FRBSF Economic Letter
2002-29; October 4, 2002
Can the Phillips Curve Help Forecast Inflation?
During the early 1960s, many economists and policymakers believed that
monetary policy could exploit a stable trade-off between the level of
inflation and the unemployment rate. One version of the hypothesized trade-off,
originally described by A.W. Phillips (1958) using U.K. data from 1861-1957,
implied that policymakers could permanently lower the unemployment rate
by generating higher inflation. Some years later, economists Edmund Phelps
(1967) and Milton Friedman (1968), argued persuasively that any such trade-off
was bound to be short-lived: once people came to expect the higher inflation,
monetary policy could not keep the unemployment rate permanently below
its equilibrium or "natural" level (i.e., the rate of unemployment
that prevails when inflation expectations are confirmed). This claim was
later borne out by the experience of the 1970s when rising U.S. inflation
did not bring about the lower unemployment rates promised by the Phillips
curve. On the contrary, higher inflation coincided with higher unemployment—a
combination that became known as "stagflation."
Though the Phelps-Friedman argument proved to be valid, there still remained
the possibility of a short-run trade-off between inflation and unemployment.
This idea led to the intellectual development of the short-run (or expectations-augmented)
Phillips curve, which says that short-term movements in inflation and
unemployment tend to go in opposite directions. When unemployment is below
its equilibrium rate (indicating a tight labor market), inflation would
be expected to rise. When unemployment is above its equilibrium rate (indicating
a slack labor market), inflation would be expected to fall. The equilibrium
unemployment rate is often referred to as the "NAIRU," i.e.,
the Non-Accelerating Inflation Rate of Unemployment.
In a recent paper, Atkeson and Ohanian (2001) challenge the usefulness
of the short-run Phillips curve as a tool for forecasting inflation. This
Economic Letter summarizes their results and discusses some evidence regarding
the empirical instability of the short-run Phillips curve.
The Atkeson-Ohanian results
Atkeson and Ohanian (2001) argue that, similar to its long-run predecessor,
the short-run Phillips curve does not represent a stable empirical relationship
that can be exploited for the purpose of constructing reliable inflation
forecasts. Their version of the short-run Phillips curve is obtained by
regressing the four-quarter change in the inflation rate on the unemployment
rate and a constant term. In each quarter, the most recent version of
the regression equation is used to construct a forecast of average inflation
over the next four quarters.
Atkeson and Ohanian (2001) show that the regression coefficient on the
unemployment rate (which measures the slope of the short-run Phillips
curve) varies substantially across different sample periods. In particular,
they demonstrate that the regression coefficient is significantly negative
in the 1960-1983 sample period, but close to zero in the post-1983 sample
period. This result is depicted in Figures 1 and 2 where we see that the
slope of the best-fit regression line is much flatter in the later sample
period. A completely flat regression line would imply that there is no
relationship between the current unemployment rate and future inflation.
Further evidence of empirical instability is shown in Figure 3 which
plots the slope obtained from a series of 15-year rolling regressions
(quarterly data from 1960 to 1974 are used for the initial regression).
The point estimate of the slope parameter varies from a low of -1.17 to
a high of +0.05. A positive value for the slope parameter turns the standard
Phillips curve intuition on its head: when unemployment is below the NAIRU
(indicating a tight labor market), inflation would be expected to fall,
not rise. Although not shown in Figure 3, the constant term obtained from
the regressions also varies substantially over the sample period. According
to the model, the ratio of the constant term to the absolute value of
the slope parameter is an estimate of the NAIRU. Hence, the regressions
imply that the NAIRU has not been stable over time. Some possible explanations
for the empirical instability of the short-run Phillips curve include:
(1) changes in the monetary policy regime that affect people's expectations
of future inflation, (2) demographic shifts in the labor force that influence
the level of the NAIRU, or (3) changes in worker productivity that affect
the pass-through of wage growth to price inflation.
the most recent estimate of the short-run Phillips curve, Atkeson and
Ohanian (2001) construct out-of-sample inflation forecasts from 1984 onwards.
They find that the Phillips curve-based forecast underperforms a naive
"no change" forecast, which says that inflation over the next
year will be the same as it has been over the most recent four quarters.
The naive forecast assumes that the current unemployment rate provides
no useful information about future inflation. (The metric for assessing
performance is the root-mean squared error of the inflation forecast.)
The authors obtain similar results for a wide array of Phillips curve
models that employ different measures of inflation, different measures
of real economic activity (as an alternative to the unemployment rate),
or additional lags of real economic activity. Finally, the authors show
that the accuracy of the naive inflation forecast is essentially identical
to the Federal Reserve Board staff's real-time inflation forecasts for
the period 1984 to 1996.
Based on the above results, Atkeson and Ohanian (2001) conclude that
inflation forecasts based on the Phillips curve should be abandoned. It
should be noted that they do not advocate the adoption of the naive model
as a structural economic relationship. Rather, they argue that policymakers
should be very skeptical of arguments to change monetary policy based
on some particular version of the short-run Phillips curve.
Robustness of the Atkeson-Ohanian results
One may wonder whether the conclusions of the Atkeson-Ohanian study
are sensitive to the post-1983 time period over which the authors compare
the out-of-sample inflation forecasts. A follow-up study by Fisher, Liu,
and Zhou (2002) examines this issue. The authors confirm that the naive
inflation forecast outperforms the Phillips curve forecast from 1985 to
2000, but find that the reverse holds true from 1977 to 1984. Over the
period 1993 to 2000, the naive forecast again outperforms the Phillips
curve forecast for a measure of inflation based on the core CPI (Consumer
Price Index), but the reverse holds true for a measure of inflation based
on the core PCE (Personal Consumption Expenditures) price index. The authors
show that the naive model consistently underperforms the Phillips curve
model when the inflation forecast horizon is shifted out to two years.
Finally, the authors demonstrate that the Phillips curve model can correctly
predict the direction of change of future inflation about 60-70% of the
time. By construction, the naive model offers no information about the
direction of change of future inflation.
The 1990s: a puzzle?
During the second half of the 1990s, the U.S. economy exhibited low and
falling inflation combined with low and falling unemployment. At the time,
many commentators and economists viewed this combination as a puzzle or
a breakdown in the short-run Phillips curve. A study by Brayton, et al.
(1999), for example, shows that the standard Phillips curve model consistently
overpredicted inflation during the late 1990s when the unemployment rate
was dropping to 30-year lows.
One factor that could help account for the late 1990s breakdown in the
short-run Phillips curve is an acceleration in the trend growth rate of
worker productivity—perhaps driven by the advent of new technologies
associated with the so-called "new economy." Recent empirical
studies by Staiger, Stock, and Watson (2001) and Ball and Mankiw (2002)
present evidence of a potential link between movements in trend productivity
growth and movements in the NAIRU. According to these authors, augmenting
the standard Phillips curve model to incorporate a declining NAIRU during
the second half of the 1990s would help account for the unusual inflation-unemployment
experience in those years (for a related study, see Lansing 2000).
During the 1970s, inflation and unemployment both trended upward for
an entire decade. This observation led economists to abandon the notion
of a stable long-run trade-off between the two variables. Nevertheless,
the evidence continued to support the existence of a short-run trade-off
between inflation and unemployment, albeit one where the slope of the
curve appears to change over time. During the second half of the 1990s,
the short-run trade-off also appeared to break down when extremely low
unemployment rates did not bring about the predicted increase in inflation.
This breakdown has drawn attention to an augmented Phillips curve model
that incorporates a time-varying NAIRU.
The need to update the short-run Phillips curve to account for changes
in slope or changes in the NAIRU (neither of which can be observed in
real time) poses a difficult challenge for anyone who wishes to use the
model for the purpose of forecasting inflation. Even within a given sample
period, the large amount of scatter around the best-fit regression lines
shown in Figures 1 and 2 reveals the fundamental imprecision of the inflation-unemployment
relationship. In light of these difficulties, the short-run Phillips curve
should be viewed as a limited tool for forecasting purposes. The evidence
suggests that the short-run Phillips curve is more likely to be useful
for forecasting the direction of change of future inflation rather than
forecasting the actual magnitude of future inflation.
Kevin J. Lansing
Atkeson, A., and L.E. Ohanian. 2001. "Are
Phillips Curves Useful for Forecasting Inflation?" FRB Minneapolis
Quarterly Review (Winter) pp. 2-11.
Ball, L., and N.G. Mankiw. 2002. "The
NAIRU in Theory and Practice." NBER Working Paper 8940.
Brayton, F., J.M. Roberts, and J.C. Williams. 1999.
Happened to the Phillips Curve?" Fed. Res. Board, FEDS Paper
Fisher, J.D.M., C.T. Liu, and R. Zhou. 2002. "When
Can We Forecast Inflation?" FRB Chicago Economic Perspectives
(1Q) pp. 30-42.
Friedman, M. 1968. "The Role of Monetary Policy."
American Economic Review 58, pp. 1-17.
Lansing, K.J. 2000. "Learning
about a Shift in Trend Output: Implications for Monetary Policy and Inflation,"
FRBSF Working Paper 2000-16.
Phelps, E.S. 1967. "Phillips Curves, Expectations
of Inflation, and Optimal Unemployment over Time." Economica 34,
Phillips, A.W. 1958. "The Relation between
Unemployment and the Rate of Change of Money Wage Rates in the United
Kingdom, 1861-1957." Economica 25, pp. 283-289.
Staiger, D., J.H. Stock, and M.W. Watson. 2001.
"Prices, Wages, and
the U.S. NAIRU in the 1990s." NBER Working Paper 8320.