View original document

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

December 1, 1990

6GONOMIG
COMMeNTORY
Federal Reserve Bank of Cleveland

How Credible are Capital Spending
Surveys as Forecasts?
by Gerald H. Anderson and
John J. Erceg

V^apital spending is one of the most
volatile sectors of the U.S. economy,
typically accounting for one of the largest
shares of the variation in GNP. Business
analysts, who often use capital spending
surveys to generate fixed-investment
forecasts, commonly rely on the one
prepared by the U.S. Department of Commerce. That survey's July-August data
show that businesses plan to increase
their capital spending in 1990 by 5.4 percent from last year's level, implying that
this sector of the economy will likely contribute only half as much to the growth
rate of GNP this year as it did in 1989.'
How reliable are these results? This
Economic Commentary evaluates not
only the accuracy of the Commerce
Department's capital spending survey,
but its usefulness in forecasting business investment.
• The Commerce Department
Surveys: An Overview
In addition to being one of the most
widely used capital spending surveys,
the Commerce Department survey is
also one of the oldest.2 Initiated in
1947, it reports actual and anticipated
U.S. plant and equipment (P&E) expenditures five times for a given year. Surveys are taken quarterly, and about
5,000 businesses generally respond.
The survey's coverage (that is, the

ISSN 0428-1276

proportion of an industry's output represented by the firms surveyed) varies
widely, ranging from 99 percent for nonferrous metals manufacturing to 12 percent for personal and business services.
The initial survey for a particular year,
taken in October and November and
published in December of the preceding
year, includes planned expenditures for
the first and second quarters and also
for the full year. Second-half projections are easily derived by subtracting
first-half spending from the annual data.
The figures are then revised by a survey
taken in January and February and published in April. However, it is not until
results are in from a third survey, taken
in April and May and published in June,
that spending predictions for each of the
four quarters are published. Surveys
taken in July and August (published in
September) and in October and November (published in December) update
both actual and anticipated expenditures
for the year.
Actual and planned expenditures are
reported to the Commerce Department
in nominal dollars. The planned figure
is only the survey respondent's best
estimate of what course a firm's future
P&E spending will take. Actual spending, however, can deviate from the initial estimates for various reasons. For

Business analysts should be aware
that the survey of capital spending
plans published by the U.S. Department of Commerce has several limitations as a forecast of quarterly and
annual fixed investment. Although
the annual expectations are relatively
reliable, the quarterly spending
projections are often less accurate
than other inexpensive and equally
accessible forecasts. This Economic
Commentary compares the reliability
of the Commerce Department survey
with that of several alternative forecasts, and suggests some underlying
reasons for the discrepancies between
the survey's projections and actual
capital expenditures.

example, a firm's board of directors
might appropriate a different amount or
alter its previous appropriation because
of a change in business conditions or in
capital stock needs. Variations can also
result from a company's inability to
arrange financing, or because the terms
of financing are not as expected. Furthermore, the timing of expenditures is
dependent upon the ability of the capital goods vendor to deliver on schedule.

TABLE 1 ACTUAL VS. EXPECTED ANNUAL
P&E SPENDING

P&E Spending"
Year

Actual
Data

Initial
Survey

Absolute Error
Initial
Survey

Naive

ARIMA

Forecast

Forecast

5.5

9.3

3.8

6.0

6.2

1971

1.9

1.4

0.5

3.6

1.5

1972

8.9

9.1

0.2

7.0

8.2

1973
In addition to publishing the data in nominal dollars, the Commerce Department
also reports real anticipated expenditures,
which it calculates using implicit price
deflators extrapolated from actual price
changes over the latest four-quarter period. Survey data are also adjusted for
reporting biases; that is, for consistent
differences between some companies'
reports of planned and actual spending.
Quarterly levels are published in seasonally adjusted annual rates.

1970

12.8

12.9

0.1

3.9

1.8

1974

12.7

12.0

0.7

0.1

13.9

1975

0.3

4.6

4.3

12.4

4.3

The average absolute difference between the percent change in spending
indicated by the initial survey and the
actual change was 2.9 percentage
points between 1970 and 1989 (table 1),
or nearly 40 percent of the 7.5 percent
average annual change in actual P&E
spending. Moreover, there is considerable variation around that 2.9 percent
average. In 10 of the 20 years between
1970 and 1989, the initial survey anticipated spending changes within 2.1 percentage points of the actual yearly
change, while in the remaining years
the difference ranged between 3 percentage points and 9 percentage points.
One way to evaluate the survey's
reliability is to compare its errors with
those of some inexpensive alternative

6.8

5.5

1.3

6.5

5.0

1977

12.7

11.3

1.4

5.9

3.5

1978

13.3

10.1

3.2

0.6

8.9

1979

17.0

11.2

5.8

3.7

11.4

1980

9.3

10.9

1.6

7.7

4.9

1981

8.7

10.8

2.1

0.6

3.5

1982

-1.6

7.4

9.0

10.3

8.8

1983
• Anticipating Annual Changes
How accurate are the initial surveys,
and how have the succeeding four surveys fared in terms of anticipating the
magnitude of spending changes for the
coming year? Since 1970, the initial
annual survey has had a mixed record.
However, accuracy tends to improve
substantially by the time the third survey is taken in April and May.

1976

-4.8

-1.3

3.5

3.2

11.3

1984

16.3

9.9

6.4

21.1

5.0

1985

9.2

8.2

1.0

7.1

2.3

1986

-2.0

2.4

4.4

11.2

8.7

1987

2.4

0.9

1.5

4.4

6.7

1988

10.3

7.3

3.0

8.1

1.0

1989

10.4

6.0

4.4

0.1

0.2

2.9

6.2

5.9

Average absolute
error (1970-89)

a. Percent change from previous year.
b. In percentage points.
NOTE: Underlying data are in nominal dollars.
SOURCES: U.S. Department of Commerce, and authors' calculations.

forecasts. Two such alternatives rely
solely on past capital spending data: the
"naive" forecast, which assumes that
each year will mirror the preceding one,
and the Autoregressive Integrated
Moving Average (ARIMA) forecast,
which is generated by a process involving all past data in the series. Both
approaches are relatively inexpensive to
develop because they rely on readily

available information and do not require
a theoretical model of investment determinants. Of the three forecasts, errors in
the capital spending survey are the smallest. For the 1970-89 period, the average
absolute error of the initial capital spending surveys was 2.9 percentage points
(as previously noted), while the respective figures for the naive and the ARIMA
forecasts were 6.2 and 5.9 percentage
points.

TABLE 2 ANNUAL P&E SPENDING
Errora of Survey Taken in:
Oct.Nov.

Jan.March

AprilMay

1970

3.8

4.3

1971

-0.5

2.4

1972

0.2

1.6

For Spending in:

JulyAug.

Oct.Nov.

2.3

1.1

1.1

0.8

0.3

0.3

1.4

0.8

0.1

1973

0.1

1.0

0.4

0.4

0.4

1974

-0.7

0.3

-0.5

-0.2

-0.5

1975

4.3

3.0

1.3

0.7

0.7

1976

-1.3

-0.3

0.5

0.6

0.7

1977

-1.4

-1.0

-0.4

0.6

1.0

1978

-3.2

-2.4

-2.1

-1.0

-0.6

1979

-5.8

-5.7

-4.3

-3.8

-2.3

1980

1.6

2.7

0.5

-0.7

-0.5

1981

2.1

1.5

-0.3

0.1

0.4

1982

9.0

8.9

3.8

2.3

1.1

1983

3.5

3.1

1.4

1.7

0.6

1984

-6.4

-2.3

-1.5

-2.0

-2.0

1985

-1.0

-0.5

0.0

-0.9

-0.8

1986

4.4

4.3

2.2

0.1

0.3

1987

-1.5

0.6

0.1

0.1

0.5

1988

-3.0

-1.5

0.4

0.3

0.1

1989

-4.4

-1.3

-0.5

-0.4

-0.1

Average Absolute Error
1970-89

2.9

2.4

1.2

0.9

0.7

1970-79

2.1

2.2

1.4

1.0

0.8

1980-89

3.7

2.7

1.1

0.9

The second survey (January-March)
improves only slightly upon the first
(reducing the average error to 2.4 percentage points), even though respondents generally alter their spending
projections to reflect changing perceptions of the investment outlook (table
2). Succeeding surveys show considerable improvement, however. Measured
in percentage points, the error is reduced
to 1.2 by the third survey (April-May),
0.9 by the fourth survey (July-August),
and 0.7 by the fifth survey (OctoberNovember).
In using the Commerce Department
data, business analysts should take
into account that the results have tended
to overestimate the strength of capital
spending during recessions and to underestimate its strength during expansions. In the recession years of 1970
and 1982, the initial surveys overshot
actual annual spending by 3.8 percentage points and 9.0 percentage points,
respectively (table 2), while they underestimated the strength of capital spending in four of the five years of the 197579 expansion and in five of the first
seven years of the expansion that began
in 1982. Only during the 1971-73 upturn did the survey avoid substantial
underestimation. For those three years,
the average error was less than 4 percent of the average annual change in
P&E spending.

0.6

a. Planned percent increase less actual percent increase.
NOTE: Underlying data are in nominal dollars.
SOURCES: U.S. Department of Commerce, and authors' calculations.

TABLE 3 SUMMARY OF FORECAST ERRORS
(Average absolute percentage points)
Results of:
P&E
Survey

Naive
Forecast

ARIMA
Forecast

Annual change, 1970-89
P&E spending
NRFI

2.9
4.3

6.2
7.7

5.9
6.7a

Quarterly change, 1979:IQ-1989:IVQ
P&E spending
NRFI

2.1
2.2

2.1
2.4

1.7
2.2

Forecast of:

a. Twenty-year moving average.
SOURCE: Authors' calculations.

One final point must be made about the
initial annual capital spending surveys:
The differences between actual and
anticipated changes appear to be increasing. In the 1970s, the average
annual difference was 2.1 percentage
points. In the 1980s, that figure rose to
3.7 percentage points, even though the
average of the actual changes was
greater in the earlier decade.
How can the larger errors of the 1980s
be explained? A major source of forecast errors may be that after a survey is
made, changes occur in the perceived
after-tax profitability of potential investments. In both the 1970s and the 1980s,
oil price shocks, recessions, episodes of
unanticipated inflation and disinflation,
and changes in tax laws all affected the

FIGURE 1 FORECASTS OF NONRESIDENTIAL FIXED INVESTMENT
(Absolute errors)
Percentage points
I

[ Initial survey
Naive forecast
Moving average forecast

1970

1972

1974

1976

1978

1980

1982

1984

1986

1988

SOURCE: Authors' calculations.

investment outlook. However, unlike
the tax changes of the 1970s, which
were designed to stimulate overall
demand, many of the tax laws passed in
the 1980s were designed as fixedinvestment incentives. (Incentives were
increased by 1981 legislation and then
reduced by 1982 and 1986 legislation.)
These differences may explain why the
annual changes in actual investment
had a greater standard deviation during
the 1980s, even though the average annual change was smaller then. During
the processes of debating, enacting,
interpreting, and discovering loopholes
in such tax changes, views about the
profitability of new capital spending
can ebb and flow, causing substantial
modifications in plans and, therefore,
larger forecast errors.
Another major problem in economic
forecasting is anticipating turning
points; that is, the dates when an economic series reaches its peak and
trough. During the last 20 years, capital
spending changed direction four times
(1982, 1984, 1986, and 1987). The survey correctly anticipated the upturns of
1984 and 1987, but failed to predict the
downturns of 1982 and 1986 (table 1).

• Anticipating Quarterly Changes
A few characteristics of the quarterly surveys should be noted. First, the initial
surveys were front-loaded during most
of the 1980s; that is, they anticipated
level or higher capital spending for the
first half of the year, followed by implied
decreases during the second half. However, in every case, actual spending
during the last six months turned out to
be greater than first-half spending. This
characteristic appears to be a reporting
bias for which the Commerce Department has not made a correction.
Since 1985, the absolute difference
between the initial survey of anticipated
first-quarter capital spending and actual
spending has averaged 3.5 percentage
points. (The initial survey for 1986:IQ
overstated the final figure by 10.4 percentage points.) As might be expected,
each succeeding survey becomes more
accurate because, as time passes, firms
gain additional information about their
sales prospects, their profits, and their
need to replace or add to capital stock.
However, the second survey, taken
during the first quarter, still has an average absolute error of 2.6 percentage
points. Even the third survey, taken in

April and May, has an average absolute
error of 0.6 percentage point. Similar
patterns of inaccuracy can be found in
the anticipations for other quarte
tiers.
In contrast to the surveys of expected
annual changes in P&E spending, the
quarterly surveys do not fare quite as
well when compared to the naive and
ARIMA forecasts. Over the last 44
quarters, forecasts of the growth rate of
P&E expenditures for the coming quarter, based on the latest survey prior to the
quarter in question, had an average absolute error of 2.1 percentage points. Naive
forecasts also had an average absolute
error of 2.1 percentage points, while the
corresponding figure for the ARIMA
forecasts was 1.7 percentage points.
Thus, taken alone, the Commerce
Department survey of quarterly P&E
spending changes seems to be no better
than the naive forecasts and is less reliable than the ARIMA forecasts.
• Forecasting NRFI
Analysts often use the latest P&E spending survey to forecast the business
fixed-investment sector of GNP. Nonresidential fixed investment (NRFI) consists of producers' durable equipment

and business structures. Businesses
make fixed investments as they seek to
bring their capital stock in line with
some desired level. The amount of
investment needed to make the adjustment varies gTeatly from one year to
the next, which in turn contributes to
wide swings in NRFI and in overall
economic output (even though fixed investment accounts for only about 11
percent to 12 percent of GNP).
P&E spending and NRFI differ in at
least two major respects, and therefore
exhibit different quarterly and annual
percentage changes. First, NRFI is a
broader series, including industries such
as farming, real estate, and professional
services. Second, NRFI is based on
construction put in place (structures)
and manufacturers' shipments of equipment (producers' durable equipment),
whereas P&E data are based on expenditures, which generally occur later.
Therefore, differences exist in the timing of the two broad series. In addition,
NRFI is organized by type of investment, while P&E survey data are organized by industry.

Despite their differences, the two investment series track reasonably well
over the long term. Nevertheless, the
annual percent changes in actual
NRFI and P&E spending have differed
by an average of 2.1 percentage points
over the last two decades. Moreover,
the absolute difference in the quarterly
percent changes has averaged 1.4 percentage points over the past 11 years.
Because of the differences between 1)
the initial and the actual yearly P&E
spending figures and 2) actual P&E
spending and actual NRFI, the initial
P&E survey is not a very accurate
forecast of the forthcoming annual
change in NRFI: The absolute error has
averaged 4.3 percentage points over the
last 20 years. Nevertheless, this is significantly less than the 7.7 percentage
points error of a naive forecast and the
6.7 percentage points error of a moving
average forecast that expects growth in
a given year to equal the average of the
preceding 20 years (see figure 1).
The same cannot be said for the quarterly surveys of P&E spending as predictors of quarterly changes in NRFI.
Those surveys are little better than the
naive forecasts and no better than the
ARIMA forecasts. Over the last 44
quarters, predictions of the coming
quarter's NRFI growth rate based on
the latest P&E spending survey had an
average absolute error of 2.2 percentage points, while the respective figures
for the naive and ARIMA forecasts
were 2.4 and 2.2 percentage points
(table 3).

• Conclusion
Although the Commerce Department
survey is often used as a forecast of
capital spending, business analysts
should be aware that its record for
accuracy is mixed.
The annual survey is a more reliable
predictor of P&E spending and NRFI
than the naive and ARIMA forecasts.
However, the survey's errors were
greater in the 1980s than in the 1970s,
and two of the four changes in the
direction of annual capital spending
went undetected.
Taken alone, the quarterly survey appears to be of little value as a forecast,
not only because the differences between anticipated and actual spending
have been large (at least until the fourth
survey for a quarter), but because the
reports have been front-loaded in recent
years. Quarterly forecasts of equal or
greater accuracy for both P&E spending
and NRFI can be obtained with naive
and ARIMA data.

•

Footnotes

1. Survey respondents report anticipated
spending in current dollars only. The U.S.
Department of Commerce estimates constantdollar projections using these current-dollar
figures and recent rates of price increase,
then publishes capital spending anticipations
in both current and constant dollars. Results
reported here are based on the survey's
current-dollar calculations and do not necessarily apply to the constant-dollar projections.
2. Commerce Department survey data are
published in Plant and Equipment Expenditures and Plans. Two other surveys are the
McGraw-Hill Annual Survey of Preliminary
Plans for New Plants and Equipment, and
the Conference Board Survey of Newly Approved Capital Appropriations. Responsibility for the Commerce Department survey
was transferred in 1988 from the Bureau of
Economic Analysis to the Bureau of the Census (both of which are in the U.S. Department of Commerce).
3. Until recently, the sample included about
12,000 businesses, with another 9,000 included in industries surveyed only annually,
such as real estate, professional services, and
forestry, fisheries, and agricultural services.

5. The naive forecasts of annual changes assume that capital spending will change in the
coming year by the same percentage as in the
year just ended. The naive forecasts of quarterly changes assume that capital spending will
change in the coming quarter by the same percentage as in the period two quarters earlier,
which would be the most recent quarter that
has a growth rate known with reasonable certainty. ARBVIA is a method of generating forecasts for a time series from the historical data
for that series.
6. Comparisons of the root mean square errors of the forecasts lead to the same conclusions, here and throughout this study, as
those reached using average absolute errors.

Gerald H. Anderson is an economic advisor
and John J. Erceg is an assistant vice president and economist at the Federal Reserve
Bank of Cleveland. The authors would like to
thank Michael Bryan and Randall Ebertsfor
helpful comments on earlier drafts of this article, Theodore Bernard for valuable research assistance, and Michael Bagshawfor
statistical consulting.
The views stated herein are those of the
authors and not necessarily those of the
Federal Resene Bank of Cleveland or of the
Board of Governors of the Federal Reserve
System.

7. Although some of these industries are surveyed annually, their reports have been excluded from most of the P&E spending
figures. In recent years, P&E expenditures
have been about 85 percent of NRFI, but that
figure increases to about 95 percent if the
annually surveyed industries are included.
8. Actual changes in NRFI appear to be random about their average, and no forecast
could be generated using the ARIMA process.

4. Price-adjusted, or real, data have been
reported for each industry, but as of June
1990, the only real figure included is the allindustries total.

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.

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