esque: monetary impulses take longer to pass to the price level.

  14. Possibly of greater significance is the issue of what this portends for monetary policy and the federal reserve forecasts of the direction of the economy.

  15. I think this is 2-0 to Mankiw. Both official inflation and the unemployment data is suspicious. Your graphs are summed up with "garbage in, garbage out".
    You need to show the philips curve is wrong using macro data that is reliable.

  16. "Washington Post columnist Robert Samuelson argues "It’s time we tear up our economics textbooks and start over." He uses my book as a prime example. Perhaps not surprisingly, I disagree."

    According to Wikipedia Mankiw has grossed 42 million from selling his text books. A while ago I priced his textbooks at Amazon and the price was over $200 for one textbook. Today, it looks like the price has gone down a bit.

    Perhaps he is doing a live economic lesson about how a captive audience pays more for goods than those that can shop on a free market.

    I, surprisingly to me, agree with Samuelson. Economics, as a discipline, does not work.

  17. Looking to the unemployment-cpi chart, it seems to me that the relationship is nonlinear: during the recessions the relationship is sound but fades away after recession.

  18. The Philips curve was made immortal, impervious to any mortal data, by NAIRU.

  19. Powell just said in Senate testimony (7/11/2019) that the relationship between unemployment and inflation has gone away. I missed the details but it was on the chyron when getting coffee.

  20. Well. Here is my stab at it all from a slightly different angle:

    First, check this out. It plots out over time the unemployment rate and the labor force participation rate. In particular, check out what transpired before and after 2008. It doesn't look like much is going on:

    However, if one adjusts the scales with the same data (All from FRED, by the way):

    You can more easily see the trend/relationship between the unemployment rate and the labor force participation rate. Notice as the labor force participation rate falls, so does the unemployment rate (a sign of structural unemployment. Thoughts start to go towards what's going on in the gig economy, too).

    Now, if we take a look at this (Yes, it was from about 6 months ago!):

    We can see over time the relationship between the unemployment rate and monetary policy via the Fed.

    As soon as unemployment hit 5%, the Fed appears to have stuck to its guns regarding NAIRU: The Fed started to increase rates.

    Now, as this relates to the Phillips Curve madness (and I have serious problems and doubts with the Phillips Curve) - and I do not believe the Fed uses the UNRATE alone to shape policy, even though it's part of their dual mandate - the UNRATE is very, very rough. It possesses some of the same problems with making decisions using an average only; something is lost/missing and doesn't tell the whole story (mean, median, sd, variance, skew, kurtosis, and on and on helps fill in the gaps). The LFPR and underemployment add important features to the employment/unemployment story.

    Also, what about cost-push and demand-pull as it relates to inflation, hmm? They can both work in the same direction. Inflation has hovered slightly below the Fed's stable price mandate of about 2%. Deflation is the real enemy, and without enough stable inflation, deflation could rear its ugly head, severely affecting consumption, employment, and aggregate demand.

    Also, check out this article:

    Here's a great blurb from that same article that gets to the heart of the problem with the Phillips Curve:

    "Gordon is one of the economists who finds it hard to contemplate a world without the Phillips curve. Without a correlation between unemployment and inflation, he said in his 2013 paper, the Fed would not be able to calculate the natural rate of unemployment or the amount of slack in the economy. 'In such a world, the Fed would be operating like a captain of a giant ocean liner operating in a fog, with no instruments to warn of icebergs to the left or to the right.'"

    UNRATE isn't enough - never has been. It's useful, but it has to be used in the right way. The Phillips Curve isn't that useful in my mind.


    1. Just found this from Mankiw:

  21. "Typically, you have to think that workers are fooled into working for what they think are higher real wages, and only later discover that prices have gone up too."
    I don't think this is the point. Suppose you are a worker, and you have more negotiating power vs. your employer thanks to tight labour markets. You want to translate that in higher real wages. What will you do? The simplest way you can use your better position is to demand higher nominal wages. The employer will then pass the extra wages into higher prices proportionally to his labour costs. Let's imagine now that (1) all workers get 5% higher wages (2) labour share of GDP is 60%: then you will have 3% price inflation. Crucially, real wages have gone up by 2%. If inflation expectations were correct, this is exactly how much workers could get. Otherwise, the process is repeated until equilibrium. So, the idea is that real wages rise, but bring up prices by a smaller amount in the process.
    This story seems extremely intuitive to me. Of course, this is an "all other things equal" story, where interest rate, exchange rates, productivity etc. do not change. As they do, the end result on price inflation could cancel out, go in the opposite direction, or just cover a smallish philips curve like effect with large uncorrelated fluctuations. Not that I'm really qualified to draw conclusions on this, but I felt you were misrepresenting the other position.

  22. Prof.Cochrane, I wonder what's your opinion on this recent ECB working paper which concludes that the Phillips curve is alive and well in the Euro zone.

    Anonymous Reader

  23. Hi John,
    The Phillips curve only looks dead because it is a business-cycle-phase dependent relationship. During most of the recovery, you are right: there is no Phillips curve. But it exists as the economy slips into recession (as in Stock and Watson 2010) and it exists as the economy enters the "overheating" phase. If we fix our coefficient estimates at their 2006:12 levels and then condition only on unemployment data, we nail the entire Great Recession inflation dynamics.

  24. I must say that I strongly disagree with the article for a couple reasons, the first being that economists who argue this point paradoxically try to look at the bigger picture, but also narrow the scope of the debate to two exogenous variables: unemployment and inflation. While these are the two variables of the phillips curve, it is negligent to argue that because there is not always a clear relation between the two on a line chart, that the phillips curve is dead. This is because other variables affect inflation. For example, a Phillips curve relationship would be cleaner if interest rates rose and fell at the same rate of unemployment. However, because interest rates do not always move directly with unemployment, the line graph becomes slightly messier. Reason two is Ceteris Paribus: all other things equal. tying into my first point, we can't expect all other variables that affect inflation to stay equal. However, if they were to stay equal, the Phillips curve relationship would be much clearer. While I respect the opinions of Cochrane, I don't think this article takes into account the presence of other relationships with inflation, causing the point to fall short for me.

  25. I think the point was to slam Greg Mankiw for attempting to slide in alternate variables to prove the curve was still relevant and in doing so demonstrate the propensity for economist to lock onto theory's that have been created post a good 10 year run and then never let go of them if they perform well for 10 years after creation. You can trust my take on this. I have a Bachelor's Degree.


Comments are welcome. Keep it short, polite, and on topic.

Thanks to a few abusers I am now moderating comments. I welcome thoughtful disagreement. I will block comments with insulting or abusive language. I'm also blocking totally inane comments. Try to make some sense. I am much more likely to allow critical comments if you have the honesty and courage to use your real name.