Full text of Economic Commentary (Federal Reserve Bank of Cleveland) : Forecasting Turning Points With Leading Indicators, October 1, 1989

View original document

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

October 1, 1989
six errors, respectively, with the two-

•

month rule.

The Composite Index of Leading Indi-

flation, and Forecasting, National Bureau of

• Implications
Outlook

cators is often used by analysts seeking
to forecast business peaks and troughs.
It can provide useful information, but
its value as a forecasting tool is quite
limited.

Economic Research Studies in Business
Cycles, No. 24, Ballinger Publishing Company, Cambridge, Mass., 1983,page 370.

for the Current

As of August 31, it appeared that the
ILl had declined in February, March,
May, and June 1989, thereby satisfying
the criterion of falling in four of the last
seven months. Thus, taken by itself, the
ILl began forecasting a business peak
when the ILl figure for June 1989 was
released on August 4. But this forecast
was confirmed by only one of the alternative indexes, so a peak was not
being forecast by the more complete
rule, which requires confirmation by
two of the altemative indexes.
On September I, the ILl for June was
revised upward to no change, and the
ILl for July was reported to have risen.
This new information, in effect, canceled the ILl's forecast of recession, because the ILl then appeared to have
fallen in only three of the last seven
months.
This episode illustrates how the possibility of data revisions reduces the usefulness of the ILl as a tool for forecasting business peaks. It also illustrates the
value of using the more complete rule,
which had not signaled and still does
not signal a peak.

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

Conclusions

The ILl has a good record in turning
down and up before business peaks and
troughs. Forecasters, however, do not
have the luxury of knowing the future
path of the ILl-they only know where
it bas been, subject, of course, to revisions. Moreover, the ILl's value as a
forecasting tool is limited by its tendency to give false signals.
For the purpose of forecasting peaks,
the ILl can be used with a four-monthsout-of-seven rule, supplemented by the
requirement that two of the alternative
indexes confirm the forecast. For
forecasting troughs, the ILl should be
discarded in favor of the Long-Leading
Index, using a two-month rule.
Based on these rules, the ILl, which has
not declined in four of the last seven
months, is not signaling a recession.

2. Geoffrey H. Moore, Business Cycles, in-

eCONOMIC
COMMeNTORY

3. See Marie P.Hertzberg and Barry A.

Federal Reserve Bank of Cleveland

Beckman, "Business Cycle Indicators:
Revised Composite Indexes," Business Conditions Digest, BEA, January 1989,pages 97102.
4. The criterion of four declining months in
a seven-month period was met 12 times in
periods of expansion during the last 42 years.
Four of these episodes were false signals,
while the other eight correctly signaled impending business peaks. Thus, the rule was
correct only two-thirds of the time.

Forecasting Turning Points With
Leading Indicators

S. Strictly speaking,the Ratio is not an

index of leading indicators.The Long-Leading Index and the Short-Leading Index were
developed by Columbia University's Center
for International Business Cycle Research
(CIDCR).Although they have been calculated back to January 1948,they were first
published by CIDeR on October 6, 1987,
and have been published in Business Conditions Digest only since April 1989.

by Gerald H. Anderson
and John J. Erceg

of the ILl and examines its record as a
forecasting tool. We conclude that,

The views stated herein are those of the
authors and not necessarily those of the

major topic of current interest to
economic forecasters is whether the
U.S. economy is headed for recession in
the near future. Some have concluded
that recession is on the horizon, or has
already begun. One piece of evidence
offered to support that view is the Composite Index of Leading Indicators
(ILl), which reached a peak in January
1989, and by mid-August was reported
to have fallen in four of the last five

Federal Reserve Bank of Cleveland or of the

months.

Economic data series that might be
helpful in identifying changes in ag-

Gerald H. Anderson is an economic advisor
and John 1. Erceg is an assistant vice president and economist at the Federal Reserve
Bank of Cleveland. They wish to thank Susan

•

Footnotes
1. For examples, see Hilary Stout, "Leading
Indicators Fell Sharply During May, Darkening Outlook," The Wall Street Journal, June
29, 1989,page A2; and "Leading Indicators
Down 0.1%," The New York Times, August
4, 1989,page 25.

Byrne for her research assistance .

Board of Governors of the Federal Reserve
System.

Generally speaking, a major task in
economic forecasting is to anticipate
turning points in economic activity points when a business expansion will
reach its zenith, or peak, or when a business contraction will reach its nadir, or

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

trough. A reliable method for forecasting turning points has continued to
elude forecasters, despite advances in
mathematical and statistical techniques.

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

The financial news media frequently
refer to "leading" indicators of economic activity as clues, if not predictors, of changes in activity. The media
commonly report the magnitude of an
increase or decline in the ILl, which,
they conclude, implies continued
growth or contraction in economic activity. Some reports also claim that the
ILl is the government's "chief economic forecasting gauge."!

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

This Economic Commentary

discusses

the composition and ostensible purpose
ISSN 0428·1276

while the ILl can provide useful information to forecasters, it is by no means
a foolproof tool for forecasting peaks or
troughs in business activity. We also
conclude that the ILl does not yet appear to be signaling the approach of a
business peak.
•

The ILl: A Historical

Perspective

gregate economic activity have been
studied by researchers at the National
Bureau of Economic Research
(NBER), a private, nonprofit economic
research organization, since the 1930s.
Of the hundreds of economic series examined, some were found to precede or
lead, others to coincide, and still others
to follow or lag turning points in the
direction of overall economic activity,
as determined by the NBER.
The first notable use of leading indicators occurred in 1937, when Secretary
of the Treasury Henry Morgenthau
asked the NBER to compile a list of economic series that would best indicate
when the 1937-38 recession would end.2
In October 1961, the Bureau of Economic Analysis (BEA) of the U.S. Department of Commerce began to publish leading, coincident, and lagging
indicators in its then-new monthly publication, Business Conditions Digest.

The financial news media frequently
point to the movement of the Composite Index of Leading Indicators
(ILl) as proof of impending growth
or contraction in economic activity. A
closer look indicates, however, that
while the ILl can provide a great deal
of useful information, its value as a
forecasting tool is limited. Its usefulness increases when it is used in cornbination with other indexes.

From among the leading indicators
originally selected by the NBER, the
BEA selected the 12 best on the basis of
six criteria: economic significance;
statistical adequacy; timing at troughs
and peaks of aggregate economic activity; conformity to past business expansions and contractions; smoothness;
and frequency and timeliness. The 12 individual series were combined into a
composite index, which was first published in November 1968.
The BEA has revised the list of components of the composite index from time
to time, as the usefulness of some
series waned, and new or improved
series became available. The composition of the ILl was last changed in January 1989 and now includes 11 series.3

• The ILl as a Forecaster of
Business Peaks
There have been eight business peaks
since World War II. The ILl has peaked
prior to each one by periods ranging
from two to 20 months, making it a
potentially good tool for forecasting
business peaks. However, closer
analysis reveals that the ILl has serious
shortcomings as a forecasting tool.
The usefulness of the ILl as a
forecaster of business peaks depends
on avoiding false signals of peaks,
avoiding the failure to forecast peaks,
and providing ample but not excessive
waming of peaks. The ILl exhibits
some substantial failings when judged
by these criteria.
First, the ILl has given from three to
seven false signals of peaks, depending
on the number of consecutive months of
decline that is taken to be a signal of a
peak. If a seven-month decline is taken
as a signal, there were false signals in
1951, 1966, and 1984, when the ILl
declined for seven, nine, and seven consecutive months, respectively. If a fivemonth decline is considered to be a signal, the ILl also gave a false signal at
the start of 1988. The financial news
media sometimes say that a three-month
decline in the index usually indicates a
recession. If so, there was also a false
signal in 1962, and perhaps also at the
beginning of 1979 and the beginning of
1981.
Second, it is stretching things to say that
consecutive monthly declines in the ILl
forecast the July 1981 peak, because the
ILl had declined from its peak for only
two months before the business peak,
according to the BEA. A less restrictive
statement of that episode is that the ILl
declined in five of the eight months
before the business peak, which could
be taken as a reasonable signal.
Two of the leads reckoned by the BEA
include periods when the decline in the
ILl was interrupted by rises of two or
more months. The 20-month lead
before the August 1957 peak includes
three gaps of two months each. Excluding those gaps, there was an uninter-

rupted lead of nine months. The 15-

periods between the time the ILl sig-

month lead before the January 1980
peak includes a gap of two months, and
three gaps of one month each. Excluding those gaps, there was no lead.

nals an impending business peak and
the peak itself. The rule selected here
on the basis of minimizing forecast errors provides leads ranging from one to
14 months.

A forecasting rule using the ILl must
specify how many months of decline
constitutes a signal of a business peak,
while also taking into account interruptions in the downward movement of the
ILL An example of such a rule is that
the ILl is forecasting a recession if the
index falls in four out of six months.

This analysis indicates that any statement suggesting that the ILl always
turns down before the economy turns
down would be true, but misleading. It
is misleading because it implies that
the ILl can be used toforecast business peaks, when in fact the index is

That rule would have yielded false signals in 1951, 1966, 1984, and 1987, and
would have failed to forecast the peak
of July 1981.

quite unreliable for that purpose. Any
forecasting rule based on the III will
give either some false signals of business peaks, or will fail to forecast

Any forecasting rule based on the ILl
must accept some trade-off between
false signals and failures to forecast a
business peak. We tested several rules

peaks, or will do both. Moreover, the
correct forecasts exhibit highly variable
lead times.
The most accurate rule identified in

based on the experience of the last 42
years. The least restrictive rule that
would not have given any false signals
is that the ILl must fall for 10 consecu-

this analysis can be said to offer the following guidance to an analyst using the
ILl to help forecast the next business
peak: If the ILl has declined in four of

tive months. However, with that rule,
the ILl would have failed to anticipate
all of the peaks. The least lenient rule

the last seven months, the chances are
two out of three that a business peak
will occur sometime within the next 14
months.4

that would have forecast all of the
peaks is that the ILl must fall in four
out of seven months. With that rule, the
ILl would have given four false signals.
Considering failure to forecast a peak,

The foregoing analysis uses a historical
series for the latest composition of the
ILl, calculated after all of the revisions
of the basic data had been made. How-

and a false signal of a peak, as equally
egregious errors, two rules appear to
minimize the total errors. One is that a
recession is forecast if the ILl falls in

ever, when actually using the ILl to
forecast, an analyst first receives an unrevised figure for the ILl for a particular month. That figure could then

any four of seven months. The other
rule uses a test of falling in any four of
eight months. Both rules would have
made four errors, giving four false signals. However, neither would have
failed to forecast any business peak.
The first of the two rules seems
preferable because it provides a longer

be revised in each of the next five
months, which adds to the uncertainty
confronting the analyst. Consequently,
the probability of false signals and

average waming time and has a shorter
average duration of false signals.
Another shortcoming of the ILl is the
variability of its warnings of business
peaks. As reckoned by the BEA, the
ILl peaks from two to 20 months
before a business peak. What is more
relevant is the range in length of the

failures to forecast might be even
greater than is implied by this analysis.
Finally, this analysis assumes that the
ILl is available during the month for
which it is issued. In fact, it is published about four or five weeks following the end of the month. Therefore,
the ILl gives forecasters that much less
lead time.

• The ILl as a Forecaster
of Business Troughs
The ILl has tumed upward prior to
each of the last eight business troughs.
By the reckoning of the BEA, the
spans between the low points of the ILl
and the troughs of the business cycles
ranged from one month to 10 months.
For six of the eight troughs, the span
was from one month to six months.
Unfortunately, it is not uncommon for
the ILl to give a false signal of the approach of a trough. In six of the eight
contractions, the ILl gave a false signal
by rising and then declining again. For
example, in the 1960-1961 recession,
the ILl rose in five consecutive months
following the business peak but then
flattened for three months, which at the
time could have been interpreted as
canceling the signal that a trough was
at hand. Again, during the 1981-1982
recession, the ILl advanced in four
months of a six-month period and then
declined, before rising again to signal
an impending end to the recession.
A rule that a one-month rise in the ILl
forecasts a trough is clearly too lenient
because it would have given false signals in six of the eight recessions. A
rule with a three-consecutive-month
criterion is probably too strict, because
while it would have given no false signals, it would have failed to anticipate
the troughs in 1958, 1970, 1975, and
1980.
The rule that seems to balance best the
risks of the two types of error states
that two consecutive rises in the ILl
forecasts a business trough. That rule
would have given two false signals and
would have failed to forecast the
troughs of 1970 and 1975. The
forecasts of impending troughs given
by the two-consecutive-month rule
would have preceded the troughs by
periods ranging from one month to
nine months, and averaging 4.7 months.
Thus, the foregoing analysis can be
said to offer the following guidance to
an analyst using the III to help forecast
the end of a recession: If the ILl has
risen for two consecutive months, the

experience of the last 42 years suggests

Using that rule, the ILl would have

that the chances are three to one that a
trough will be reached in the next one
to nine months. But if the ILl has not
risen in both of the last two months, the

forecast all eight peaks. The average
lead time falls slightly to 4.4 months
from the 5.6 months obtained by using
the ILl alone. However, the range of
lead times is narrowed to one to nine

analyst must nevertheless keep in mind
that the rule has failed to forecast two
of the last eight business troughs. If the
ILl has risen for three consecutive

months from one to 14 months, thereby
reducing the uncertainty about when
the peak will arrive.

months, the experience of the last 42
years suggests that a trough will be
reached within the next six months or
has already occurred within the last
two months.

The rule also would be somewhat better
in that it would have given three false
signals instead of the four that are given
by forecasting with the ILl alone. The

• Forecasting Business Peaks
Using Other Indexes in Combination
With the ILl

false signal of a business peak following the stock market collapse in October 1987 would have been avoided. The
rule does not reduce the nine-month

In addition to the ILl, three other com-

average duration of the false signals.

posite indexes of leading indicators are
published in Business Conditions
Digest. They are the ratio of the Composite Index of Coincident Indicators

• Forecasting Business Troughs
Using Other Indexes
We used two rules to examine the

to the Composite Index of Lagging Indicators (the Ratio), the Long-Leading
Index, and the Short-Leading Index.5

forecasting accuracy for business
troughs of both the ILl and the alternative indexes. Because recessions have
been much shorter than expansions, the

Rules based on these measures have the

forecasting rules can be much simpler
than those used to forecast peaks. The
two rules tested are (1) a one-month
rise in a measure forecasts a trough,

same shortcomings as the ILl: a rule
lenient enough to predict all business
peaks will also give false signals, and a
rule strict enough to avoid false signals
will fail to forecast most of the peaks. If
four declines in the last seven consecutive months is considered a forecast of a
business peak, the three alternative in-

The Long-Leading Index with the twomonth rule would have giver) the best

dexes would have forecast all of the last
eight peaks, with one exception: the
Short-Leading Index would have failed
to forecast the July 1953 peak. How-

performance during the last eight recessions in terms of minimizing forecast
errors. It forecast all eight troughs and
gave only one false signal. Its range of

ever, the Ratio, the Long-Leading
Index, and the Short-Leading Index
would have given false signals nine,
five, and eight times, respectively.

lead times, one to nine months, is
wider than some of the other choices,
but that greater variability seems a

Despite the shortcomings of the alternative indexes, analysts Can improve their
forecasts to some extent by using the altemative indexes in conjunction with
the ILl. A rule for doing so is to say that
the ILl is forecasting a peak only when

and (2) rises in two consecutive
months forecast a trough.

small price to pay for its much greater
forecasting reliability.
Thus, the Long-Leading Index appears
to be superior to the ILl, which gave
two false signals and failed to forecast

two conditions are met: (I) the ILl has
declined in four of the last seven

two troughs with the
and gave seven false
one-month rule. The
Short-Leading Index

two-month rule,
signals with the
Ratio and the
are much less reli-

months, and (2) at least two of the alternative indexes are also signaling a peak
by the sarne four-out-of-seven test.

able measures for forecasting troughs,
making 10 and nine errors, respectively,
with the one-month rule, and five and