* this is a revised version of the March 29, 2007 speech to the Virginia Association of Economists

Let me begin by telling you about some recent experiences. I had the opportunity earlier this year to guest-teach a couple of business school economics classes. I opened my discussions with a pair of questions, asking students to put themselves in the place of a monetary policymaker choosing a target for the federal funds rate. First I gave them a set of hypothetical facts about the state of the economy: a slowdown in housing in the wake of multi-year housing boom; rising mortgage default rates; preliminary indicators of a slowing in business investment. And then I asked them: "What are you going to do?" The students dutifully responded that this situation could call for a reduction in the funds rate. They'd obviously been doing their homework.

Next, I gave them a set of hypothetical facts about inflation: core PCE inflation, on a year-over-year basis, has been above 2 percent for nearly three straight years; after some signs of moderation, recent months' inflation numbers have moved higher; energy prices have been fluctuating around historically elevated levels and labor compensation is rising after a relatively flat period. Same question: "What are you going to do?" Once again, their response came right out of the textbook: an increase in the funds rate is needed to counter rising inflation, other things equal.

The trick of course is that both sets of hypothetical facts are drawn from the same period – basically right now. My objective was to underscore the fact that sometimes monetary policy decisions are not obvious, and that figuring out the appropriate policy action requires as complete a picture as possible of the state of the economy. Interpreting that picture and drawing policy conclusions from it can be a challenging task.

The situation I presented to the students represents a policymaking dilemma. The actions needed to bring down inflation could work against our desire to see the real outlook solidify. The facts would appear to present the policymaker with a tradeoff. You can address one – inflation or real growth – but that puts the other at risk. The obvious approach might be to decide how much "weight" one puts on each – low inflation and low unemployment – and then try to conduct policy so as to minimize the weighted average of the two. There is a superficial attraction to this popular view of the situation as a simple tradeoff. But that characterization is also, I think, an extreme over-simplification and can be highly misleading. I'd like to devote my remarks today to the relationship between inflation and the real side of the economy and to what I think that relationship implies for policymaking. As always, the views expressed are my own, and not necessarily the same as others in the Federal Reserve.

The relationship between inflation and the real economy can be described in terms of a number of alternative measures of real activity. Perhaps the most popular is the unemployment rate, because it captures the extent to which the expansion in employment generated by growth in real output is rapid enough to absorb additions to the labor force. As an indicator of the economy's use of its resources, the unemployment rate is commonly used as rough measure of the extent to which real activity is giving rise to "inflation pressures," but I will have much more to say about this later on.

Unemployment and inflation have, together, been at the center of macroeconomics for at least as long as there has been a field called macroeconomics. The relationship between these two variables is usually summarized by what's called the Phillips curve, named for A.W. Phillips, the economist who in 1957 documented an inverse relationship between unemployment and wage inflation in nearly 100 years of data for the United Kingdom.¹But the notion that rising inflation might at times be associated with rising real growth and falling unemployment had been recognized and discussed by early economists, a fact that has been emphasized by many scholars, including Robert Lucas in his Nobel lecture and by my long-time Richmond Fed colleague, Thomas Humphrey, who retired in 2005.3

Since Phillips' original paper, his curve has played a critical role in the evolution of thinking about macroeconomics and monetary policy. It captures the notable correlations between inflation and unemployment, although those correlations vary over time in important ways. But more importantly, it embodies compactly a theoretical understanding of the interplay between inflation and real economic forces. Because of its importance, and because the modern version of the Phillips curve is in some respects starkly different from the early edition, I think it will be worthwhile to briefly review some of the Phillips curve's history, before examining the role it plays today in thinking about monetary policy.

A.W. Phillips' curve

The Phillips curve began as an atheoretical relationship drawn to fit the data. The form with which people are most familiar – linking unemployment to price inflation – was first set down in 1960 by Paul Samuelson and Robert Solow for U.S. data.4 Following Samuelson and Solow, the Phillips curve was interpreted as describing a set of choices available to society each time period. According to this view, if the data suggested that price stability tended to coexist, on average, with 5 percent unemployment, we would have to live with higher inflation to enjoy unemployment persistently lower than 5 percent. This thinking led to descriptions of policy as either stimulating real activity at the cost of rising inflation, or fighting inflation by restraining the real economy. This understanding of the Phillips curve seems to have contributed to a political sentiment that, at least when inflation was relatively low, the costs of a little more inflation were worth the return in reduced unemployment.

But an alternative understanding of the Phillips curve was emerging in the 1960s. Milton Friedman and Edmund Phelps, separately, focused on the role of expectations in the relationship between inflation and unemployment.5 Specifically, they argued that while inflationary policy actions that were not anticipated by the public could have a temporary stimulative effect on the economy, fully anticipated inflation would not affect real activity. Similarly, surprise disinflation could have a temporary contractionary effect but fully anticipated disinflation would not. This meant that the observed correlation in the data between inflation and unemployment must have come largely from episodes in which changes in the inflation rate were not expected by the public.

According to the expectation-augmented Phillips curve developed by Friedman and Phelps, changes in inflation, and by implication monetary policy, could not have persistent, lasting effects on real economic activity. Over the medium to long run, economic growth and unemployment would tend to return to rates that were determined by productivity growth, population dynamics, and other characteristics of the markets for goods and labor.

To illustrate, suppose a policymaker consults a Phillips curve estimated from historical data that tells him that 3 percent unemployment can be achieved at 5 percent inflation. If the policymaker then eased interest rates in order to bring unemployment down, the policy might initially have the intended effect, provided the public continues to expect the lower previously prevailing inflation rate. But, the Friedman-Phelps framework argues, a sustained effort to reduce unemployment by maintaining inflation at 5 percent would ultimately lead the public to adjust their expectations for inflation. In the long run, unemployment would rise again to its "natural" level consistent with the real structure of the economy. Monetary policy can only have a transitory effect on unemployment.

In retrospect, many observers have labeled the Friedman-Phelps analysis prescient. They point to the 1970s as a confirmation of the implications of the expectations-augmented Phillips curve, because the period of high and volatile inflation in the 1970s brought no sustained improvement in real economic activity. In fact, not coincidentally, the general performance of the real economy during that period was poor.

The analysis of Friedman and Phelps focused attention on the critical macroeconomic role of expectations. They assumed adaptive expectations, meaning that households and firms based their expectations of future inflation on their observations of recent past inflation. (For example, models of that era frequently represented inflation expectations as a simple weighted average of past inflation.) In this view, an increase in the inflation rate may catch people by surprise, but over time they would learn about the altered policy stance. But why wouldn't people try to look ahead and foresee what the central bank was likely to do, rather than rely on the mechanical adaptive-expectations forecasting formula? For example, why wouldn't they expect the central bank to try to exploit a short-run Phillips curve, even before inflation rises? Such anticipation could rob inflationary policy of even its short-run stimulative effects.

In 1972, Robert Lucas provided an alternative, rational expectations analysis of the relationship between inflation and real activity. Under rational expectations, people's expectations are based not just on their past observations, but also on their assessment of how the economy is likely to behave, including their knowledge of the process driving policymakers' choices. The rational expectations analysis retained the Friedman-Phelps implication that only unexpected inflation would be associated with falling unemployment. But rational expectations implies that the public's reaction to policy is more forward looking than in the case of adaptive expectations.

In separate work, Lucas also showed how rational expectations presented a challenge to making policy choices based on statistical estimates of such relationships as the Phillips curve.6 In what has famously become known as the Lucas Critique, he showed that shifts in the pattern of policy behavior would cause such relationships like the Phillips curve to shift as well, so that statistical estimates from historical data would no longer be relevant for predicting the economy's response to changing policy.

Later in the 1970s, Finn Kydland and Edward Prescott built on Lucas' work and analyzed the problem faced by a policymaker when the public is forward-looking.7 They studied the temptations faced by a central bank choosing policy period-by-period. In any given period, what the public expected the central bank to do has already been determined. Given those beliefs, the policymaker can pull down unemployment with a little more unanticipated inflation. But people understand that the policymaker will be tempted to induce unanticipated inflation, and thus they don't believe prices will be stable – the inflation will be anticipated. The result is higher inflation with no gain in real activity. What the policymaker would like to do is find a way to commit to price stability, that is, commit to not give in to the temptation to attempt to reduce unemployment by inducing unanticipated inflation. The work of Kydland and Prescott highlighted to role of a central bank's credibility, in other words, the extent to which the public believes their commitment to price stability. Their work highlights the extent to which establishing credibility requires, indeed is virtually identical to, sacrificing future flexibility.

The linchpin of the link between inflation, unemployment and monetary policy is thus the public's expectations for inflation. If a run-up in inflation has been correctly anticipated, then it up will have little or no effect on unemployment. Similarly, if people expect falling inflation, then unemployment will not increase as much as it would if the disinflation were unanticipated. Thomas Sargent demonstrated this dramatically in his analysis of the ends of hyperinflations in a number of countries.8 Very large reductions in inflation were achieved at much less cost than would be predicted by a standard Phillips curve, when those reductions were part of a comprehensive package of monetary and fiscal reforms.

The role of expectations figured prominently in the disinflation that took place in the early 1980s under Fed Chairman Paul Volcker. The Fed had delayed taking strong enough action against inflation before Volcker took office in 1979, in part out of a belief that the slope of the Phillips curve was such that a fairly large increase in unemployment would be required to reduce inflation. The cost of the Volcker disinflation turned out to be substantially less than predicted however. The recent release of the FOMC transcripts from that era reveals that the public's inflation expectations were quite prominent in the Committee's discussions.9

The Modern Phillips Curve

Prior to the 1970s and 1980s, a significant methodological divide separated macroeconomics and microeconomics. That divide broke down when economists learned how to study models of the aggregate economy that were built on sound microeconomic foundations – general equilibrium models, in other words – but were also capable of addressing macroeconomic issues. The latter required models that were dynamic, because investment and interest are central to macroeconomics, and stochastic as well, because business cycle fluctuations seem to be to some extent unanticipated.10 The first generation of such models had no substantive interaction between inflation and real economic activity, and they displayed business cycles that were driven entirely by real phenomenon. The challenge was to build models that captured the inflation-unemployment link in a compelling way.

The modern Phillips curve emerged out of one approach to understanding the links between inflation and real economic activity in such well-grounded models. The approach involves specifying price-setting frictions that make a firm's choice of the price of its goods a dynamic decision that depends on expectations of future inflation. In addition, a monopolistic competition feature provides firms with the scope for meaningful price setting decisions. Under common forms of this friction, only a fraction of sellers reset their prices each period. Anticipating a length of time before the next opportunity to adjust their price, sellers will choose a price depending on what they think will happen to the overall level of prices during that interval. Aggregating across sellers, one finds that current prices (and thus current inflation) depend on expected future inflation.11 This comports with common sense intuition – money has value only because it is expected to have value in the future. So the value of money today depends on value it is expected to have in the future.

In this class of models, current inflation also depends on real economic variables, particularly the real marginal cost of production, since relative prices are set as markups over marginal costs. Under certain heroic assumptions, a one-for-one relationship emerges between the real marginal cost of production and a measure of the scale of aggregate economic activity, like output or the unemployment rate. That case results in an equation one could describe as a Phillips curve, relating current inflation to current real activity and expected future inflation.12

Although this approach has broad acceptance, in fairness I should note that it is not without its critics. Some economists view the price-setting frictions that are at the core of this approach as ad hoc and unpersuasive. Moreover, there are alternative frictions, such as spatial separation and limited information, that can also rationalize monetary non-neutrality. Nonetheless, a Phillips curve derived from price-setting frictions has become the leading model for applied central bank policy analysis.

This modern form of the Phillips curve closely resembles the expectations-augmented Phillips curve I discussed earlier, and thus shares many of the same properties. For example, a marked movement in inflation will be associated with a move in unemployment only if the inflation is different from what the public expected.

Furthermore, inflation expectations in these models are forward-looking, so expected inflation, just like inflation itself and unemployment, is an endogenous variable determined by the interaction of the conduct of monetary policy with private sector decisions and shocks to the economy.

When economists take this new Phillips curve to the data, they often find that past inflation helps explain inflation dynamics, even after attempting to control for expected inflation. This finding has led some to formulate versions of the Phillips curve in which both forward-looking and backward-looking price-setting behavior play a role.13 Backward-looking price-setters are assumed to form expectations as a weighted average of past inflation, consistent with the old adaptive expectations assumption in Friedman and Phelps, or else set prices using a simple rule of thumb based on recent inflation which amounts to the same thing.

This so-called "hybrid" Phillips curve implies an intrinsic persistence to inflation, beyond that implied by the persistence of external shocks, the effect of expected inflation on current inflation, or the conduct of policy. Moreover, in the presence of significant backward-looking price-setting behavior, inflation would be prone to respond more slowly to changes in policy. Consequently, a backward-looking component implies that the real economic costs of bringing down inflation may be higher than would the case with a purely forward-looking Phillips curve. For the same reasons, backward-looking features would mean that inflation does not respond as rapidly to a change in policy as it otherwise might, even if that change itself is well explained by the policymaker and well anticipated and understood by the public.

In these so-called "hybrid' Phillips curves, the extent to which price-setting and expectations are backward looking can matter a great deal for monetary policy. As a result, a small cottage industry of economists is now devoted to estimating hybrid Phillips curves to try to find the appropriate weight to put on backward-looking price-setting. Common estimates are that around 25 percent of agents form expectations in a backward-looking fashion, although estimates of up to 60 percent have been obtained. Many such estimates assume that the conduct of monetary policy has been constant over this period, in the sense that it was guided by a single consistent pattern of behavior. If instead one allows for the possibility of shifts in monetary policy, then the estimated weight on a backward-looking component in the Phillips curve is far lower – in fact, often zero. The intuition is straightforward. By not allowing persistent swings in monetary policy, the standard approach can mistakenly attribute inflation persistence to backward-looking expectations. The observed persistence of inflation thus might arise from forward-looking behavior combined with uncertainty in the public's mind about policy trends. If people are uncertain about the policymaker's objectives or strategy, then their inflation expectations will adjust more slowly as they learn about these features of policy by observing actual policy outcomes.

Policy

While there is reasonably strong statistical evidence of shifts in monetary policy, the premise that inflation expectations are forward-looking and that the conduct of monetary policy has evolved over time is also broadly consistent with the postwar history of U.S. monetary policy. Moreover, that history is intimately intertwined with the evolution of our understanding of the relationships embodied in the Phillips curve.14 Widely held views about the Phillips curve in the 1960s suggested that tolerating a small amount of inflation would allow permanently lower unemployment. As trend inflation steadily rose into the early 1970s, the public came to expect higher inflation to persist and the Phillips curve shifted out. Policymakers had overlooked the endogeneity of inflation expectations, and their influence on future inflation outcomes. In the 1970s, policymakers were reluctant to attack inflation aggressively out of a belief, based on the estimated slope of the Phillips curve, that high sustained unemployment would be required to reduce inflation. Belief in backward-looking expectations led policymakers to underestimate the extent to which they could influence the evolution of expectations and thereby reduce the cost of disinflation. Unemployment did increase during the disinflation that Chairman Volcker initiated in 1979, but by substantially less than had been predicted by backward-looking Phillips curves. On several occasions after that, identified as "inflation scares" by my former colleague Marvin Goodfriend, inflation expectations rose and the Fed responded aggressively by raising real interest rates above what otherwise would have been warranted.15 These episodes helped gain credibility for the Fed's commitment to low inflation. Realized inflation fell in the 1990s. Between November 1995 and March 2004 (except for a six-month period around September 11, 2001), 12-month core PCE inflation was between 1 and 2 percent. The stabilization of inflation expectations at low levels was crucial to that success.

I have given you an overview of the evolution of economists' understanding of the Phillips Curve, that is, the links between inflation and unemployment. To briefly summarize, the Phillips curve began life as an atheoretical relationship, drawn to fit historical correlations. It then became a static menu of inflation-unemployment options available to policymakers. But that approach neglected the dynamic, forward-looking nature of the decisions underlying the observed statistical relationship. The static Phillips curve broke down in the late 1960s and 1970s, just when policymakers began to rely on it, and in fact partly because they began to rely on it. The Phillips curve relationship was rebuilt from first principles in 1980s and 1990s. In the resulting modern Phillips curve the forward-looking expectations of price-setters plays a dominant role.

The modern Phillips curve has several important implications. First, the conduct of monetary policy is best understood as a pattern of behavior, or a "rule" in the broad sense of the word, not necessarily an algebraic formula. This contrasts with the view of monetary policy as a sequence of one-shot choices of policy rate that was inherent in the early version of the Phillips curve and the way my students were thinking about it. The reason it makes sense to think of policy this way is that expectations about future policy play a key role in the decisions people make today.

An immediate corollary is the importance of credibility, because low and stable inflation today requires that people believe inflation will be low and stable in the immediate future. And it deserves emphasis that credibility means giving up some flexibility in the future. Another immediate corollary is the value of central bank communications, especially communications aimed at helping people understand likely future outcomes for inflation. The value of communications is what has led several central banks to announce explicit numerical objectives for inflation.

The central theme, however, is that just as central banks are now widely acknowledged to be responsible for the behavior of inflation, they are just as responsible for the behavior of inflation expectations, because expectations are central to inflation dynamics. Related, central banks should not take expectations for granted by acting in ways that are inconsistent with those expectations, without taking into account that those expectations may change as a result. For example, economists have detected a decline in inflation persistence – that is, an increase in the tendency for inflation to return to trend after a deviation from trend. One might be tempted to count on inflation to return to trend more rapidly now, and infer that less policy response is needed when inflation departs from trend. But the fall in persistence surely reflects the fact that policy reacts more forcefully than before. Thus, a more rapid reversion to trend may be predicated on an implicit belief that the Fed will act to bring that about. Failing to do so risks a shift in expectations that makes inflation less likely to revert to trend.

The modern Phillips curve is particularly relevant to the dilemma I posed for my business school students, that is, for evaluating alternative policy strategies for restoring price stability. Would reducing inflation require large increases in unemployment? Again, the key is the behavior of inflation expectations, which now seem to hover between 2 and 2-1/2 percent, a bit below inflation itself, and somewhat higher than rates consistent with price stability. One might be pessimistic about the Fed's ability to reduce inflation below 2 percent without significant increases in unemployment if one takes the current level of inflation expectations as given. On the other hand, a strategy of clear and forceful communications about policy intentions, if successful, could bring inflation and inflation expectations down at significantly less cost. In this case, a more rapid return to price stability could be achieved, and would require less unemployment and less policy tightening than would otherwise be the case.

How responsive are inflation expectations to policymaker influence? General prescriptions should not be expected, since results will depend on the nature of central bank communication, the actions that accompany them and the context in which they are received. There are many examples of significant shifts in expectations induced by convincing people of a break from past practice. Examples include the fiscal reforms accompanying the ends of hyperinflations I mentioned earlier, the operational regime shift adopted by the Volcker FOMC in 1979, and the adoption of explicit inflation objectives by several foreign central banks. In many recent instances, FOMC actions or statements have induced short-run movements in market participants' expectations regarding the path of the federal funds rate and inflation. Fall 2005 and spring 2006 stand out as noteworthy examples. Although there may be no precise historical analogs for potential communications and actions to restore price stability in circumstances like the present day, these examples suggest, to me at least, that policymakers can have a significant effect on inflation expectations.

In any case, the centrality of inflation expectations in the modern Phillips curve reinforces the importance of consistency and credibility in monetary policymaking, since these are traits that reduce people's uncertainty about future policy and stabilize expected inflation. Central banks should not underestimate the degree to which they are capable of influencing the evolution of inflation expectations. To paraphrase the late Milton Friedman, inflation is always and everywhere an expectational phenomenon. To put it another way, inflation expectations are an outcome of monetary policy, not an autonomous help or hindrance. Central banks are as responsible for the behavior of inflation expectations as they are for the behavior of inflation. That recognition is the key to sound monetary policy.

Again, thank you for inviting me to speak here today.