View original document

The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.

January 1, 1991

Federal Reserve Bank of Cleveland

Forecast Accuracy and
Monetary Policy
by Michael F. Bryan and William T. Gavin

At has been suggested that the purpose
of economic forecasting is to make
weather forecasters look good by comparison. Despite their inaccuracies,
though, forecasts of the economy must
be useful, given the large number of
them available and the relatively high
cost of producing such information. Indeed, if forecasters can even marginally
reduce uncertainty about future business
conditions, the savings to business is
potentially huge.
But why do policymakers, specifically
the Federal Reserve, use forecasts?
Monetary policy is generally thought
to influence business conditions, but
only after a lag, and therefore an activist policy requires some prevision of
the economy.
Yet, the forecasting record of economists suggests that near-term real GNP
projections are of limited usefulness to
monetary policymakers. However, it
may be that monetary policy can affect
forecast accuracy. This Economic Commentary examines the accuracy of
macroeconomic forecasts and discusses
some implications for monetary policy.
• The Nature of the Data
The dimensions of the job faced by forecasters can be put into perspective by
examining tendencies in the data. Consider the pattern of quarterly rates of real
GNP growth and inflation over the past
35 years (figure 1). The average quarterly growth rate of real GNP was 3.1 per-

ISSN 0428-1276

cent—1.5 percent less than the average
quarterly rate of inflation. Yet, quarterly
real growth rates showed a huge dispersion, from a minimum of -9 percent to a
maximum of 14 percent, with a standard
deviation of 4.0 percent. Quarterly rates
of inflation were relatively less variable,
ranging from 0 to 15 percent, with a
standard deviation of 2.8 percent.
Note, however, that as the number of
quarters in the time period increases, the
variability of real growth narrows substantially. For example, the four-quarter
growth rate of real GNP has only 65 percent of the volatility of the one-quarter
growth rate (a standard deviation of 2.6
percent compared with 4.0 percent).
The average real growth rate continues
to become less variable as the time
period lengthens: Five-year real GNP
growth rates have only 25 percent of the
volatility of quarterly growth rates, and
ten-year growth rates have less than 18
percent of the volatility of quarterly
rates (standard deviations of 1.0 percent
and 0.7 percent, respectively).
Inflation, however, has not shown a
similar inclination to become less variable over long periods. The standard
deviation of the ten-year inflation rate
is still about 70 percent the size of the
quarterly inflation rates (a standard
deviation of 2.0 percent compared with
2.8 percent).
Moreover, historical patterns indicate
that the ten-year rate of real GNP growth
has predominantly been between 2 and 3

The policy forecaster, on the other
hand, necessarily focuses on those
aspects of the economy that policy
most directly influences. For example,
it is generally agreed that monetary
policy affects the general price level in
the long run, and aggregate output
and employment in the short run.
These are variables by which the success of monetary policy most often is
judged. Consequently, they are the
variables of primary interest to the
policy forecaster.
Alan Greenspan, on the distinctions between private and public economic
forecasters, from remarks presented at
the Annual Meeting of the National
Association of Business Economists,
Washington, D.C., September 24, 1990.

percent. Alternatively, there is no unique
tendency in the long-run inflation rate.
That is, the ten-year rate of inflation
was just as likely to be low (1 to 2 percent) as high (6 to 7 percent). Thus, the
observed patterns of real GNP growth
imply that the process that generates
output gravitates to a particular value,
while the data reveal no comparable
forces anchoring the inflation rate.
There are two forecasting implications
of these tendencies in the data. Because
deviations in quarterly changes in real
GNP growth are 43 percent greater than
for quarterly inflation rates, it seems reasonable to assume that forecasting quar-

Percent, relative-frequency distribution

One-quarter change

Four-quarter change

Five-year change

Ten-year change

Percent, relative-frequency distribution

+ 1% to +12%

One-quarter change


Four-quarter change

+2% to +9%

+2% to +7%

Five-year change

Ten-year change

NOTE: Frequency distributions are calculated using data from 1955:IQ to 1989:1VQ. Inflation is measured using the GNP implicit price deflator.
SOURCE: U.S. Department of Commerce, Bureau of Economic Analysis.

terly inflation rates is an easier proposition than forecasting quarterly real GNP
growth." Yet, a forecaster's ability to
predict real GNP growth should improve
substantially as the forecast period increases (annual growth rate predictions
should be more accurate than quarterly
growth rate predictions, and so forth).
This is not true for inflation. Consequently, forecasts of trend real GNP
growth (over five- or ten-year intervals)
are likely to be more accurate than forecasts of the trend inflation rate.
• The Forecast Record
How accurate have economic forecasts
been? Forecasts have clearly reduced
uncertainty about the economy. For example, quarterly forecasts up to one year
ahead reduced uncertainty about the
growth rate of real GNP by roughly 14
percent and by 52 percent for inflation.
The forecasting record of real GNP
growth and inflation more or less reflects
the characteristics of the data. For example, quarterly forecasts of inflation have,
in fact, been more accurate than quarterly forecasts of real GNP growth for horizons at least as distant as two years.

Of course, forecast accuracy diminishes
as the forecast horizon increases. That
is, a forecast of next quarter is almost
certain to be more accurate than a forecast of some distant quarter, since our
knowledge about the present is likely to
be of more value in predicting the near
future than a more distant future. Yet,
quarterly forecasts of inflation as much
as one year ahead have been more accurate than quarterly real GNP forecasts
one quarter ahead. Consider, for instance, the root mean square error
(RMSE) of quarterly real GNP forecasts
with quarterly inflation forecasts for the
years 1968 to 1979 (table I).5 Quarterly inflation forecasts one year ahead had
10 percent smaller errors (on an RMSE
basis) than one-quarter-ahead real GNP
forecasts (0.98 percentage points and
1.09 percentage points, respectively.)
The relative forecast accuracy of real
GNP growth and inflation also depends
on the forecast period** Because real
GNP growth has a strong inclination to
a particular trend rate—and inflation
doesn't—errors in quarterly real GNP
forecasts tend to cancel one another
over time, while quarterly inflation
forecast errors tend to accumulate.

Consequently, trend inflation forecasts
(for periods of two years or longer)
have tended to be less accurate than
trend forecasts of real GNP growth.
As an example, consider the cumulative quarterly forecasting errors for real
GNP and inflation from 1976 to 1987
(figure 2). While one-quarter-ahead
real GNP forecast errors were substantially larger than one-quarter-ahead inflation forecast errors (mean absolute errors [MAEs] of 2.8 and 1.1 percentage
points, respectively), cumulative fourquarter real GNP forecast errors were
only marginally larger than the cumulative four-quarter inflation forecast errors
(MAEs of 1.6 vs. 1.3 percentage points).
And for eight-quarter intervals, cumulative real GNP forecasts were superior to
inflation forecasts (MAEs of 1.2 vs. 1.9
percentage points).
• Accuracy and Policy
Are forecasts accurate enough to be
useful guides for monetary policy? It is
widely held that monetary policy can
influence the growth rate of real GNP
only in the short run, but affects the
price level in the long run. Therefore, a
reasonable assumption is that policy

(Root Mean Square Errors)1


Quarters Ahead






Real GNP





a. Data at quarterly rates.
SOURCE: Victor Zarnowitz, "The Accuracy of Individual and Group Forecasts from Business Outlook Surveys," National Bureau of Economic Research, Working Paper No. 1053, December 1982.

has been established on the basis of a
combined near-term real GNP/longterm inflation outlook.

realized growth rate ranged from -1.6
percent to 6.8 percent roughly 68 percent of the time. 10

Unfortunately, near-term real GNP forecasts are unlikely to show whether the
economy will be strong or weak, even
over the immediate future. Indeed, on
average, the most accurate forecasters
cannot predict at the beginning of a
quarter whether the economy will be
receding or booming that quarter with
any reasonable degree of certainty.

How should a policymaker respond to
an average forecast, if the range of precision is so wide that it includes both
economic decline and rapid expansion?
Although the large errors in quarterly
real GNP forecasts do not necessarily
preclude some countercyclical policy,
they do, however, suggest that policy
actions based on near-term forecasts
should be conservative. Simply, the
greater the uncertainty associated with
the forecast, the smaller the policy
response the forecast should induce.

One way to measure our confidence in
the near-term real GNP forecast is to examine the size of the typical forecast
error relative to the average forecast. For
example, the average quarterly growth
rate of the economy between 1968 and
1985 was 2.6 percent (at an annual rate),
and the average one-quarter-ahead root
mean square forecast error was about
4.2 percent. That is, if the quarterly
real GNP forecast was 2.6 percent, the


1976 TO 1987







Real GNP




1 1 1

2 3 4 5 6 7
Forecast period (quarters)


SOURCE: Stephen K. McNees, "How Accurate
Are Macroeconomic Forecasts?" Federal Reserve
Bank of Boston, New England Economic Review,
July/August 1988, pp. 15-36.

All of this assumes that the response of
the economy to monetary policy is
known and invariant. But because of an
uncertain and probably variable lag between policy action and its impact on
the economy, the large errors associated
with the typical forecast make it impossible to be certain that policy based on
near-term forecasts will not aggravate
the business cycle.
• Policy and Accuracy
We have considered what forecast accuracy implies for monetary policy
decision-making; policy is made in a
very uncertain environment. But can
monetary policy reduce that uncertainty?
Probably not, if its intent is to offset
short-run fluctuations in real GNP.
Although there is no clear agreement on
the mechanism that links monetary policy to the real economy, it is generally
understood that the connection between
the two depends importantly on how
the public forms expectations; monetary
policy affects the economy most noticeably when it produces an unexpected

change in the inflation trend. Inflation
rates above the expected trend, for example, are thought to increase employment and production temporarily, while
reductions in the inflation rate below
the expected trend produce an opposite
effect. Consequently, policy is unlikely
to improve near-term forecast accuracy
because monetary policy seems to be
most effective when it is unanticipated.
Monetary policy also cannot reduce trend
real GNP forecast errors, because longrun real growth is determined by nonmonetary forces such as population
growth, labor-force participation, capital
accumulation, and changes in technology. If monetary policy influences these
"real" factors, it is only through indirect
channels that are very difficult to predict.
It seems clear, though, that monetary
policy can influence the predictability
of the long-term inflation trend. In principle, the Federal Reserve can make the
trend in the price level follow any path.
In practice, policymakers pursue multiple objectives—price stability and maximum sustainable real growth. The principal policymaking body of the Federal
Reserve, the Federal Open Market Committee (FOMC), meets periodically to
consider policy options, each time
weighing the risks associated with more
long-term inflation against the risks associated with slower near-term growth.
The fact that inflation has not followed
a predictable trend implies something
about the conduct of monetary policy:
The FOMC has often judged that the
need to stimulate the real economy in
the near term has outweighed the
benefits of a stable inflation rate.
Uncertainty about the trend in inflation
can be reduced by committing to a longrun target for the price level. Such a policy may even reduce some near-term uncertainties about the economy as it
reduces the frequency of monetary "surprises." For example, the announcement
of monetary targets and liberalization of
price/credit controls in the mid-1970s
have been shown to correspond with
smaller forecast errors for the Japanese
economy. That is, under conditions
where markets were encouraged to

operate more freely and the monetary
authorities adhered to a predictable
policy, forecasters were better able to
anticipate quarterly real economic
growth and inflation.



Economic forecasts have reduced some
of the uncertainty about the future course
of the economy. Yet, forecast errors are
still too large to justify basing monetary
policy on the near-term real GNP outlook alone. This is not an indictment of
the tools or the craftsmanship of forecasters. It merely acknowledges that there is
a great deal of uncertainty inherent in the
economy over short periods of time.
The long-term real growth rate, which
is determined by real forces, has a predictable trend. And the long-term inflation rate, which is ultimately an outcome of monetary policy, does not.

• Footnotes
1. All data are annualized unless otherwise
2. The real growth rate trend is not stationary, but has probably varied within a very
narrow range over the postwar period.
3. It is not always the case, of course, that
unpredictability follows variability. Variability may be very predictable in some cases,
but these seem likely to be the exceptions.
4. We compare the root mean square error
of 79 individual forecasts from 1968 to 1979

Federal Reserve Bank of Cleveland
Research Department
P.O. Box 6387
Cleveland, OH 44101

Address Correction Requested:
Please send corrected mailing label to
the above address.

Material may be reprinted provided that
the source is credited. Please send copies
of reprinted materials to the editor.

relative to the root mean square value of the
data. Data are from Victor Zarnowitz, "The
Accuracy of Individual and Group Forecasts
from Business Outlook Surveys," National
Bureau of Economic Research, Working
Paper No. 1053, December 1982.
5. Root mean square error is a common
gauge of forecast error and is similar to the
mean absolute error except that it penalizes
errors as they increase in size. Data are from
Zamowitz, ibid.
6. The terminology here can be somewhat
confusing. The forecast period refers to the
frequency of the data, or span of the forecast;
for example, quarterly vs. annually. This
should not be mistaken for the forecast horizon, which refers to the date of a forecast for
a specific forecast period, for example, a
quarterly forecast four quarters into the future.
7. Average RMSEs from five "early-quarter"
forecasts. From Stephen K. McNees, "How
Accurate Are Macroeconomic Forecasts?"
Federal Reserve Bank of Boston, New England Economic Review, July/August 1988,
pp. 15-36.
8. See Allan H. Meltzer, "Limits of ShortRun Stabilization Policy," Presidential Address to the Western Economic Association,
luly 3, 1986, Economic Inquiry, vol. 25
(January 1987), pp. 1-14.
9. Data for the 1968 to 1979 period are
from Zamowitz, "The Accuracy of Individual and Group Forecasts." For the 1980
to 1985 period, data are 12 early and midquarter forecasts from Stephen K. McNees,
"Forecasting Accuracy of Alternative Techniques: A Comparison of U.S. Macroeconomic Forecasts," Journal of Business and
Economic Statistics, vol. 4, no. 1 (January
1986), pp. 5-15.

10. Forecasts of real GNP growth improve
as more information becomes available, but
by surprisingly little. Even "late-quarter"
forecasts of the economy are so inaccurate
that they have little relevance for policy
deliberations; that is, they are generally incapable of accurately distinguishing whether
we are in a period of boom or bust. In fact,
analysis of forecast revisions reveals that adjustments to real GNP forecasts on the basis
of incoming data frequently result in lessaccurate forecasts; adjustments to purely
statistical models on the basis of information
added by the forecaster were in the wrong
direction about 25 percent of the time during
the 1980s. See Stephen K. McNees, "Man vs.
Model? The Role of Judgment in Forecasting," Federal Reserve Bank of Boston, New
England Economic Review, July/August
1990, pp. 41-52.
11. See Meltzer, "Limits of Short-Run
Stabilization Policy."

Michael F. Bryan is an economist and
William T. Gavin is an assistant vice president and economist at the Federal Reserve
Bank of Cleveland. The authors would like to
thank John B. Carlson, Stephen K. McNees,
H. Gregory Pelt, and Katherine A. Samolyk
for helpful comments.
The views stated herein are those of the
authors and not necessarily those of the
Federal Reseve Bank of Cleveland or of the
Board of Governors of the Federal Reserve

U.S. Postage Paid
Cleveland, OH
Permit No. 385