The Phillips Curve: A Primer

Economists have debated the relationship between inflation and unemployment at least since A.W. Phillips’s study of U.K. data from 1861 to 1957 was published 60 years ago. The idea that a tight or slack labor market should result in faster or slower wage gains seems like a natural corollary to standard economic thinking about how prices respond to deviations of demand from supply. But, over the years, disputes about this Phillips curve relationship have been and remain fierce.

As the U.S. labor market tightens, and unemployment approaches levels we have not seen in more than 15 years, the question is whether inflation is going to make a comeback. More broadly, how useful is the Phillips curve as a guide for Federal Reserve policymakers who wish to achieve a 2-percent inflation target over the long run?

To anticipate our conclusion, despite evidence of a negative relationship between wage inflation and unemployment, central banks ought not rely on a stable Phillips curve for setting monetary policy.

To understand the controversy surrounding the Phillips curve, and the limits of its usefulness as a policy guide, we briefly look at some data before turning to the history of this economic idea. The following chart shows the relationship between the civilian unemployment rate, shown on the horizontal axis, and wage inflation two years later, plotted on the vertical axis. The observations are divided into three distinct periods: the 1960s in red; the 1970s, 80s and 90s in blue; and 2000 to 2015 in black.  Amid the scatter plots, we show simple regression lines for each period. While there appears to be a relationship in the early and later periods (the lines fit well), the three intervening decades are another story altogether.

Wage inflation and the unemployment rate (quarterly), 1960-2015

Source: FRED.

With this background, we now turn to the intellectual history of the Phillips curve. Based on experience through the early 1960s, many U.S. observers concluded that there was a stable, negative relationship between the level of unemployment and the level of inflation. As a result, policymakers could choose where along this tradeoff they wanted the economy to operate. This is the standard interpretation of  Paul Samuelson and Robert Solow’s original 1960 paper examining the relationship in the United States. (For a recent discussion, see here.)

But, to the extent that there was a consensus, it was short-lived. In his 1968 Presidential address to the American Economic Association, Milton Friedman made two key points: (1) Phillips had failed to distinguish real wages from nominal wages, and; (2) it is deviations of unemployment from some normal (or natural) rate that matters. The first point means that it is essential to take account of expected inflation, while the second implies that changes in economic structure can lead to movements in unemployment that are not going to put pressure on inflation in one direction or the other. (Edmund Phelps developed this second point simultaneously.)

Friedman and Phelps had not come to their conclusions in a vacuum. While the first half of the 1960s was characterized by extremely low inflation—prices rose roughly 1¼ percent per year and wages 3 percent—the second half of the 1960s already looked quite different. By the end of the decade, prices were going up at a rate of more than 4 percent per year and wages were rising at a 6 percent annual rate.

Taking these lessons to heart, researchers sought to patch their estimated Phillips curves, adding various bells and whistles. A common approach was to treat expected inflation as backward-looking—estimated from the recent past—while allowing the natural rate of unemployment to change slowly over time. This treatment implied that the change in inflation, rather than the level, was related to deviations of the unemployment rate from this time-varying natural rate. Economists then estimated the natural rate as the level of unemployment consistent with stable inflation.

The expectations-augmented Phillips Curve, as it was called, distinguished between the short run, when wages and prices could be sticky, and the long run when they are flexible. The implication is that, in the long run, after prices and wages have had time to adjust, inflation is unrelated to labor market conditions. This is the corollary of Friedman’s adage that “inflation is always and everywhere a monetary phenomenon”—real quantities like unemployment cannot permanently affect aggregate prices. So, while there may be a short-run relationship in which changes in inflation are negatively associated with deviations of unemployment from the natural rate, there is no long-run relationship.

But the 1970s brought major new challenges for Phillips-curve adherents. First, there was the combination of high inflation and high unemployment that came to be known as stagflation (these are the blue triangles in the upper part of the chart). Second, there were new theoretical reasons to view the Phillips curve relationship as a mere artifact that was inherently unstable. Following the Friedman and Phelps logic, in his seminal theoretical work, Robert Lucas emphasized the importance of focusing on deviations of inflation from what is expected. From this, he derived what has come to be known as the Lucas supply curve, which can be thought of as another version of the Phillips curve, but with a profoundly different interpretation. In the Lucas model, the observed inflation-unemployment relationship is unlikely to be stable and is not something policymakers can exploit.

The policy lessons following from these new ideas were profound. For example, if inflation dynamics are dominated by forward-looking expectations, then, in contrast to the world of the backward-looking Phillips curve, it should be possible to lower inflation without inducing high levels of unemployment. Instead, a credible policy commitment, which convinces people that a resolute central bank will set policy to lower inflation, can itself be sufficient to lower inflation. Put differently, if people believe policymakers will act to lower inflation, the authorities can achieve this goal without inducing a recession.

Today, economists agree that commitment is key to promoting price stability. There also is evidence that an aggressive, speedy effort to lower inflation (buttressed by a policy commitment) is less costly in terms of unemployment than a gradual one. In practice, however, commitment has its limits. The Volcker disinflation of the early 1980s is a case in point. In 1979, U.S. inflation rose into double digits, peaking at nearly 15 percent, a peacetime record. A concerted effort by the Fed brought this down below 5 percent in three years. But the cost was high: unemployment rose by almost 5 percentage points from 6 to nearly 11 percent. As bad as this is, it was much less than the 8-percentage-point increase the naïve relationship estimated using 1960s data predicts. In other words, the Volcker commitment appears to have paid some dividend. 

To continue our intellectual historical narrative, the workhorse (New Keynesian) macroeconomic models in use today still link inflation to slack in the labor market. That is, they incorporate a Phillips curve. However, unlike their antecedents, these formulations do not posit an exploitable tradeoff between inflation and unemployment. Instead, they imply that there is a tradeoff between the volatility of inflation and the deviations of unemployment from the natural rate (or of production from its norm). Put differently, the models that frequently guide central bank thinking today allow policymakers to choose only the relative speed at which they wish to return inflation to its target or the unemployment rate to its norm.

This brings us back to the chart at the beginning of this post. There, we noted that over the past 15 years or so, the relationship between the level of inflation and the level of unemployment once again seems stable. A simple explanation for this appearance is that the Fed has built up substantial credibility, helping to anchor inflation expectations at something like 2 percent. When expectations don’t change, the challenges that arose during the high inflation period of the 1970s and 1980s recede, making the Phillips Curve stable. (For recent discussions of the Phillips Curve, see the papers by Olivier Blanchard, Robert Gordon and Laurence Ball and Sandeep Mazumder.)

However, this apparent stability does not imply an exploitable tradeoff. The very crude estimates shown in the chart suggest that each percentage point deviation of unemployment from the natural rate alters wage inflation by 0.3 to 0.4 percentage point. This is less than one-third of the estimated 1960s impact. But even this weak estimated response looks to be too large. In a recent report that we co-authored, the impact on the trend of price inflation is much smaller: less than 0.1 percentage-point change in inflation per percentage-point change in the unemployment rate. This minute response is what analysts mean when they say that the Phillips curve has flattened.

What does such a flat Phillips curve mean for monetary policy going forward? You might think that it is an invitation for policymakers to let unemployment fall well below the natural rate, and that doing so would not risk a significant pickup in inflation. In our view, that temptation would be very risky. First, there is no guarantee that as inflation goes up, expectations will stay anchored where they are today. If, instead, expectations rise with the level of inflation, then the pattern of the 1970-2000 period (when the Phillips curve was, at best, elusive) could return. Second, the relationship between inflation and unemployment could be nonlinear. More specifically, what would happen if unemployment fell below 4 percent for any length of time? Will inflation start to rise rapidly? Put differently, a curve that looks flat at normal or high unemployment rates could turn very steep at low levels (see here and here).

So, like many economists and policymakers, we still consult models that incorporate a relationship between inflation and unemployment. And, we take some solace in nearly two decades of stable inflation. However, the re-appearance of what seems to be a stable relationship between the level of unemployment and the level of inflation is probably just another sign that effective monetary policy has given rise to stable inflation expectations. As a result, a robust monetary policy that aims to limit big risks (including big policy errors) should not assume that the Phillips curve is stable. If expectations shift, the Phillips curve will, too.

The bottom line: past success in restoring price stability provides little room for complacency. As a guide for policy, the relationship between inflation and the unemployment rate is no more reliable today than it was 50 years ago.