The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
R E G IO N A L E C O N O M IC IS S U E S
W o r k in g P a p e r S e r ie s
A W i n d o w of Opportunity O p e n s
for Regional E c o n o m i c Analysis:
B E A Releases Gross State Product Data
Alenka S.Giese
FE D E R A L R E SE R V E
O F C H IC A G O
B A N K
WP- 1989/3
Ta bl e of Cont en ts
I.
II.
List of Tables and Graphs
Introduction
III.
The Deficiencies of Traditional Measures of Regional Economic Activity
IV.
The Definition of G SP
V.
VI.
VII.
VIII.
IX.
B E A ’s Methodology
The Strengths and Weaknesses of B E A ’s Methodology and a
Comparison to the K-J Methodology
Comparing G SP to Other Estimates of Output
Conclusion
Footnotes
FRB CH ICAGO W orking Paper
February 1989, W P -1 989-3
1
List of Tables a n d G r a p h s
Graphs
1.
Comparison between New England GSP and K-J
Estimates—Manufacturing
2.
Comparison between New England GSP and K-J Estimates— Services
3.
Comparison between New England GSP and K-J Estimates—FIR E
4.
Comparison between U.S. Manufacturing GSP and Value-Added
5.
Comparison between U.S. Manufacturing GSP and Purchased Services
Table
1.
Trends in Manufacturing GSP, Value-Added, and Purchased Services
FRB CH ICAGO W orking Paper
February 1989, W P -1989-3
2
A
W in d o w
E c o n o m ic
S ta te
o f O p p o r t u n it y
A n a ly s t s :
P ro d u ct
B E A
O p e n s fo r R e g io n a l
R e le a s e s
G ro ss
D a ta
Alenka S. Giese*
Introduction
One o f the formidable challenges facing regional analysts has been over
coming data limitations. A paucity of data has often impeded rigorous
economic analysis or resulted in findings that are burdened with caveats
about data problems at the regional level. O f major concern has been the
difficulty in accurately gauging regional economic activity using available
data on output. With the release of the Bureau of Economic Analysis’
(BEA) gross state product (GSP) estimates, the problems with measuring
regional output have been alleviated significantly. These estimates have
several major advantages in terms of coverage and accuracy over other re
gional data such as value-added and employment.
The purpose of this paper is to evaluate the GSP data. The paper is divided
into five sections. The first section examines the problems associated with
using traditional regional measures of economic activity. The second sec
tion provides a description o f the G SP data and their components. The
third section describes B E A ’s methodology, and the fourth section discusses
the methodology’s strengths and weaknesses and compares it to the
Kendrick-Jay cox (K-J) methodology.1 The final section presents the results
of comparative analyses of the G SP data and other measures of output.
T h e deficiencies of traditional measures of regional
e c o n o m i c activity
The important contribution of B E A ’s GSP data is that they provide a more
accurate and comprehensive measure of regional output than other regional
data. The regional data that have traditionally been used to measure eco
nomic activity (e.g., value-added, personal income, and employment) have
proven to be either insufficient or misleading.
*Alenka S. Giese is an associate economist at the Federal Reserve Bank of Chicago. The au
thor thanks Robert Schnorbus, W
illiam Testa, and Ted (Edward A.) Trott for helpful com
m
ents and suggestions.
FRB CH ICAGO W orking Paper
February 1989 , W P -1989-3
3
Employment data are the most commonly used data in regional economic
analysis, which is understandable given that they are available at a fine level
o f industry and geographic detail. They are not, however, an accurate
proxy for economic activity, especially for goods-producing industries (i.e.,
agriculture, mining, construction, and manufacturing). For example, in the
Seventh District (comprised of Illinois, Indiana, Iowa, Michigan, and
Wisconsin), manufacturing’s share of total employment has been declining
gradually, though at an increasing rate, since World War II. In contrast,
manufacturing’s share of total output has been relatively stable. Thus, it
would be misleading to interprete the relative decline in manufacturing
employment as a decline in activity.2 Most of the shift in employment from
manufacturing to services does not reflect decline but rather reflects differ
ences in labor productivity growth and the substitution of capital for labor
in manufacturing.
The problem with Census value-added data is not so much one of accuracy
but rather one of coverage.
Census value-added data cover only the
goods-producing sectors which account for just under a third of national
output. Even though manufacturing is regarded as the engine of economic
growth, manufacturing output paints only a partial picture of regional
output and may not accurately reflect overall economic performance. Some
regional analysts believe that tertiary sectors (e.g., transportation and ser
vices) have been becoming less dependent upon manufacturing activity.
Thus, their performance has become less to tied to manufacturing output.
Evidence supporting this view is seen in the diverging trends in manufac
turing and services GSP.
The third regional data series, B E A ’s regional personal income, offers data
on all two-digit SIC (Standard Industrial Classification) industries but is
not comprehensive because it omits indirect business taxes and capital
charges.3 When only income data are used to gauge economic activity, sig
nificant distortions can occur. For example, measurement error would oc
cur for industries in which earnings account for a relatively small share of
total income (e.g., real estate, oil and gas extraction, and petroleum refin
ing). This problem would be particularly pronounced in states where these
industries are dominant such as Texas.
T h e definition of G S P
In order to understand why GSP data offer a better measure of regional
economic activity than the three data series examined above, it is necessary
to, first, define GSP and then discuss the methodology used to estimate it.
Gross state product represents a state's contribution to gross domestic
FRB CH ICAGO Working Paper
February 1989 , W P -1 989-3
4
product (G DP) and is the most comprehensive measure of a state’s output.
It can be formally defined in alternative ways. It can be viewed as the gross
market value of the goods and services produced by a state’s labor, capital,
and land net o f purchases o f intermediate products (i.e., materials) and
services. Alternatively, it can be defined as the sum of factor and nonfactor
charges incurred in producing these goods and services net of purchases of
intermediate products and services.
It is important to note that there are fundamental and significant differ
ences between the concept of G SP and that of Census value-added. Firstly,
G SP data exclude intermediate purchases of both materials and services
whereas Census value-added data only nets out intermediate purchases of
materials. Secondly, the Census method does not estimate value-added di
rectly but derives it as a residual of shipments less estimates of purchased
materials. In contrast to Census methodology, B E A ’s methodology builds
up G SP by component.
The G SP data are disaggregated into four components, corresponding to
factor and nonfactor charges:
1. Compensation of employees (wages and salaries and supplements to
them).
2. Farm and nonfarm proprietors’ income with inventory valuation ad
justment and capital consumption allowances.
3. Indirect business taxes (IBT) and nontax liability.
4. Capital charges, primarily corporate profits with inventory valuation
adjustments and capital consumption adjustments, capital consump
tion allowances with capital consumption adjustments, and
net interest.
The G SP data are annual and begin in 1963. They are available at the one
and two-digit SIC code level for a total of 61 industries. For manufactur
ing, aggregates of durable and nondurable goods are available, and the
transportation equipment industry is disaggregated into SIC 371 and SIC
37 less 371.
The G SP data are available in current and constant (1982) dollars. To
calculate the constant dollar series, the current dollar series is deflated using
national industry implicit price deflators (discussed in detail below). Both
series are useful depending on the type of analysis to be undertaken. The
current dollars series is useful in measuring relative regional strengths in
factors o f production and regional effects of changes in relative output
prices. The constant dollar series provides the necessary data to calculate
FRB CH ICAGO W orking Paper
February 1989 , W P -1989-3
5
growth rates of total regional economic activity (i.e., output) and regional
factor productivities.
B E A ’ methodology
s
The methodology that B E A developed to estimate the four components of
G SP for the benchmark years (i.e., 1963, 1967, 1972, 1977, and 1982) is the
focus of this analysis.4 It is through this methodology that B E A has over
come many of the problems that have degraded the accuracy of previous
methodologies. The methodology differs across sectors, depending upon
raw data availability.
Similar methodology, however, exists for major
sectorial groupings such as goods-producing. Because of the differences in
methodology by sectorial groupings, the clearest way to describe B E A ’s
methodology is by G SP component and by industry.
The largest components, compensation and proprietors’ income, pose no
estimation problems because they can be derived from B E A ’s annual State
personal income series (see footnote 3). Although the IBT estimates are
not as easy to derive, the problems are surmountable. For the benchmark
years and for years after 1982, IBT are estimated using data on state taxes
by type from the census of governments and data on taxes by type and in
dustry from B E A ’s National Income and Wealth Division.
Estimation of capital charges, the fourth component, poses the most diffi
cult problems. For goods-producing industries, capital charges are derived
as a residual of total G SP less compensation, proprietors’ income, and IBT.
Total G SP is estimated from Census value-added by industry data that have
been adjusted to conform to the definition of value-added in the National
Income Product Account.
Three major adjustments are made to Census value-added data. Firstly,
adjustments are made to account more accurately for the geographical dis
tribution of central administrative office (CAO) costs (e.g., payroll and
nonpayroll value-added, and intermediate purchases). The need for this
adjustment arises because the Census Bureau does not estimate separately
C A O value-added and instead counts C A O costs as a component of valueadded of the establishments that the C A O s operate.5 In order to redistribute
C A O costs correctly across states, B E A undertakes a two step process.
Firstly, B E A estimates Census value-added ( V A CB) net of C A O costs, which
results in V A X . B E A then adds its own estimates of C A O costs based on
state location (C A O bea), resulting in V A 2 6
N OTE:
script.
All variables are by industry by state unless there is a us super
FRB CH ICAGO Working Paper
February 1989, W P -1989-3
6
1)
V A X = V A cb * (ratio of V A less C A O costs to total V A )
where ratio = [V A ™ - { P C A O 15 + N P C A O f EA +
4
IC A O & J W *"
2)
V A 2 = V A X + C A O bea
where C A O bea = P C A O -f N P C A O bea
N P C A O bea = N P C A O f EA x { P C A O IP C A O ™ )
Available from published sources:
VA™ = U.S. Value-added
P C A O ™ = U.S. C A O payroll
P C A O = State C A O payroll
Estimated by BEA:
N C A O f EA = U.S. C A O nonpayroll
I C A O f EA = U.S. intermediate purchases
The second adjustment that B E A makes is to estimate the purchased ser
vices component (.P S C B A and subtract this cost from V A 2 To approxi
E)
mate PSC, B E A multiplies V A 2 by the ratio of B E A national value-added
which excludes purchased services to the sum across all states of V A 2 which
includes purchased services. The result is V A ,.
3)
VA, = VA2 -
P S C BEA
where P S C BEA = V A 2 x ( V A f EA/ I , V A 2 across all states)
There are two contrasting views on the importance of excluding purchased
services. Although these two views bring up some important consider
ations, they do not come into play in the use of G SP data to examine
overall regional economic activity or to determine sectorial contribution to
total output. One viewpoint maintains that their exclusion in the GSP data
make these data superior to Census value-added data. The reasoning be
hind this view is that manufacturing GSP estimates strictly reflect manu
facturing output due to production as opposed to secondary activities such
as services that are purchased externally. In addition, because purchased
services output is distributed across the services industries, nonmanufac
turing G SP is not underestimated and total GSP is purged of double
counting. A contrary viewpoint states that inclusion of purchased services
allows one to gauge the total contribution of manufacturing activities to
regional output (i.e., all activity linked to manufacturing). Another ad
vantage o f this measure is that it may be more consistent over time, that is,
it is not distorted by the recent trend to outsource more services. As man
ufacturing firms purchase an increasing amount of services from external
FRB CH ICAGO W orking Paper
February 1989 , W P -1989-3
7
sources, manufacturing G SP growth is negatively impacted whereas valueadded growth is not.
The third adjustment of Census value-added is to account for differences
between B E A and Census industrial classifications of payrolls. This ad
justment involves two steps. First, V A 3 is multiplied by the ratio of BE A
compensation ( P A Y bea) to Census payrolls (.P A Y Cs), resulting in V A 4
Second, V A 4 is multiplied by the ratio of B E A ’s Gross Domestic Product (
G D P bea ) in the industry to the sum of V A 4 across all states, resulting in
VAS :
4)
VA4 =
V A 3 x C A Y bea / P A Y cb )
P
5)
VA5 =
V A 4 x ( G D P BEA/ 1 , V A 4 across all states)
For nongoods-producing industries, a less complex methodology is used to
estimate capital charges. Capital charges are estimated directly for the
benchmark years and indirectly for the non-benchmark years (excluding
real estate G SP which is estimated directly for all years; see footnote 4).
Benchmark year data are collected from several sources. For real estate,
data from the population and housing censuses and the U.S. Department
of Agriculture are used. For regulated distributive and service industries,
data are obtained from financial reports filed by firms with regulatory
agencies. For unregulated distributive and service industries, economic
census data on business receipts/sales and on compensation are used to
distribute capital charges. For government enterprises, capital charges are
estimated in two different ways depending upon level of government. A t
the federal level, surplus or deficit data specific to each enterprise are as
signed to capital charges. For state and local government enterprises, data
from the census of governments on current revenues and expenses by type
of enterprise (e.g., transit and water) are used to allocate the surplus or
deficit.
The
strengths a n d
weaknesses of B E A ’ m e t h o d o l o g y a n d
s
a
c o m p a r i s o n to the K - J m e t h o d o l o g y
Before B E A estimated G SP by component, regional output was usually
estimated using the K-J methodology, which is also known as the “blow
up” technique (see footnote 1). The K-J methodology uses the ratio of state
earnings to national earnings to distribute on a state by state basis national
data on the non-earnings components of GSP. For example, to estimate
IBT at the state level, the K-J methodology distributes national totals of
state and local taxes using state earnings shares. National data on the
non-earnings components come from B E A ’s G D P by industry series.
FRB CH ICAGO W orking Paper
February 1989, W P -1989-3
8
There are four major reasons why B E A ’s methodology is believed to be
superior to that of K-J. First, a significantly greater percent of GSP is di
rectly estimated by B E A than by K-J. According to BEA, about 96 percent
o f G D P (sum o f total G SP across all states) for the benchmark years is di
rectly estimated by B E A ’s methodology and allocated to a particular state
and industry whereas only 70 percent is directly estimated by the K-J
methodology.7 Second, through the use of value-added at the two-digit SIC
code level, the G SP estimates for the goods-producing industries reflect the
particular industry mix of each state.
The third advantage of B E A ’s methodology over the K-J methodology is
that B E A avoids making the K-J assumption that the relationship between
the earnings and non-earnings components by industry at the national level
is the same at the state level. The validity of this assumption has been
challenged because significantly different relationships between earnings
and non-earnings components at the state level have been found. For ex
ample, the distribution of IBT across states is often not at all related to the
distribution of payroll, particularly for states with low or no corporate in
come tax. Under the pure K-J methodology, the IBT estimates for a state
without corporate income tax (e.g., Florida) would be overestimated.
In analyzing their G SP estimates, B E A showed that the K-J assumption
does not hold in many cases and can result in regional distortions (see
footnote 4). Two of B E A ’s findings, in particular, support this conclusion.
First, B E A found that there is a positive relationship between fast output
growth and low per capita income, which is the result of location decisions
based on minimizing costs. Second, B E A ’s results revealed a positive re
lationship between fast output growth and high share of capital charges to
total GSP, which is the result of location decisions based on maximizing
profit levels. The implication is that for economically expanding states and
industries where earnings is a relatively small portion of G SP (e.g., real es
tate, oil and gas extraction, and petroleum refining), output will be under
estimated by the K-J methodology. For example, when the price of oil was
rising in the mid-1970s, which translated into increasing profits, the ratio
of G SP to G D P for states with an important oil industry (e.g., Texas) was
increasing at a faster rate than the ratio of these states’ earnings to national
earnings.
The fourth advantage offered by G SP data is that they expand the scope
of regional economic analysis that can be undertaken. In addition to a
more accurate measure of G SP by industry, the G SP estimates permit the
analysis of differential growth rates in output and total factor productivity
across states and industries attributable to factors of production outside of
earnings such as profit rates. For example, B E A in its article "
‘Gross State
FRB CH ICAGO W orking Paper
February 1989 , W P -1989-3
9
Product by Industry, 1963-1986,” presents an analysis of manufacturing
activity in the Great Lakes region and finds that the poor performance of
the Great Lakes’ manufacturing sector stems partially from the relative
decline in its manufacturing profit rates, as reflected by a fall in the capital
charges component.8
Even though B E A ’s methodology overcomes many of the problems that
undermine the accuracy of the K-J methodology, the G SP estimates have
some faults as well. Several problems that arise in both the K-J and B E A
methodology result from the deflators used to convert current dollars to
constant (1982) dollars. The deflator problems can be categorized into two
groupings. First, there are problems specific to the accuracy of the national
industry implicit price deflators used (IPDs at the two-digit SIC code level).
Second, there are problems specific to using national deflators at the re
gional level.
There are two concerns regarding the accuracy of the national deflators.
Firstly, the accuracy of IPDs for certain service industries has been chal
lenged (e.g., advertising and banking). Criticism has revolved primarily
around the relatively steep rise in these IPDs.9 The second issue concerns
the constant dollar GSP estimates for goods-producing industries. As ex
plained above, the current dollar GSP estimates are based on Census
value-added data and are deflated using IPDs. The potential problem lies
in the use of value-added data (derived by subtracting purchased materials
from shipments) and the fact that the purchased materials component in
cludes both imported and domestic inputs whose prices behave differently.
The question is whether the constant dollar GSP estimates are distorted by
not accounting for and separately deflating imported purchased materials
when value-added data are used. Unfortunately, it is impossible to deter
mine the extent to which this may be a problem in the GSP estimates. BE A
has, however, been able to approximate the effect of this problem on its
constant dollar gross product by industry series (see footnote 9). BEA has
estimated that if separate import price deflators had been used during the
1980-1985 rise in the dollar (i.e., when import prices rose relatively slowly),
the value of the constant dollar purchased materials component would have
increased and as a result, growth in constant dollar manufacturing gross
product would have decreased by half a percentage point or more per year.
The second category of deflator problems arise from the use of national
deflators at the regional level. Because regional industry mix and prices
often differ from the national figures, the use of national price deflators can
result in two types of estimation errors. Firstly, regional cost differentials
are not accounted for in the IPDs. This deficiency creates estimation errors
for industries whose costs tend to vary regionally (e.g., energy, con
struction, and real estate) and for states dominated by these industries.
FRB CH ICAGO W orking Paper
February 1989, W P -1989-3
10
Secondly, region specific deflator problem stems from the use of IPDs at
the two-digit SIC code level and differences in regional industry mix and
in the price behavior of three-digit SIC industries. By using two-digit SIC
IPDs, B E A assumes that the national three-digit SIC industry mix is con
stant across states. This is often not the case because many industries tend
to be regionally concentrated. Two striking examples of this are electronic
computing equipment (SIC 3573) and semiconductors (SIC 3674). A l
though B E A incorporates the drop over the past decade in computer and
semiconductor prices into the IPDs for SIC 35 and 36, it assumes by using
national IPDs that the three-digit SIC industry mix of SIC 35 and 36 is the
same across all states.1 This, of course, is not the case because SIC 3573
0
and 3674 are highly geographically concentrated. As a result, for states
with relatively large computer and semiconductor industries, the 1980s
two-digit IPDs for SIC 35 and 36 are overstated. Thus, the constant dollar
series of SIC 35 and 36 in these states to be undervalued. GSP underesti
mation is likely to occur for states such as California whose computer in
dustry accounts for 112 percent more of SIC 35 employment (1985) than
it does in the nation and whose semiconductor industry accounts for 33
percent more of SIC 36 employment, Massachusetts whose computer in
dustry accounts for 359 percent more of SIC 35 employment, and Texas
whose semiconductor industry accounts for 205 percent more of SIC 36
employment. Conversely, the GSP for these two-digit SIC industries are
overestimated in states without a large computer and/or semiconductor in
dustry but with large nonelectrical and electrical machinery industry (e.g.,
Illinois and Michigan).
In addition to the deflator problems, another problem with the G SP esti
mates occurs in the adjustment for cost of purchased services. Again, the
problem stems from using aggregated data at the two-digit SIC code level.
Some subindustries of the two-digit industries are geographically concen
trated and have different patterns of purchased service inputs than the in
dustry overall. The example given by B E A is the printing and publishing
industry (see footnote 4). The book and magazine publishing subindustry
tends to be geographically concentrated and purchases a significant amount
of services whereas the printing subindustry tends to be spatially dispersed
and purchases relatively fewer services. The result is that states with pub
lishing centers such as New York tend to have an overestimated capital
charges component for their printing and publishing industry.
A third problem arises from the use of Census value-added data to estimate
total G SP for the goods-producing sectors and stems from the way these
data are geographically allocated. Census value-added data are derived
from shipment data which are spatially allocated according to location of
final production. As long as the production and nonproduction activities
of an establishment are located in close proximity, there is no problem with
FRB CH ICAGO W orking Paper
February 1989, W P -1 989-3
11
this allocation procedure. Studies have shown, however, that headquarters
and R & D and tech-intensive activities tend to locate in northern urban
areas while production activities tend to locate in southern areas.1 Thus,
1
if these findings are accurate, implying that spatial separation of production
and nonproduction activities exists, value-added may not be correctly allo
cated across regions. For example, value-added for regions that tend to
specialize in nonproduction activities could be underestimated. B E A ac
counts for part of this problem when it adjusts Census value-added by re
distributing C A O costs by location as explained above. Its algorithm,
however, has a few deficiencies. Firstly, by using national C A O costs and
value-added data (see p. 6), it assumes that the ratio of the operating es
tablishment cost of maintaining C A O s to the operating establishment
value-added does not vary by state. The ratio for manufacturing overall,
however, does vary by state because it reflects each state’s industry mix (i.e.,
the ratio is an aggregate across the state’s two-digit SIC industries). A n
other problem with B E A ’s algorithm is that it neglects the nonproduction
activities of non-CAOs.
C o m p a r i n g G S P to other estimates of output
In order to do a quantitative analysis, the GSP data are compared with
other measures of output. Two different comparisons are made: the first
is between G SP estimates for New England and the Federal Reserve Bank
of Boston’s New England K-J estimates, and the second is between manu
facturing G SP and Census value-added. The objective of these compar
isons is to examine the trends of these estimates and check whether they are
displaying similar growth patterns. The purpose is not to gauge whether
or not the levels differ, which is to be expected given the different method
ologies applied.12 The results of these comparisons reveal the important
differences between GSP estimates and other measures used to track re
gional output. In addition, the findings point out that these other measures
can be misleading indicators of regional economic growth.
In order to capture any differences in trends across regions, the value-added
comparisons were conducted for five B E A regions: New England, MidAtlantic, Great Lakes, Plains, and Pacific.1 In addition, the Seventh Dis
3
trict and the five states that comprise it were included (i.e., Illinois, Indiana,
Iowa, Michigan, and Wisconsin).
The first comparative analysis was made between New England G SP esti
mates and the Federal Reserve Bank of Boston’s New England K-J esti
mates for the period 1969 to 1986.14 This analysis allowed the evaluation
of the K-J methodology vis-a-vis the B E A methodology. Comparisons
were made for total G SP and the G SP o f four sectors: manufacturing,
FRB CH ICAGO W orking Paper
February 1989 , W P -1989-3
12
nonmanufacturing, services, and finance, insurance and real estate (FIRE).
The results reveal that the K-J methodology produced output estimates that
are highly correlated with B E A estimates for every sector under study ex
cept FIRE. For both the constant and current dollar series, the correlations
were .99 (the correlation for F IR E was .96).
In terms of trends in the constant dollar series, the K-J estimates tracked
GSP estimates relatively closely at the total, manufacturing, and nonman
ufacturing levels. Graph 1 displays the trends for manufacturing which are
comparable to those of the two other aggregates. It is not surprising that
the K-J total and manufacturing estimates moved in lock step with the GSP
estimates because New England’s industry mix tends to be composed of
industries for which earnings is a relatively large share of GSP. Thus,
manufacturing earnings are a good proxy for manufacturing GSP.
Trends in services output diverged rather dramatically over the double dip
recession in the early 1980s, though for the other years they moved in tan
dem (Graph 2). Unlike the above sectorial trends, the trends in FIR E
output diverged significantly (Graph 3). These results suggest that earnings
are not an accurate proxy for services GSP and are particularly inaccurate
for FIR E GSP. For services, the divergence between the K-J and GSP
trends is most pronounced between 1980-82 when the K-J estimates show
a drop in output while the G SP estimates show continued growth. This
disparity exemplifies the errors in K-J estimates that arise from the as
sumption that the relationship of earnings to nonearnings at the national
level holds at the state level, as discussed above. In the New England case,
it appears that its share of national earnings in services took a dip whereas
its share of national nonearnings rose. Because rises such as these are not
captured by the K-J methodology, the value of IBT and capital charges for
New England’s services sector were underestimated by the K-J methodol
ogy and thereby resulted in an underestimation of services output.
For FIRE, the K-J estimates are off the mark across the entire 1969-1986
time period with the trend in GSP estimates significantly below the trend
in the K-J estimates. The probable reason for this disparity is that the K-J
methodology is weak at estimating real estate output because earnings
comprise a relatively small portion of the total. In the New England case,
it appears that the K-J methodology over “ blew-up” the output estimate for
real estate and thus the output estimate for FIR E as well.
Comparing manufacturing G SP to Census value-added allows one to com
pare direct manufacturing output estimated by manufacturing G SP to di
rect and indirect output (i.e., PSC) combined represented by Census
value-added.1 Because manufacturing GSP is based in part on Census
5
value-added, it is not surprising that trends in both measures are highly
FRB CH ICAGO Working Paper
February 7989, W P -1989-3
13
Graph 1
Comparison between New England G S P and
K-J estimates - Manufacturing
-
1969 70 71 72 73 74 75 76 '77 78 79 ’80 ’81 ’82 '83 '84 '85 ’86
FRB CH ICAGO Working Paper
February 1989, W P -1989-3
14
Graph 2
Comparison between New England G S P and
K-J estimates - Services
1969 70 71 72 73 74 75 76 ’77 78 79 ’80 '81 ’82 '83 '84 ’85 ’86
FRB CH ICAGO Working Paper
February 1989, W P -1989-3
15
Graph 3
Comparison between New England G S P and
K-J estimates - FIRE
1969 70 71 72 73 74 75 76 '77 78 79 '80 ’81 ’82 '83 '84 ’85 ’86
FRB CH ICAGO W orking Paper
February 1989, W P -1989-3
16
correlated (e.g., .93 and above for all the regions under study). The trend
in Census value-added, however, has been outpacing the trend in manu
facturing GSP, as exemplified at the national level (Graph 4). The spread
between the two began around the early 1970s with the largest gap occur
ring in the late 1970’s when manufacturing was at its historical peak.
The pattern of divergence seen at the national level is duplicated across all
the regions and states under study. Regressions of manufacturing G SP on
value-added with a 1974 dummy variable statistically confirm this diver
gence. For all the regions and states examined, the coefficient on the
dummy variable is significant and negative while the value-added coeffi
cient was on average below 0.75.
Differences in the trends are also evidenced by the differences in the annual
growth rates o f manufacturing GSP and value-added (Table 1). Annual
growth rates were calculated for two periods: 1) 1963-1973, a period when
manufacturing activity was expanding strongly with a minor dip in 1970
and 2) 1973-1984, a period during which manufacturing activity peaked and
plunged. During both periods, value-added across all regions and states
T a b le 1
T r e n d s in M a n u f a c t u r i n g G S P , V a lu e - A d d e d ,
a n d P u r c h a s e d S e rv ic e s : A n n u a l G r o w t h R a te s
1963-1976
Area1
VA
PSC
GSP
-- -percent- - - Z Z )
U.S.
4.9
4.1
New England
Mid-Atlantic
Great Lakes
Plains
Pacific
3.5
3.3
4.6
3.0
2.7
4.0
5.3
3.5
Seventh District
Illinois
Indiana
Iowa
Michigan
Wisconsin
60
.
4.5
5.2
4.4
4.9
6.5
60
.
5.0
1973-1984
4.1
3.9
4.1
5.7
4.1
4.4
VA
GSP
PSC
( ......... ---percent--
9.1
16
.
1.7
28
.
5.4
3.1
.5
.3
3.1
2.9
1.9
3.4
3.5
1.3
60
.
8.5
8.7
9.3
14.5
6.7
9.2
8.9
(2)
7.9
28
.
3.1
.
6
-2
.
.4
22
.
.3
26
.
.
1
-.4
26
.
3.6
-2
.
-.7
-.5
22
.
-1.0
22
.
4.0
1.5
3.6
2.4
(2)
4.4
1 See footnote 13 for definition of regions.
2 See footnote 16 regarding Michigan data problem.
FRB CH ICAGO W orking Paper
February 1989, W P -1989-3
17
Graph 4
Comparison between U.S. manufacturing
GSP and value-added
1963
'65
’67
’69
’71
’73
’75
'77
’79
’81
’83
'85
NOTE: Regional value-added data were available only up to 1984.
FRB CHICAGO W orking Paper
February 1989, W P -1 989-3
18
under study grew stronger than manufacturing GSP. Among the regions,
the percentage point difference between the two rates ranged from 1.1 in
the Seventh District to 0.5 in New England. Michigan recorded the largest
difference of 1.9 percentage points.
The reason for these growth rate differentials is that the purchased services
component (PSC) of value-added has grown at a faster rate than manu
facturing GSP. The PSC can be estimated as a residual of value-added less
G S P .1 An examination of the trend in the PSC estimates reveals that
6
purchased services have been growing at a much stronger pace than GSP.
Over the 1963-1973 period, P S C ’s annual growth rate was twice that of
G SP for the U.S. and four o f the six regions under study (Table 1). During
1973-1984, P SC ’s growth continued to exceed that of GSP, though it
slowed down significantly along with manufacturing G SP growth, reflecting
the nationwide decline in manufacturing activity. Although P S C ’s growth
decelerated, it did not turn negative as was the case with manufacturing
GSP for some regions and states. Graph 5 displays the national pattern in
manufacturing G SP and PSC trends which represents relatively well the
regional and state patterns.
In summary, the comparisons between GSP and K-J estimates and valueadded reveal that these various measures of output tend to track each other
closely at the national level (to be expected given the level of aggregation)
but tend to diverge at the regional and state level. The greatest divergence
occurs between nonmanufacturing G SP and K-J estimates and illustrates
the deficiencies of the K-J methodology. A relatively smaller divergence in
trends exists between manufacturing GSP and value-added and can be ex
plained by the exceptional growth in PSC. The results o f these analyses
suggest that caution should be used when interpreting trends in K-J esti
mates and value-added data.
Conclusion
Despite inherent problems, the G SP estimates are a substantial improve
ment over K-J estimates. By estimating directly a greater percentage of
total G SP in terms of earnings and non-earnings components, BE A avoids
making the erroneous assumption implicit in the K-J methodology that the
relationships between the non-earnings and earnings components at the
national level hold at the state level. Thus, GSP estimates account more
accurately for the distribution across states of IBT and capital charges. In
addition, the GSP estimates have advantages over other measures used to
track regional economic activity. Firstly, they offer more comprehensive
industry coverage than value-added data. Moreover, they net out pur
chased services from manufacturing output and distribute it across the ser-
FRB CH ICAGO Working Paper
February 1989, W P -1989-3
19
Graph 5
Comparison between U.S. manufacturing
GSP and purchased services
1963
'65
’67
’69
'71
73
75
'77
79
'81
'83
'85
NOTE: Because the scale of the graph changed, the trend in manufacturing
appears flatter than it did in Graph 4.
FRB CH ICAGO W orking Paper
February 1989, W P -1989-3
20
vices industries.
Secondly, they provide a more accurate measure of
regional economic activity than personal income and employment data. In
sum, they open the door to more accurate statistical research on differential
regional output growth, factor productivities, industrial restructuring, and
deindustrialization. In addition, they allow the disaggregation of regional
output growth into earnings and noneamings components. As a result,
regional growth differences in nonearnings components such as profits can
be examined.
Footnotes
1 John W. Kendrick and C. Milton Jaycox, “The Concept and Estimation of Gross
State Product,” S o u t h e r n E c o n o m i c J o u r n a l , October 1965, pp. 153-168.
2 See Robert Schnorbus and Alenka Giese, “Is the Seventh District Deindustrial
izing?,” FRB Chicago, E c o n o m i c P e r s p e c t i v e s , November/December 1987, pp.
3-10.
3 See Bureau of Economic Analysis (BEA)(U.S. Department of Commerce),
P e rs o n a l In c o m e :
E s tim a te s f o r
S ta te
1 9 2 9 -8 2 a n d a S ta te m e n t o f S o u rc e s a n d M e th o d s,
BEA REM 84-101, 1984.
4 See BEA S t a f f P a p e r 42, “Experimental Estimates of Gross State Product by
Industry,” U.S. Government Printing Office, Washington, D.C., 1985.
For the non-benchmark years, GSP estimates are calculated using both BEA’s
methodology and a form of the K-J methodology (excluding GSP estimates for
agriculture, real estate, and 1983-1984 manufacturing, see below). To estimate the
GSP components, BEA uses two approaches. Firstly, when there are raw data
available, BEA’s methodology—as described in the text—is used (raw data avail
able include earnings, rental income, and post-1982 IBT). Secondly, when there
are no raw data available, BEA uses a version of the K-J methodology and na
tional control totals from gross domestic product (GDP) data. For example, IBT
and capital charges estimates are interpolated using the K-J methodology based
on movements in earnings components and on national totals from GDP data.
For agriculture, real estate, and 1983-1984 manufacturing, capital charges are es
timated directly (i.e., without the use of the K-J methodology). For the real estate
industry two types of data are used, depending upon year (benchmark or non
benchmark). For the benchmark years, data from the population and housing
censuses and the U.S. Department of Agriculture are used.
For the non
benchmark years, data on rental income from the State personal income series are
used. For agriculture, estimates are approximated using U.S. Department of
Agriculture data. 1983-1984 manufacturing estimates are derived from data from
the Census’ Annual Survey of Manufactures (ASM).
5 The difficulty in distributing CAO costs correctly across states stems from the
fact that data on these costs are only available at the firm level (i.e., firm meaning
a legal entity such as a corporation) and not at the establishment level (i.e., eco-
FRB CH ICAGO W orking Paper
February 1989 , W P -1989-3
21
nomic unit or single physical location). Thus, the problem is how to distribute
CAO costs of multiestablishment firms. Questions that arise include whether
profits arise at headquarters or at the producing establishments and how to allo
cate overhead. See BEA S t a f f P a p e r 42.
6 Although BEA’s methodology solves the problems with allocating CAO costs
outlined in footnote 5, another problem remains regarding industry classification.
The industry classification of CAO costs by state are based on the industry clas
sification of the bulk of the operating establishments (across all states) that the
CAOs serve. Problems arise when the industry classification of operating estab
lishments in a certain state do not fall within the bulk industry classification or
when CAOs serve non-manufacturing establishments.
7 BEA M e m o r a n d u m , “Status of the BEA Gross State Product Estimates,” De
cember 1987.
8 Vernon Renshaw, Edward A. Trott, and Howard L. Friedenberg, “Gross State
Product by Industry, 1963-86,” S u r v e y o f C u r r e n t B u s i n e s s , May 1988, pp. 30-46.
9 “Gross Product by Industry: Comments on Recent Criticisms,”
r e n t B u s i n e s s , June 1988, pp. 132-133.
S u rv ey
o f C u r
1 The producer price indices for integrated circuits, the primary inputs of com
0
puters, began to spiral downward in the mid-1970s due to the learning curve ef
fect. It was not, however, until around 1982 that the impact of the relative decline
in their prices was reflected in a decline in the IPD of SIC 35.
11 For a discussion of the spatial separation of front-end activities (e.g., adminis
trative) and production-end activities see: Alenka S. Giese and William A. Testa,
“Can Industrial R&D Survive the Decline of Production Activity: A Case Study
of the Chicago Area,” E c o n o m i c D e v e l o p m e n t Q u a r t e r l y , November 1988, pp.
326-338.
12 Significant level differences exist in several types of industries. Firstly, for those
industries which are burdened with relatively higher IBT, K-J estimates tend to
be significantly below BEA estimates. For example, underestimation occurs in the
K-J estimates for the food and kindred products industry in Kentucky which is
impacted by “sin” taxes on alcohol. Secondly, for those industries that are highly
capital-intensive, K-J estimates tend to fall substantially short of BEA estimates.
For example, in the oil production industry, when output increases due to com
pletion of new construction or additions to capacity, employment and thereby
payroll usually remain relatively flat (except for inflationary adjustments) whereas
output increases, sometimes doubling or tripling. Thus, “blowing up” payroll data
would result in a significant underestimation of the increase in output.
1 Definition of the BEA regions are: New England = Connecticut, Maine,
3
Massachusetts, New Hampshire, Rhode Island, and Vermont; Mid-Atlantic =
New Jersey, New York, and Pennsylvania; Great Lakes = Illinois, Indiana,
?
Michigan, Ohio, and Wisconsin; Plains = Iowa, Kansas, Minnesota, Missouri,
Nebraska, North Dakota, and South Dakota; Pacific = Alaska, California,
Hawaii, Oregon, and Washington.
14 Anne E. Kinsella and Deanna M. Young,
G ro ss
S ta te
P ro d u ct
New
E n g la n d
1969-1986, Federal Reserve Bank of Boston, September 1988.
FRB CH ICAGO W orking Paper
February 1989, W P -1989-3
22
15 Value-added data are from the Bureau of the Census: Census o f Manufactures
and the Annual Survey o f Manufactures. Data were in historical dollars and were
deflated using the IPDs for manufacturing by state that were used by BEA to
deflate GSP. Externally purchased services (PSC) are assumed to be acquired
primarily from establishments located in the same state as the manufacturer.
Thus, PSC income is assumed to accrue to that state’s GSP.
16 The accuracy of these residual estimates of PSC needs to be qualified. Firstly,
for the benchmark years, the definitions and calculations of manufacturing GSP
and Census value-added differ. Manufacturing GSP is built up from data on each
component whereas value-added is derived as the residual of shipments less pur
chased materials. A second reason why the two series differ is that GSP estimates
are based on 1972 SIC code definitions while value-added for 1963-66 and 1967-71
are based on 1958 and 1967 definitions, respectively. Thirdly, BEA makes two
other adjustments to value-added in addition to netting out purchased services
(see text). Given these comparability problems, the level estimates of PSC may
not be accurate. This is the case for the state of Michigan for which the difference
between value-added and GSP is negative between 1963 and 1971. This problem
is mitigated at higher levels of geographic aggregation.
A better way to estimate PSC would be to: 1) Take national BEA value-added
data and distribute them across states and industries using regional shares of
value-added by industry calculated from Census value-added data. 2) Subtract
GSP from the state value-added by industry estimated in 1). Unfortunately, time
constraints did not allow the undertaking of this procedure.
FRB CH ICAGO W orking Paper
February 1989 , W P -1 989-3
23
@FedFRASER