Every academic discipline has dirty secrets. Those of economics include the fact that some of our best known principles are based on very thin data. The Phillips curve, which is relevant to much of today’s economic news, is one of these.
This graph, developed by New Zealand-born economist William Phillips, relates changes in wage rates or general inflation to rates of unemployment. It thus bears directly on Federal Reserve policymaking as the central bank tries to return interest rates to normal levels with unemployment low but with both wage rates and general inflation beginning to rise after years of little change.
This has a very real impact, as seen in the dramatic stock market volatility seen earlier this month following the January employment report. Jobs were up, wages were up, the report showed — seemingly good news that nonetheless spurred U.S. stock indexes to drop sharply. Why? Fear of inflation and higher interest rates. Enter Bill Phillips.
Phillips’ curve is intended to show a tradeoff between these variables. High unemployment can fall at first without there being any increase in wage rates. But as it falls farther, wages start to rise. At some point, wages could increase sharply but unemployment would bottom out at some “natural rate” minimum.
Looking at it another way, if wage increases or inflation changes are large, these can be reduced at first without any increase in unemployment. But as such inflation is reduced further, unemployment begins to rise. At some point, unemployment can become extreme, but inflation ceases to fall.
This curve appears in every introductory macroeconomics text. And it is the subject of a recent bulletin from Jim Paulsen, chief investment strategist at Leuthold Group in Minneapolis, who looks at recent data and relates it to current questions. Moreover, it implicitly is in the mind of every policy maker in every central bank in the world. So it is as salient an idea as when Phillips first set it out 60 years ago this coming fall.
But despite its impact, both the underlying data and the economic theory are pretty thin.
Phillips was the rare economist who actually had an interesting life and practical experience, including stints as a movie theater manager, crocodile hunter and RAF fighter pilot before settling on economics. If his health had not failed at age 55, in large part due to over three years in a Japanese prisoner of war camp, he might have gotten a Nobel Prize.
His study, based on analysis of 96 years of British wage and unemployment data starting in 1861, was largely historical. But other economists, including eventual Nobelists Paul Samuelson and Robert Solow, broadened the focus from wage changes to general inflation and looked at data from other nations. Most reached similar conclusions. There was a tradeoff between inflation, either of wages or consumer goods, and unemployment. This tradeoff was small at either extreme, but real in the ranges at which most economics operated.
This related directly to the Keynesian ideas that had revolutionized economic theory and were gaining weight in central bank and general government policy making. Standard economics long held that government should avoid trying to manage an economy. But in 1936, British economist John Maynard Keynes argued that in times of low output and high unemployment, government could vary taxes, spending and the money supply to spur growth and restore jobs. If instead the problem was high inflation, the same levers could be used to return to price stability.
Phillips’ curve seemed to mesh with this perfectly. There was a tradeoff between inflation and unemployment that government could manipulate to optimize economic conditions. This is still accepted wisdom, at least at some level, in central banking and in much of government.
In intro textbooks, the curve is always a smooth sweep. But if you look at plots of actual data in Phillips’ paper, or in Paulsen’s bulletin , they often look more like random dots scattered across a page.
A fundamental problem is seldom discussed: It is dangerous to use data on a simplistic relationship taken over long periods of time to make inferences about the future. If the underlying social or economic or physical structure changes, such inferences can be wrong.
Take an example from climate. There is data on water flow levels in the Minnesota River and rainfall in the basin it drains going back 150 years. You can plot them against each other as Phillips plotted wage increases against unemployment.
However, the relationship between rain events and river stages for the Minnesota was very different in 1870, when most of the land was still prairie or woods, and few acres had been drained, than in 1920, when much land had been farmed, largely in grains like wheat, oats and barley or as hay. Corn was only one crop among many. And while steam dredges had drained wetlands, there still was little artificial drainage.
This situation in turn was distinct from 1960, when corn was more important and it and soybeans were displacing wheat, oats and other grains. Mechanically-installed clay or concrete tile drained many acres.
This in turn was distinct from 2017, when there were few acres other than corn or soy and installing corrugated plastic tile with a differential GPS controlled attachment on a big tractor was easy.
Land use patterns changed. What was true decades ago is not a valid basis for predictions of what will happen now.
This sort of problem occurs again and again. It is treacherous to treat data from a series of observations over time as random samples from some unchanging population if you want insight on how some cause-and-effect will work in the future. Measurements of tides in the Bay of Fundy over the last 200 years may serve as reliable forecasts of them in the next decade, but that is not true for rainfall-riverflow relations in Minnesota.
And a pattern of economic relationships that seemed to be true in one country over a century may not be either.
Of course, Phillips’ study was the first of many. Similar ones were done for other nations over other time periods. And the analytical techniques used have been far more sophisticated than potting dots on a piece of paper. But the dangers of making inferences from time-series data remain larger than many economists recognize.
Ironically, economic theory, being reconstructed as Phillips was dying in 1975, concluded that his curve was bunk. The Rational Expectations school of thought, based largely at the University of Chicago and University of Minnesota, argued that there was no unemployment-inflation tradeoff except perhaps in the very short term. One could not reduce unemployment by accepting a bit higher inflation or vice versa.
This group won the day within economics. By 1990, nearly every economics textbook presented the Phillips curve in a very different way than it had in 1975. A half dozen economists involved in this “revolution” in economic thought, including four associated in some way with the University of Minnesota or the Federal Reserve Bank of Minneapolis, got Nobel prizes for their work.
Yet in further irony, despite what happened in economic theory, the logic of the Phillips curve underlies central banking. The research department at the Minneapolis Fed was a bastion of rational expectations theory, but the last three presidents of this bank, going back over more than 30 years, have made their voting decisions at FOMC meetings in a Phillips curve framework. Neel Kashkari, the incumbent and Narayana Kocherlakota, his immediate predecessor, have been very explicit in this. And every time the policymaking open-market committee issues a statement saying that it will or will not tighten the money supply based on wage or CPI changes versus unemployment, Bill Phillips’ ghost waves a little flag.
The same is true every time the stock market bobbles in reaction to news about wages, consumer prices and unemployment. The next few years will give us more data and perhaps some econ texts will get revised back to where they were in 1980.