View original document

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

Announcement
1999
ank Structure
& Competition
Conference

Federal Reserve Bank
of Chicago

First Quarter 1999

perspec ives
2

Competitive analysis in banking:
Appraisal of the methodologies

16

Bank Structure Conference announcement

18

Agglomeration in the U.S. auto supplier industry

35

The new view of growth and business cycles

1C
perspectives
11

President
Michael H. Moskow

Senior Vice President and Director of Research
William C. Hunter

Research Department
Financial Studies
Douglas Evanoff, Vice President

Macroeconomic Policy
Charles Evans, Vice President

Microeconomic Policy
Daniel Sullivan, Vice President

Regional Programs
William A. Testa, Vice President

Administration
Vance Lancaster, Research Officer

Economics Editor
David Marshall

Editor
HelenO’D. Koshy

Production
Rita Molloy, Kathryn Moran, Yvonne Peeples,
Roger Thryselius, Nancy Wellman

Economic Perspectives is published by the Research
Department of the Federal Reserve Bank of Chicago. The
views expressed are the authors’ and do not necessarily
reflect the views of the Federal Reserve Bank of Chicago
or the Federal Reserve System.
Single-copy subscriptions are available free of charge. Please
send requests for single- and multiple-copy subscriptions, back
issues, and address changes to the Public Information Center.
Federal Reserve Bank of Chicago, P.O. Box 834, Chicago,
Illinois 60690-0834, telephone 312-322-5111
or fax 312-322-5515.

Economic Perspectives and other Bank
publications are available on the World Wide Web
at http:Zwww.frbchi.org.

Articles may be reprinted provided the source is credited
and the Public Information Center is sent a copy of the
published material. Citations should include the following
information: author, year, title of article, Federal Reserve
Bank of Chicago, Economic Perspectives, quarter, and
page numbers.

ISSN 0164-0682

Contents
First Quarter 1999, Volume XXIII, Issue 1

2

Competitive analysis in banking: Appraisal of the methodologies
Nicola Cetorelli
How do we measure market power in the banking industry? This article provides an
overview of the methodology currently used in competitive analysis and highlights an
alternative technique that could be used to complement this methodology. Given the
ongoing process of consolidation in U.S. banking, assessing the competitiveness of
financial services markets is an important issue for policymakers.

16

Bank Structure Conference announcement

18

Agglomeration in the U.S. auto supplier industry
Thomas H. Klier
Analysis of a large set of plant-level data shows the auto supplier industry to be highly
spatially concentrated. Data on location of the plants’ customers, however, suggest that
immediate proximity to assembly plants is not necessary, despite a production system
that emphasizes low inventories and tight linkages.

35

The new view of growth and business cycles
Jonas D. M. Fisher
Evidence on the cost of business equipment investment supports a new way of under­
standing growth and business cycles. The equipment price has been falling for most of the
last 40 years and it tends to fall more the faster the economy is growing. This suggests
that technological change embodied in new capital equipment has a substantial effect on
growth and business cycles.

Competitive analysis in banking:
Appraisal of the methodologies
Nicola Cetorelli

Introduction and summary
Over the last 20 years, the U.S. banking industry has
experienced significant structural changes as the
result of an intense process of consolidation. From
197 5 to 1997, the number of commercial banks decreased
by about 35 percent, from 14,318 to 9,215. Since the
early 1980s, there have been an average of more than
400 mergers peryear (see Avery et al., 1997, and Sim­
mons and Stavins, 1998). The relaxation ofintrastate
branching restrictions, effective to differing degrees
in all states by 1992, and the passage in 1994 of the
Riegle-Neal Interstate Banking and Branching Effi­
ciency Act, which allows bank holding companies to
acquire banks in any state and, since June 1, 1997, to
open interstate branches, is certainly accelerating the
process of consolidation.
These significant changes raise important policy
concerns. On the one hand, one could argue that
banks are merging to fully exploit potential economies
of scale and/or scope. The possible improvements in
efficiency may translate into welfare gains for the
economy, to the extent that customers pay lower prices
for banks’ services or are able to obtain higher quality
services or services that could not have been offered
before.1 On the other hand, from the point of view of
public policy it is equally important to focus on the
effect of this restructuring process on the competi­
tive conditions of the banking industry. Do banks
gain market power from merging? If so, they will be
able to charge higher than competitive prices for their
products, thus inflicting welfare costs that could
more than offset any presumed benefit associated
with mergers.
In this article, I analyze competition in the bank­
ing industry, highlighting a very fundamental issue:
How do we measure market power? Do regulators
rely on accurate and effective procedures to evaluate
the competitive effects of a merger?

2

The U.S. Department of Justice, the Federal
Reserve System, the Federal Deposit Insurance Corpo­
ration (FDIC), and the Office of the Comptroller of the
Currency (OCC) enforce the antitrust laws in banking.
The procedures to evaluate the competitive impact of
a proposed merger may differ in some details among
the agencies, but they all share the same approach,
based on structural analysis of the banking market
affected by the merger. The basic guideline, established
by the Justice Department, requires the evaluation of
the concentration of deposit market shares held by
banks operating in the affected market. The importance
of market concentration finds its theoretical justifica­
tion in the so-called structure-conduct-performance
paradigm (Bain, 1951), which postulates that fewer
and larger firms (higher concentration) are more likely
to engage in anticompetitive conduct. For example, a
small number of large firms may be able to cooperate
and act as a monopoly {cartel}. Alternatively, one or
more firms together may be large enough to set higher
than competitive prices (acting as a dominant firm),
while the other (smaller) firms would act as a competi­
tive fringe, following the dominant firm’s behavior.
The most common measure of concentration,
and the one used by regulators, is the HerfindahlHirschman Index (HHI), which is defined as the sum
of the squared market shares of all banks in the market
(box 1 explains how the index is calculated).2 Accord­
ing to the current screening guidelines, if the post­
merger market HHI is lower than 1,800 points, twit/the
increase in the index from the pre-merger situation is

Nicola Cetorelli is an economist at the Federal Reserve
Bank of Chicago. The author thanks Eli Brewer,
Betsy Dale, Bob DeYoung, Doug Evanoff, Hesna
Genay, David Marshall, and Paula Worthington for
their comments.

Economic Perspectives

less than 200 points, the merger is presumed to have
no anticompetitive effects and is approved by the reg­
ulators. Should those threshold values be exceeded,
the regulators will check for the existence of potential
mitigatingfactors that would make it unlikely that
the merger could result in anticompetitive behavior.
The regulators also seek to identify those extreme
cases in which the potential welfare loss from the exer­
cise of market power would be smaller than the loss
produced by maintaining the status quo (for example,
the merger might prevent the failure of one of the
parties involved, thus preserving the stability of the
market).3 If the mitigating factors are not enough to
justify the merger, the regulators may require the dives­
titure of some branches and offices, in order to bring
the concentration indicator closer to or below the
threshold level. If divestiture would not accomplish
this goal, the merger application is denied.4 If the merg­
er does not violate the 1,800/200 rule,5 the application
is approved without further investigation.
Over the years, very few mergers have been
denied. However, this fact should not lead one to
conclude that the rules are not sufficiently stringent.
The official statistics do not show attempts to file
merger applications that were abandoned because of

a voluntary decision of the banks involved or informal
dissuasion by the regulators.
Does the ongoing merger and consolidation pro­
cess represent a real competitive threat? A survey of
local markets shows that concentration is a wide­
spread characteristic of the banking industry. For ex­
ample, in 1994, about 40 percent of metropolitan
statistical areas (MSAs) had HHIs greater than 1,800
(Rhoades, 1995b). Ifindeed high concentration im­
plies noncompetitive conduct, then policy concerns
about the welfare effects of future mergers may be
justified.
First, I review the appropriateness of the use of
the HHI as a main screening factor in merger analysis.
I examine the theoretical foundations of the market
concentration-market power relationship and how fo­
cusing on market structure to infer firms’ conduct
may lead to ambiguous or even misleading conclu­
sions about the potential effects of a merger.
Next, I survey the state of the art of the empirical
literature. If there are consistent and convincing em­
pirical results confirming the existence of the market
concentration-market power relationship, then it may
be appropriate to use it in policy analysis, even in the
absence of a solid theoretical explanation. While

BOX 1

Calculation oftheHerfindahl-Hirschman Index (HHI)
The HHI formula is

«
HHI = 2
z=l

9

z

where MS. is the market share of bank i and n is the
number of banks in the market.
Suppose a market has five banks. The share
of total deposits of each bank is as follow s:

Deposit
market share
Bank 1

30

Bank 2

25

Bank 3

21

Bank 4

16

Bank 5

8

TheHHI = 302 + 252 + 212+162 + 82 = 2,286.
Suppose that banks 3 and 5 merge. After the
merger, the HHI = 302 + 292 + 252 + 162 = 2,622, with
a post-merger increase AHHI = 336. In antitrust

Federal Reserve Bank of Chicago

evaluation this merger may be rejected, because it
violates the 1,800/200 rule.
By construction, the HHI has an upper value
of 10,000, in the case of a monopolist firm w ith 100
percent share of the market, and tends to zero in
the case of a large number of firms w ith very small
market shares.
The HHI sy nthesizes information on both the
distribution of market shares and the number of
banks in the market. With some manipulation it
could be rewritten as

77

w here Fis the coefficient of variation of deposit
market shares, and n is the number of firms in the
market. This feature of the HHI makes it more pop­
ular than other concentration indicators, such as
the n-firm ratio, calculated as the sum of the mar­
ket shares of the n largest firms in the market,
where n is usually 3 or 4.

3

there have been important contributions confirming a
positive and significant relationship between market
concentration and the exercise of market power, other
recent work has cast doubt on the overall empirical
strength of such a relationship.
1 then describe an alternative methodology of
competitive analysis that does not infer banks’ con­
duct through the analysis of market structure. This
methodology recognizes that firms’ behavior differs
depending on whether they operate in a perfectly
competitive market, a monopolistic market, or any
other prevalent market structure. 1 survey the appli­
cations of this methodology, which is based on the
estimation of a direct indicator of firms’ behavior, for
the banking industry.
Finally, 1 present some results of a specific empir­
ical application of this methodology to the Italian
banking industry. The analysis of Italy is relevant
because the Italian banking industry has experienced
a similar pattern of structural and regulatory changes
as U.S. banking. In particular, as the result of an on­
going process of consolidation, the Italian HH1 has
been steadily increasing. The results of my empirical
analysis indicate a steady convergence toward com­
petitive conditions, providing evidence that changes
in market concentration may not always provide correct
information about the exercise of market power.

Theory behind the HerfindahlHirschman Index
As discussed above, the use of concentration
ratios to evaluate competitive conditions relies on
the theoretical predictions of the structure-conductperformance paradigm. According to this paradigm,
structure affects the conduct of firms, which ultimately
determines their performance. Concentration of market
shares will facilitate the adoption of collusive conduct
and, ultimately, the setting of prices departing from
the perfectly competitive benchmark. In a perfectly
competitive market, firms are considered too small to
have an individual impact on the price of the good
they produce. From the point of view of social welfare,
perfect competition represents an ideal benchmark,
since consumers (in this case bank customers) pay the
lowest possible price for the product they demand.
Any situation in which firms command some degree
of market power and are therefore able to set higher
than competitive prices implies a social cost in terms
of welfare loss for consumers.
The structure-conduct-performance paradigm
predicts that there is an increasing relationship be­
tween the level of market concentration and market
power. Some authors are more precise in stating that

4

the relationship, while it is increasing, may not be linear.
One would expect that at low levels of concentration,
conduct is close to competitive, and an increase in
concentration would generate a substantial increase
in market power. At high levels of concentration, con­
duct is already very far from the competitive bench­
mark, and an additional increase would not increase
market power very much. Given this argument, the
market concentration-market power relationship
should be S-shaped, as shown in figure 1 (Carlton
andPerloff, 1989).
Is it possible to derive an optimal behavior rule
from a model of industrial organization theory that
predicts an increasing relationship between market
concentration and market power? Can we rely on such
a model to find a theoretical justification for, say, the
1,800/200 rule? The answer is yes, but only if one makes
strong, restrictive assumptions about firms’ behavior,
such as assuming that firms behave as Cournot
oligopolists. Under Cournot conduct, a firm makes
the simplistic assumption that all other firms have no
reaction to a change in its behavior (see the technical
appendix for the analytical derivation of this result).
However, in more general (and plausible) theoretical
models that allow for active interactions among firms,
the market concentration-market power relationship
is less obvious.
Thus, it seems that we cannot rely too much on
theory to justify the postulated market concentrationmarket power relationship. Before surveying the
approach taken in the profession, which has been to
turn to a direct empirical corroboration of the postulated
relationship, 1 present some simple numerical exam­
ples showing that, in the absence of a complete theory
that can explain the market concentration-market

Economic Perspectives

banks behaving as a competitive fringe, adjusting to
the noncompetitive choices of the dominant firm.
In the second example, the pre-merger market has
15 banks, two with 15 percent market shares, one with
Numerical examples
10 percent, and 12 with 5 percent (see table 1, exam­
These examples demonstrate the following two
ple 2). The two larger banks, Bj and B2, taken sepa­
assertions: First, even when the 1,800/200 rule is not
rately, may still be too small to behave as dominant
firms. In addition, tacit or explicit collusion between
violated, a merger may generate anticompetitive con­
duct. Second, a merger may be procompetitive even
them to act together as a dominant firm may still be un­
when the 1,800/200 rule is violated.
likely, given the fact that the combined market share
In the first two examples, the basic guidelines are
may not generate the market power and extra profits
not violated. However, the mergers may generate the
necessary to offset the costs associated with collu­
right conditions for monopoly power, not necessarily
sion.7 The HH1 of 800 may therefore be correct in
exercised only by the banks involved in the merger.
characterizing a competitive market.
Table 1 summarizes the examples.
Suppose banks B3 and B]5 merge. The post-merger
In a pre-merger market with 20 banks, each with
structure now has three banks with a 15 percent market
a 5 percent market share (see table 1, example 1), the
share each and 11 banks with 5 percent each. The
HH1 (52 + 52 4---- H 52 = 500) characterizes a market with
post-mergerHHlisnow950. As in the first example,
a relatively large number of banks with equal and small
according to the guidelines the market would still be
market shares and is presumably associated with a low
considered unconcentrated. However, the three major
likelihood of anticompetitive behavior. Suppose five
banks may now be able to coordinate (explicitly or
of the banks are involved in a series of mergers. When
tacitly) their action, thus producing adverse competitive
all the mergers are completed, the market has one bank
conditions. (Note also that the two larger banks in
the pre-merger market are benefiting from a merger
with a 25 percent market share and 15 banks with 5
percent each. The post-merger HH1 of 1,000 would
that did not directly involve them).
still be considered (borderline) unconcentrated.6
The third example describes a market in which
However, the newly created bank, with a 25 percent
some degree of collusive behavior might have been
market share, may be able to act as a dominantfirm,
observed prior to the merger (see table 1, example 3).
setting noncompetitive prices, with the remaining 15
The merger could create conditions under which the
stable collusive agreement would break
down, thus restoring market competition.
TABLE 1
However, since the basic guidelines are
violated, the merger could be rejected and
Examples of pre- and post-merger markets
the exercise of market power preserved.
Example 1
Pre-merger market (20 banks)
The pre-merger market has seven
Bank
B,
b2
b3
b20
banks, three with 20 percent market
Marketshare (%)
5
5
5
5
shares, two with 15 percent shares, and
Post-merger market (1 6 banks)
two with 5 percent shares. The HH1 of
Bank
B,
B,e
b2
b3
1,700, classifying the market as moder­
Marketshare (%)
25
5
5
5
ately concentrated, may not fully account
for a situation in which the three largest
Example 2
Pre-merger market (15 banks)
banks, Bp B2, and B3, may be able to col­
Bank
B,
B,e
b4
b2
b3
lude. In the event of a merger between
Marketshare (%)
15
15
10
5
5
banks B4 and Bs, the post-merger market
Post-merger market (14 banks)
would have six banks, one with a 30 per­
Bank
B,
B3
b2
b4
b14
cent market share, three with 20 percent
Marketshare (%)
15
15
15
5
5
each, and two with 5 percent each. The
Example 3
post-merger HH1 of 2,150 identifies this
Pre-merger market (7 banks)
as a highly concentrated market. In addi­
Bank
B,
Be
B7
b2
b3
b4
b5
tion,
since the change in the HH1 would
Marketshare (%)
20
20
20
15
15
5
5
be
more
than 200 points, there are grounds
Post-merger market (6 banks)
Bank
B,
Be
for
the
regulator
to reject the merger
b4
b2
b3
b7
Marketshare (%)
30
20
20
20
5
5
application. However, the stability of a

power relationship, it is possible to generate ambigu­
ous or even incorrect predictions about the effects of
a structural change on competition.

Federal Reserve Bank of Chicago

5

collusive agreement is known to decrease with the num­
ber of participants. In the new market structure, with
four large players, the collusion might break down. In
that case, the merger would actually be procompetitive.
In considering whether to reject the merger appli­
cation, the regulator may impose some degree of dives­
titure on the banks involved in the mergers. Ironically,
banks Bp B ,. and B3, which were not involved in the
merger, could benefit in this case, as the post-divesti­
ture B4 may not be strong enough to undermine the
stability of their pre-merger collusive agreement.
The market dynamics described in these numerical
examples are all hypothetical. My point is that whether
a merger will generate (undetected) anticompetitive
conditions or actually improve competition cannot be
determined unambiguously just by looking at market
structure. Banks’ behavior can only be measured accu­
rately through direct empirical analysis.

Empirical evidence
The empirical evidence for the existence of the mar­
ket concentration-market power relationship is mixed.
Some influential papers have suggested a positive rela­
tionship between concentration and the degree of mar­
ket power. For example, Berger and Hannan (1989)
analyze a cross-section ofbanking markets in 1983-85.
After controlling for various factors affecting price-setting behavior, the authors find that deposit rates are sig­
nificantly lower in the most concentrated markets.
Other work compares the time-series behavior of
the deposit interest rate (and/or the loan rate) with the
benchmark money market rate, which is not controlled
by the banks. If banks have market power, they will,
for example, quickly lower the deposit rate when the
money market rate decreases, but the deposit rate will
be sluggish when the money market rate increases.
Conversely, in perfect competition one should expect
quick reactivity in both cases. Hannan and Berger
(1991) and Neumark and Sharpe (1992) find evidence of
deposit rate rigidity and, thus, evidence of market power
in the U.S. banking industry. Importantly, they find a
higher level of rigidity in markets with higher HHIs.
However, recent research casts doubt on the mar­
ket concentration-market power relationship. Review­
ing Berger and Hannan’s (1989) results, Jackson (1992b)
suggests that the market concentration-market power
relationship may not be monotonic. He finds that such
a relationship already holds at low levels of concentra­
tion, but in markets with middle levels of concentration
the relationship vanishes, and it actually changes sign
in highly concentrated markets (although this is a
less robust result). In other words, at higher levels
of concentration, an increase in concentration may

6

imply less anticompetitive behavior, as suggested in
example 3 of table 1.
In another work focusing on the rigidity of depos­
it rates, Jackson (1997) presents additional evidence
that the market concentration-market power relation­
ship may not be monotonic. He finds that while it is
true that at high levels of concentration price rigidity
increases, this is also the case at low levels of concen­
tration. This suggests a U-shaped relationship between
market power and market concentration which is not
consistent with the structure-conduct-performance
hypothesis.
Similarly, Rhoades (1995a) observes that structur­
al characteristics may vary widely for markets exhibit­
ing similar HHI levels. In particular, the market share
distribution may differ substantially. As shown in
example 1 above, firms’ conduct may be very different
depending on market share distribution. Rhoades
shows that market share inequality and the number
of firms in the market have an effect on banks’ profit­
ability that is independent of the HHI, despite the fact
that (as shown in box 1) the HHI incorporates informa­
tion on both market share variability and the number
of firms. Finally, in an analysis similar to Berger and
Hannan’s (1989), Hannan (1997) extends Rhoades’s
(1995a) contribution by analyzing the impact of these
two factors on deposit rate levels. His results for a
cross-section ofbanking markets using November
1993 data show, first, that the HHI was not significant
in explaining deposit rates and, second, that it was
not able to take into account the separate importance
of market share inequality and the number of firms.
Thus, a lack of strong theoretical foundations
and mixed empirical evidence motivate the search for
alternative methodologies to investigate firms’ com­
petitive behavior.

Oligopoly theory and the measurement
of market power
Methodologies in the “new empirical industrial
organization” literature analyze firms’ conduct directly,
instead of relying on observation of the market struc­
ture.8 Following this approach, the relationship between
theory and firms’ conduct becomes unambiguous.
For instance, as mentioned earlier, if banks are behav­
ing as Cournot oligopolists, the market concentrationmarket power relationship would be theoretically
grounded and the use of the HHI to infer firms’ con­
duct would be appropriate. This alternative method­
ology allows us to test whether indeed banks behave
as Cournot oligopolists. However, the methodology
is flexible enough to allow us to test for behavior
that could be consistent with alternative models of

Economic Perspectives

oligopoly theory. In such a case the market concen­
tration-market power relationship would not be as
clearly identified as in the Cournot case, but one
would still be able to quantify the departure from per­
fect competition and, hence, to assess the degree of
market power exercised in the industry.
The technical appendix provides details of the
methodology. The following example illustrates the
intuition. Suppose there is an exogenous increase in
the demand for bank loans. In response, banks will
take into account the cost they would incur in increas­
ing the quantity of loans, the reactivity of demand
itself to possible increases in the loan rate, and the
expected reaction of the other banks in the market to
their chosen course of action. In particular, the degree
of interaction with the other banks in the market could
dilfer substantially, depending on whether banks are
in perfect competition with each other or enjoy some
degree of market power. More precisely, the parameter
of banks’ interaction should be equal to 0 if the market
is perfectly competitive, equal to 1 if it is monopolis­
tic, and should take intermediate values between 0
and 1 if banks are neither perfectly competitive or
monopolistic but still exercise a positive degree of
market power. Using appropriate econometric model­
ing techniques, one can estimate this parameter of
interaction and, therefore, a quantifiable measure of
market power.
The advantage of this approach is that it is rigor­
ously based on theory and does not require indirect
(and perhaps ambiguous) inferences about market
power through measures of market concentration. The
major limitation of the approach is that it requires de­
tailed information, mainly on cost and demand condi­
tions at the firm level.

Applications to the banking industry
Spiller andFavaro (1984) estimate the parameter
of banks’ interaction for the Uruguayan banking indus­
try in a period characterized by a significant relaxation
of entry regulations. They apply a refinement of the
methodology proposed by Gollop and Roberts (1979)
to see whether different groups of banks have differ­
ent reactions to other groups’ change in behavior
They reject Cournot conduct and find evidence of
dominant firm-competitive fringe behavior, with a
significant degree of oligopoly power, although this
is substantially reduced after deregulation. Gelfand
and Spiller (1987) extend the analysis ofUruguayan
banks, treating the banks as multiproduct firms, the
products being loans in the domestic currency and in
U.S. dollars. They find evidence of noncompetitive
behavior and, in particular, behavior consistent with
mutualforbearance, whereby firms avoid changing

Federal Reserve Bank of Chicago

behavior in one market fearing retaliation in another
market, and with spoiling, whereby firms adopt pred­
atory strategies. Applying the methodology to the
Norwegian banking industry, Berg and Kim (1994)
find that Cournot behavior is strongly rejected by the
data and that instead banks behave as if they expect
retaliation from their competitors in response to a
change in their own behavior. Berg and Kim (1996)
also investigate Norwegian banks as multiproduct
firms, distinguishing between the retail and corporate
banking markets. They find banks’ degree of oligopoly
power to be relatively high in the retail market and
lower in the corporate market. Interestingly, the
Herfindahl indicators for the two markets analyzed
suggest opposite findings. Shaffer (1989), using
aggregate data for the U.S. banking industry, finds
no evidence of oligopoly power. Similarly, in a study
of Canadian banking, Shaffer (1993a) finds that despite
structural and regulatory changes, Canadian banks
operate in a market exhibiting perfect competition.
Shaffer and Di Salvo (1994) focus on a local market in
Pennsylvania with only two banks. They find that
banks’ conduct is imperfectly competitive, but closer
to perfect competition than one would expect, given
the very high degree of concentration in that market.

Measuring market power: Results from an
application to the Italian banking industry
Next, I present some results from an application
of the methodology outlined above to the Italian bank­
ing industry. The remainder of the section is based
onAngelini and Cetorelli (1998).
As mentioned in the introduction, there are at
least two reasons the evolution of the Italian banking
industry is of interest. First, the Italian banking indus­
try is experiencing a similar pattern of regulatory and
structural changes as that observed in the U.S. In the
late 1980s, the requirement that Italian banks obtain a
specific authorization from the central bank to open
an additional branch was eliminated. Consequently,
from 1983 to 1993 the number of branches increased
by 67 percent. At the same time, mainly based on the
anticipated opening of Italy’s national borders to in­
ternational competition, widespread merger activity
reduced the number of banks by more than 10 per­
cent, to a total of approximately 900. It is not clear a
priori whether such changes have actually enhanced
competition. Second, the results for Italy highlight
the possibility that changes in market concentration
may provide misleading information on the exercise
of market power.
To determine an average indicator of banks’ in­
teraction, Angelini and Cetorelli (1998) analyze the
market for commercial loans in 1983-93, pooling data

7

on all individual banking institutions, in substance
treating the market for commercial loans as having a
national dimension. It is usually argued that, especially
for wholesale loans, the market boundaries are indeed
very wide. Given that Italy is about as large as a mid­
size U.S. state, using such a broad market definition
seems appropriate. Also, performing the analysis at
the national level increases the potential for finding
evidence of perfect competition. This is true at least
in terms of the structure-conduct-performance ap­
proach, since, as we will see below, market concentra­
tion is very low at the national level. With a possible
bias in the study toward a finding of competition, there­
fore, evidence of noncompetitive behavior would be a
strong result.
Angelini and Cetorelli (1998) make the following
observations about the level of concentration of the
Italian banking industry. First, the HH1, calculated
on both deposits and loans, remained practically un­
changed in the first part of the sample period, but in­
creased noticeably after 1990, clearly due to the wave
of bank mergers mentioned above. Second, in absolute
terms the HH1 remains very low, going from about 200
to 260 points over the entire period. Figure 2 plots the
HH1 time series for both deposits and loans. Follow­
ing the predictions of the structure-conduct-perfor­
mance paradigm, these two observations would imply
that, given the extremely low level of concentration,
the Italian banking industry should exhibit a very high
degree of competition over the entire sample period,
but with gradual movement toward conditions more
appropriate to the exercise of market power.
In fact, the results of the econometric estimation
contradict both predictions of the structure-conductperformance paradigm. Figure 3 shows the estimates
of the parameter of banks’ interaction for each year
between 1983 and 1993, a period including years before
and after the regulatory changes. As explained earlier,
the parameter should take values between 0 and 1,
with 0 representing the perfectly competitive bench­
mark and 1 the monopolistic benchmark. However, the
results show the parameter is significantly different
from 0 (and from 1) for almost the entire sample period,
thus rejecting the hypothesis that the Italian banking
industry is perfectly competitive (as well as the hypoth­
esis that it is a perfect monopoly). This finding con­
tradicts the inference one would draw from the HH1.
Indeed, given the very low level of concentration,
one might expect the market for commercial loans at
the national level to be very competitive.
A further observation is that the parameter is
well above 0 in the initial part of the sample, prior to
deregulation, and shows an approximately steady

8

decline throughout the rest of the sample period.
This can be seen as evidence that the regulatory and
structural changes have indeed enhanced the overall
competitiveness of the banking industry. Finally, the
parameter approaches 0, suggesting the presence of
perfectly competitive conditions, toward the end of
the period. This represents a second element of con­
tradiction with the information in the HH1, which is
increasing in the final years of the sample period.
In addition to the estimation of the parameter of
interaction, Angelini and Cetorelli (1998) estimate a
parameter measuring the elasticity of demand for com­
mercial loans. As mentioned earlier, in deciding on be­
havior, banks have to take into account not only the

Economic Perspectives

expected reaction of other banks but also the reaction
of customers. Whether the market for loans exhibits
a high or low elasticity to changes in the loan rate is
crucial to banks’ ability to exercise market power and
affect profits. The intuition is simple. Suppose the
parameter ofinteraction is very high, close to 1, approxi­
mating ideal conditions for the exercise of market power.
Banks would attempt to keep a high loan rate, or to
increase it, to maximize their profits. However, if market
demand elasticity is also high, borrowers are likely to
reduce substantially their demand for loans in the
case of a price increase. In such a case, banks will be
constrained in their ability to profit from their market
power. The opposite would be true in the case of a
rigid demand schedule.
This consideration is important, therefore, if we
are interested in exploring the actual welfare cost of
market power, in terms of how high the loan rate is
relative to what it would be under perfect competition.
To obtain a quantifiable measure of this, Angelini and
Cetorelli (1998) compute the ratio of the parameter of
banks’ interaction and the parameter measuring demand
elasticity. When this ratio is close to 0, it means that
the market exhibits competitive conditions, regardless
of banks’ potential ability to exercise market power.
Figure 4 reports estimates for this ratio for every year
in the sample period. Between 1984 and 1986, interest
rates on loans charged by banks were about 2 percent­
age points above the level that would have been charged
under competitive conditions (interest rates on loans
averaged around 21 percent). This gap declined to
about 1 percentage point in 1987-89, then dropped to
practically 0 at the beginning of the new decade. This
provides evidence that the Italian banking industry

Federal Reserve Bank of Chicago

has changed substantially as a result of the process
of deregulation and consolidation that began in the
late 1980s.

Conclusion
This article has presented an overview of the
methodologies used in competitive analysis of the
banking industry. Given the ongoing process of con­
solidation in the U.S. banking industry, properly iden­
tifying the conditions for the exercise of banks’ market
power is highly relevant for policy analysis.
1 have briefly outlined the antitrust analysis pro­
cedure currently followed by the regulators. Drawing
on the existing literature, 1 have highlighted some chal­
lenges to the theoretical foundations of the current
approach, which is based on the identification of an
increasing, monotonic relationship between market
concentration and market power. Only under rather
strong, restrictive assumptions about the behavior
of banking firms is this relationship identifiable. As
shown in the numerical examples, relying on concen­
tration measures alone to infer industry conduct may
lead to possibly incorrect conclusions. The empirical
evidence on the existence of the market concentrationmarket power relationship is rather mixed, in light of
several recent works that cast doubt on the robust­
ness of such a relationship.
An alternative methodology for the identification
of parameters of firms’ conduct and the degree of mar­
ket power, which does not rely on indirect inferences
of market structure analysis, requires an econometric
estimation of market demand and supply conditions.
The testable implications associated with this approach
allow us to unambiguously identify firms’ conduct.
The results from an empirical application of this meth­
odology to the Italian banking industry provide evi­
dence that contradicts the inferences of the
structure-conduct-performance approach.
Adopting this alternative methodology to identi­
fy the parameter of banks’ interaction brings a higher
rigor to the antitrust analysis, implicit in the econo­
metric exercise required to extract information from
industry data. This is, however, also its principal
shortcoming, in terms of the need for more detailed
data and the greater difficulty associated with the im­
plementation and interpretation of the econometric
work. Conversely, the main advantage of the current
approach to competitive analysis is that HHI indica­
tors are relatively easy to compute and allow the reg­
ulators to formulate objective statements (for example,
setting the 1,800/200 guideline) and deliver opinions
that are less subject to arbitrary judgements. None­
theless, it is important to recognize the potential

9

shortcomings of the current approach and to test
for accountability when developments in economic
research provide the appropriate tools.
For example, the alternative methodology pre­
sented in this article could be applied to markets in
which mergers have been approved to analyze banks’
conduct before and after the change in market struc­
ture.9 In addition to an “after the fact” analysis, the

methodology could be used routinely to overview
market conditions and to provide ex ante information
that could be used by regulators when a merger appli­
cation is filed, perhaps to resolve potential ambiguities
associated with mere observation of market structure.
In this way, the methodology could be adopted to com­
plement the current procedure for antitrust analysis.

TECHNICAL APPENDIX

Details of the methodology
Estimating market power
The basic elements of the methodology can be
illustrated as follows.1 In an industry producing a sin­
gle good, let p be the market price of product y and
let y be the quantity produced by firm j,j= 1,..., /«,
and Zy. =y. Let the demand function, written in
inverse form, be p =p (y, z), where z is a vector of ex­
ogenous variables affecting demand. In addition, let
C(_v, co) be the cost function for firmj, where cois
the vector of the prices of the factors of production
employed by firmj.
Firms in this industry behave as profit maximizers.
The profit maximization problem for firmj is written as

1)

Maxpy,z)y; - C(yy, CO,.).
•V7

If firms were in perfect competition with each other,
they would set their optimal quantities at the point
where the marginal cost of production would equal
the market price, that is,

4) p = C\yp cop

where the parameter 0 is an index of oligopoly con­
duct, quantifying the departure from the competitive
benchmark. Equation 4 is a very general expression
embedding various models of oligopoly behavior,
which can be estimated econometrically. To appreci­
ate its generality, it is perhaps convenient to interpret
0as a parameter measuring the “conjectured” or “per­
ceived” response of the entire industry to a change in
quantity operated by firmj. Solve the maximization
problem in equation 1 in more extensive form as

5) ^+^p^y,-C'(y,,cop = 0.
3y dy.
Multiply and divide the second term of equation
5 by y. Then, rearranging terms, the equation can be
rewritten as

6)
2)

p = C\yj,tOj),

where C'(v co.), is the marginal cost of firmj.
At the opposite extreme, suppose there is only
one firm in the industry, operating as a monopolist.
In such a case, we know that the firm would set
quantities to a level where marginal revenue equals
marginal cost, or

3)

/?= C'(y7, cop--P,

/>= C'(y, co)-^-y,
dy

where p + —y is the monopolist marginal revenue

where

7)

e=-^-l,e<0

op y

is the semi-elasticity of demand and

8)

=

y’j
y

is the so-called conjectural elasticity, that is, the per­
centage variation in aggregate output due to firmy’s

. In intermediate oligopolistic structures,
with m firms operating in the market, conduct would
be summarized by the general expression

10

Economic Perspectives

change in y.. It should be clear that one does not
need to impose any a priori restriction on 0 , that is,
any behavioral model is a priori plausible, and the
more appropriate one can be tested and identified
econometrically. For example, if firms were Cournot
= 1. Recall that under Cournot

oligopolists, then

behavior, firmj expects that all other firms will not
adjust their quantities to a change iny. Therefore,
since y = Zj) incorporates firmj quantity, the total
variation in output to a change in y. must equal unity.
Thus, under Cournot, 0. would reduce to the market
share of firmj.
If firms were instead in perfect competition,

= 0, hence 0 = 0. In the case of monopoly,
yj
—— = 1 andy =y, hence 0 =1. Therefore, the
O',
>
J
convenient feature of this approach is that it specifies
well-defined boundaries in terms of industry equilib­
rium conditions (perfect competition at one end and
monopoly at the other), within which it is possible
to identify the actual underlying characteristics of
firms’ conduct.
Given the generality of the methodology, one
can also test whether 0, 0, where// = 1,..., / and i =
1,..., n and l+n = m. This would allow us to test, for
example, whether firms behave according to dominant
firm or leader-followers models.

Assume now that all firms form the same, identi­
cal conjecture about how the rest of the industry
would react to a change in their own quantities. In
addition, assume that these identical conjectures will
also stay the same over time and over changes in
market structure (for example, distribution of market
shares and number of firms). Under these conditions,

= -p\i j, where y is a given constant.

Consequently

then

Analytical derivation of the market concentrationmarket power relationship
We can also see now under what behavioral restric­
tions it is possible to identify a relationship between
market concentration and market power.2 Define the
degree of market power of firm/' as
y/-C(y.,ca.)

9)

e

p

where e =——(e<0) is the elasticity of demand.
tyy

12)

a = -yHHI.
e

The Cournot model, where y= 1, is an example
of a model that would identify a proportional relation­
ship between market concentration and market power.
However, we have already remarked that the Cournot
conjecture is rather restrictive. It seems even more
restrictive to assume identical conjectures equal to
some arbitrary constant y. Moreover, note the impor­
tance of the assumption that the identical conjectures
will have to stay unchanged over time and in case of
a change in market structure. This implies assuming
that y and HHIare independent from each other. Yet,
as we argued earlier, a change in market structure,
such as the one determined by a merger, whereby the
distribution of market shares and the number of firms
operating in the industry vary, will have an effect on
how firms perceive the conduct of one another. This
effect on conduct will not necessarily be the same
for all firms (see, for example, the numerical examples
section of the text). Therefore, the behavioral restric­
tions required to derive the market concentrationmarket power relationship from theory would indeed
seem too strong to be accepted.
In the more general (and more plausible) case
where
^,j * i, the expression for a does
not allow7one to derive the HHI. Therefore, under

Now define the degree of market power of the in­
dustry as a firm average, weighted by firms’ relative size,

10)

a=s
j

p-C/(y,,0),)'l Jfy =

p

J y

■> e y '

Given the definition of 0.J we can rewrite this last
expression as

Federal Reserve Bank of Chicago

these more general conditions, we cannot rely on the
HHI to make predictions regarding firms’ conduct.
Nonetheless, as stated above, we can test economet­
rically whether the Cournot or the constant y restric­
tions can be rejected against alternative theoretical
specifications. As Bresnahan( 1989, p. 1031) stated,
“Only econometric problems, not fundamental prob­
lems of interpretation, cloud this inference about
what has been determined empirically.”

11

Details ofthe empirical implementation
As we saw above, estimating the degree of market
power means being able to identify the conduct
parameter 0 in equation 4, here rewritten for conve­
nience of exposition as

/’ = c'Cv7,®7)-JvA,

C,
C'(v.,co)= — al + a, ln( v) + X b, ln(co„) .
i=l

In addition, the parameter e is recovered by
estimating simultaneously a loans demand function,
specified as
ln(v) = <f+d1p + <f ln(z) + <V4[ln(z)/?],

wherep now indicates the interest rate on commercial
loans, y indicates the quantity of commercial loans,
and co, the vector of factor prices, includes labor cost,
capital expenses, and the interest rate on deposits.
For the identification of the parameter of conduct
0., we need information on the marginal cost function
C'(y, co ) and on the inverse of the semi-elasticity on
loans demand, -L = ^-v. One can obtain this
e
additional information at different degrees of refine­
ment, depending in practice on data availability.
Angelini and Cetorelli (1998) estimate the parameters of
the marginal cost function using the widely used trans­
log specification, deriving the following expression:

where z is an exogenous shifter of demand, such as
investments or GDP.
Finally, although it would be feasible in terms of
data availability to test various models of oligopoly,
thus identifying distinct parameters of conduct, 0,, 0.,
Angelini and Cetorelli (1998) focus on the determination
of an average indicator of conduct, 0 (see Bresnahan,
1982, for details). Such an indicator gives a first approxi­
mation of the overall conditions for the exercise of
market power in the industry. Since such a study has
never been conducted before for the Italian banking
industry, 1 believe there is high informational value in
the average indicator 0.
‘The remainder of the section is based on Appelbaum (1982)
and Bresnahan (1989).

2The derivation is based on Cowling and Waterson (1976).

NOTES
Examples of research work on the impact on efficiency of
bank mergers include Berger and Humphrey (1992), DeYoung
(1997), Hughes et al. (1996), Rhoades (1993b), and Shaffer
(1993b). Other authors have sought to evaluate the impact on
profitability (for example, Berger and Humphrey, 1992; Cor­
nett and Tehranian, 1992; Pilloff, 1996; and Akhavein et al.,
1997) and on production decisions, in particular on lending to
small business (for example, Berger et al., 1997).

2An alternative measure also used in research is the sum of the
market shares of the largest firms in the industry, usually the
largest three or four firms.
3For a thorough description of the use of mitigating factors in
antitrust analysis, see Holder (1993a).
4For a detailed description of the official guidelines for competi­
tive analysis in banking, see, for example, Bureau of National
Affairs (1984, 1992), Litan (1994), Holder (1993a, 1993b),
and Di Salvo (1997).
5To be precise, thrift institutions are currently included in the
calculation of the HHI. Their market shares, however, have only
a 50 percent weight (20 percent for the Justice Department’s
evaluation procedure), which in any case always determines a
reduction in the HHI calculated on banks only. Because of the
inclusion of the thrift institutions, the 1,800/200 rule is some­
times called the 1,800/200/50 rule.

12

6The Justice Department’s horizontal merger guidelines define
markets with a post-merger HHI below 1,000 as unconcentrated
and unlikely to present anticompetitive concerns. Markets
with a post-merger HHI between 1,000 and 1,800 are defined
as moderately concentrated. In such markets a variation in the
HHI of less than 100 points is unlikely to present anticompet­
itive concerns. Markets with a post-merger HHI above 1,800
are defined as highly concentrated, and a variation of the HHI
greater than 50 points is thought to have adverse competitive
consequences. In the past several years, however, the Justice
Department has not challenged a merger unless the post-merger
HHI was at least 1,800 and the change in the HHI at least 200
points (see Litan, 1994).
7A firm joining a collusive agreement always has an incentive
to abandon the agreement (or “cheat”) and set prices and/or
quantities that maximize its own profits. The costs associated
with the collusive agreement are therefore expressed either in
terms of the losses suffered by participants in the event that
one of them cheats, or in terms of the punishment that a firm
would sustain in the event it is caught cheating (for instance,
all firms revert to competitive pricing forever after collusion
breaks down, hence the deviating firm will no longer be able to
make positive profits.)
important methodological contributions include Iwata (1974),
Appelbaum (1979 and 1982), Bresnahan (1982 and 1989),
Gollop and Roberts (1979), and Roberts (1984). Applications
to the banking industry include Spiller and Favaro (1984),
Gelfand and Spiller (1987), Berg and Kim (1994 and 1996),
Shaffer (1989 and 1993a), and Shaffer and Di Salvo (1994).

Economic Perspectives

9Prager and Hannan (1998) examine a cross-section of such
markets, finding that hanks operating in markets where a
merger produces a substantial increase in concentration have

deposit rates that are lower than those set by banks not operat­
ing in such markets. They interpret the result as evidence that
these mergers lead to increased market power.

REFERENCES

Amel, D., 1997, “Antitrust policy in banking: Current

Berger, A., A. Saunders, J. Scalise, and G. Udell,

status and future prospects,” in Proceedings of a
Conference on Bank Structure and Competition,
Federal Reserve Bank of Chicago, pp. 166-179.

1997, “The effects of bank mergers and acquisitions
on small business lending,” in Proceedings of a
Conference on Bank Structure and Competition,
Federal Reserve Bank of Chicago, pp. 241-255.

Akhavein, J., A. Berger, and D. Humphrey, 1997, “The

effects of megamergers on efficiency and prices: Evi­
dence from a bank profit function,” Review of Industri­
al Organization, Vol. 12, No. 1, February, pp. 95-139.

Bingaman, A., 1995, “Antitrust and banking,” paper
presented at the Conference on Antitrust and Banking,
Office ofthe Comptroller ofthe Currency, November 16.

Angelini, P., and N. Cetorelli, 1998, “Assessing

Bresnahan, T. F., 1989, “Empirical studies of indus­
tries with market power” in Handbook ofIndustrial
Organization, R. Schmalensee andR. D. Willig (eds.),
Amsterdam: NorthHolland, pp. 1011-1057.

competition in the Italian banking industry,” Federal
Reserve Bank of Chicago, mimeo.
Appelbaum, E., 1982, “The estimation of the degree of

oligopoly,” Journal ofEconometrics, Vol. 19, August,
pp. 287-299.

________ , 1982, “The oligopoly solution is identified,”
Economics Letters, Vol. 10, pp. 87-92.

_______ , 1979, “Testing price taking behavior,” Journal
ofEconometrics, Vol. 9, February, pp. 283-294.

Bureau of National Affairs, Inc., The, 1992, Daily
Reportfor Executives, Special Supplement, April 3.

Avery, R., R. Bostic, P. Calem, and G. Canner, 1997,

_________ , 1984, Daily Reportfor Executives,
Special Supplement, June 15.

“Changes in the distribution of banking offices,”
Federal Reserve Bulletin, September, pp. 707-725.
Bain, J., 1951, “Relation of the profit rate to industry

Carlton, D., 1986, “The rigidity of prices,” American
Economic Review, Vol. 76, No. 4, pp. 637-658.

concentration: American manufacturing, 1936-1940,”
Quarterly Journal ofEconomics, Vol. 65, pp. 293-324.

Carlton, D., and J. Perloff, 1989, Modern Industrial

Berg, S. A., and M. Kim, 1996, “Banks as multioutput

Organization, Glenview, IL: Scott Foresman/Little,
Brown.

oligopolies: An empirical evaluation of the retail and
corporate banking markets,” in Proceedings of a
Conference on Bank Structure and Competition,
Federal Reserve Bank of Chicago, pp. 183-201.
________ , 1994, “Oligopolistic interdependence and
the structure of production in banking: An empirical
evaluation,” Journal of Money, Credit, and Banking,
Vol. 26, No. 2, pp. 309-322.

Cornett, M., and H. Tehranian, 1992, “Changes in

corporate performance associated with bank acquisi­
tions,” Journal of Financial Economics, Vol. 31,
April, pp. 211-234.
Cowling, K., and VI. Waterson, 1976, “Price-cost
margins and market structure,” Economica, Vol. 43,
August, pp. 267-274.

tration relationship in banking,” Review ofEconomics
and Statistics, Vol. 71, May, pp. 291-299.

DeYoung, R., 1997, “Bank mergers, X-efficiency and
the market for corporate control,” Managerial Finance,
Vol. 23, No. 1, pp. 32 47.

Berger, A., and D. Humphrey, 1992, “Megamergers in

Di Salvo, J., 1997, “Federal Reserve market defini­

banking and the use of cost efficiency as an antitrust
defense,” Antitrust Bulletin, Fall, pp. 541-600.

tions methodology by District,” Casework Conference,
Federal Reserve Bank of Philadelphia, October
23-24.

Berger, A., andT. Hannan, 1989, “The price-concen­

Federal Reserve Bank of Chicago

13

Gelfand, M., and Spiller, P., 1987, “Entry barriers and
multiproduct oligopolies,” International Journal of
Industrial Organization, Vol. 5, March, pp. 101-113.
Gollop, F., and M. Roberts, 1979, “Firm interdependence
in oligopolistic markets,” Journal ofEconometrics,
Vol. 10, August, pp. 310-331.
Hannan, T., 1997, “Market share inequality, the num­
ber of competitors, and the HH1: An examination of
bank pricing,” Review of Industrial Organization,
Vol. 12, February, pp. 23-35.
Hannan, T., and A. Berger, 1991, “The rigidity of prices:
Evidence from the banking industry,” American
Economic Review, Vol. 81, No. 4, pp. 938-945.

Kravitz, P., 1997, “Antitrust policy in banking:

Comments,” in Proceedings of a Conference on
Bank Structure and Competition, Federal Reserve
Bank of Chicago, pp. 180-183.
Kwast, M., M. Starr-McCluer, and J. Wolken, 1997,

“Market definition and the analysis of antitrust in
banking,” Antitrust Bulletin, Winter, pp. 973-995.
Litan, R., 1994, “The ABC of Justice’s antitrust

assessment of bank acquisitions,” Banking Policy
Report, Vol. 13, No. 10, pp. 16-20.
Neumark, D., and S. Sharpe, 1992, “Market structure

and the nature of price rigidity: Evidence from the
market for consumer deposits,” Quarterly Journal
ofEconomics, Vol. 107, May, pp. 656-680.

Hannan, T., and J. Liang, 1993, “Inferring market

power from time-series data. The case of the banking
firm,” International Journal of Industrial Organiza­
tion, Vol. 11, June, pp. 205-218.
Holder, C., 1993 a, “The use of mitigating factors in
bank mergers and acquisitions: A decade of antitrust
at the Fed,” Economic Review, Federal Reserve Bank
of Atlanta, March-April, pp. 32-44.

_________ , 1993b, “Competitive considerations in
bank mergers and acquisitions: Economic theory, legal
foundations, and the Fed,” Economic Review, Federal
Reserve Bank of Atlanta, January-February, pp. 23-33.
Hughes, J., W. Lang, L. Mester, and C. Moon, 1996,

“Efficient banking under interstate branching,”
Journal ofMoney, Credit, and Banking, Vol. 28,
No. 4, Part 2, pp. 1045-1071.
Iwata, G., 1974, “Measurement of conjectural varia­
tions in oligopoly,” Econometrica, Vol. 42, September,
pp. 947-966.

Jackson, W., Hl, 1997, “Market structure and the

speed of adjustment: Evidence of non-monotonicity,”
Review ofIndustrial Organization, Vol. 12, February,
pp. 37-57.

________ , 1992a, “Is the market well defined in
bank merger and acquisition analysis?,” Review of
Economics and Statistics, Vol. 74, November, pp.
655-661.
________ , 1992b, “The price-concentration relation­
ship in banking: A comment,” Review ofEconomics
and Statistics, Vol. 74, May, pp. 373-376.

14

Peristiani, S., 1995, “Do mergers improve the X-efFi-

ciency and scale efficiency of U.S. banks? Evidence
from the 1980s,” Federal Reserve Bank of New York,
working paper.
Pilloff, S., 1996, “Performance changes and share­

holder wealth creation associated with mergers of
publicly traded banking institutions,” Journal of
Money, Credit, and Banking, Vol. 28, No. 3, pp.
294-310.
Prager, R., and T. Hannan, 1998, “Do substantial hor­

izontal mergers generate significant price effects?
Evidence from the banking industry,” Journal of
Industrial Economics, December.
Radecki, L., 1997, “The expanding geographical

reach of retail banking markets,” Federal Reserve
Bank of New York, mimeo.
Rhoades, S. A., 1995a, “Market share inequality, the
HHI, and other measures of the firm composition of a
market,” Review ofIndustrial Organization, Vol. 10,
No. 6, pp. 657-674.

________ , 1995b, “Competition and bank mergers:
Directions for analysis from available evidence,”
Board of Governors of the Federal Reserve System,
mimeo.
_________, 1993a, “The efficiency effects of horizon­
tal bank mergers,” Journal of Banking and Finance,
Vol. 17, April, pp. 411-422.
________ , 1993b, “The Herfindahl-Hirschman
Index,” Federal Reserve Bulletin, Vol. 79, March,
pp. 188-189.

Economic Perspectives

Roberts, M., 1984, “Testing oligopolistic behavior,”
International. Journal of Industrial Organization,
Vol, 2, December, pp, 367 383

Simmons, K., and J. Stavins, 1998, “Has antitrust
policy in banking become obsolete9,” New England
Economic Review, March/April, pp. 13-26.

Shaffer, S., 1993a, “A test of competition in Canadian
banking,” Journal of Money, Credit, and Banking,
Vol. 25, No. l,pp. 49-61.

Spiller, P., and E. Favaro, 1984, “The effects of entry
regulation on oligopolistic interactions: The Uruguayan
banking sector,” The Rand Journal of Economics,
Vol. 9, No. 2, pp. 305-327.

________ , 1993b, “Can megamergers improve bank
efficiency?,” Journal of Banking and Finance, Vol.
17, April, pp. 423 436. '
________ , 1989, “Competition in the U.S. banking
industry,” Economics Letters, Vol. 29, No. 4, pp.
321-323.
_________ , 1983, “Non-struetural measures of
competition: Toward a synthesis of alternatives,”
Economics Letters, Nq\. 12, pp. 349-353.
Shaffer, S., and J. Di Salvo, 1994, “Conduct in a
banking duopoly,” Journal of Banking and Finance,
Vol. 18, December, pp. 1063-1082.

Federal Reserve Bank of Chicago

Suominen, M., 1994, “Measuring competition in
banking: A two product model,” Scandinavian
Journal ofEconomics, Vol. 96, No. 1, pp. 95-110.
Whalen, G., 1995, “Non-local concentration, multimar­

ket linkages and interstate banking,” paper presented
at the Conference on Antitrust and Banking, Office of
the Comptroller ofthe Currency, November 16.
Wolken, J., 1984, “Geographic market delineation: A
review of the literature,” Board of Governors of the
Federal Reserve System, staff study, No. 140, pp. 1-38.

15

The ffth Annual Conference on Bank Structure and Competition

GLOBAL FINANCIAL CRISES
Implications for Banking and Regulation

May^-y, Z999

16

Economic Perspectives

The Federal Reserve Bank

of

Chicago

■ A luncheon address by John B. McCoy, President

will hold its 35th annual Conference on Bank Structure

and Chief Executive Officer of Bank One Corporation,

and Competition at the Westin Hotel in Chicago,

the nation’s fifth largest bank holding company re­

May 5—7, 1999. Attended each year by several hundred

sulting from the 1998 merger of Bank One and First

financial institution executives, regulators, and academics,

Chicago/NBD Corporation.

the goal of the conference has been to generate an

■ A panel discussion on the need for regulatory reform

ongoing dialogue on public policy issues affecting the

and alternative perspectives on the appropriate form

financial services sector.

The major theme for the 1999 conference will be an

this should take.
■ A discussion of the potential benefits and problems

analysis of financial crises. What appears to have started

associated with bank mergers and reasons for the

as a currency devaluation for a relatively small Asian

apparent lack of evidence supporting the contention

country has seemingly led to a general upheaval in financial

that significant cost and efficiency gains should result

markets throughout the world. Suddenly, global capital

from mergers.

market integration, which had been credited with pro­

■ A discussion of the appropriate response to financial

longing and enhancing economic growth through the

crises emphasizing the role of the lender of last resort

1990s, has been criticized as a major cause of the crisis.

and the impact of debt forgiveness.

What are the implications of these crises for U.S. firms?

What are the appropriate public policy responses?

As usual, the Wednesday sessions will showcase

a wide array of technical research papers of primary

Finally, what are the most effective means to prevent

mterest to researchers in academia and government. The

such crises—including an evaluation of the need for a

Thursday and Friday sessions are designed to discuss

new regulatory architecture?

issues that focus on the interests of a broader audience.

These events raise a number of issues for bankers—

If you currently are not on our mailing list or have

both large and small—as well as for alternative providers

changed your address and would like to receive an in­

of credit, and public policymakers. Many of these issues

vitation to the conference please contact:

will be discussed at the 1999 conference. As in past years,
much of the program will be devoted to the primary

theme of the conference, but there also will be a number

of additional sessions on topical issues of financial
structure and regulation. Some highlights of this year’s

scheduled program include:
■ The keynote address by Federal Reserve Board
Chairman Alan Greenspan.

Portia Jackson

Conference on Bank Structure and Competition

Research Department
Federal Reserve Bank of Chicago
P.O. Box 834
Chicago, Illinois 60604-1413

Telephone: 312-322-5115

e-mail: portia.jackson@chifrh.org

■ A discussion of the conference theme by a panel of

industry experts. The participants scheduled to attend
include Carter Golembe from CHG Consulting Inc.,

William McDonough, President of the Federal Reserve
Bank of New York, John Heimann from Merrill

Lynch and recently appointed to chair the Financial
Stability Institute of the Bank for International Settle­
ments, Andrew Sheng from the Hong Kong Securities

and Futures Commission, and Allan H. Meltzer from
Carnegie Mellon University.

FEDERAL RESERVE BANK
OF CHICAGO

Agglomeration in the U.S. auto supplier industry

Thomas H. Klier

Introduction and summary
The General Motors (GM) strike during June and
July 1998 showed the extent to which lean manufac­
turing production methods, such as efforts to keep
inventories low and reduce the number of parts sup­
pliers, have taken hold in the U.S. auto sector. As
observers tried to assess the ramifications of this
event, it became apparent that we know much more
about the spatial structure of light vehicle assembly
operations and Big Three (Ford, GM, and Chrysler)
owned parts plants than of the large number of inde­
pendent parts suppliers. In an environment of tightly
linked supply chains, it is important to understand
the spatial nature of these linkages. Such knowledge
would help policymakers assess the economic impact
of regional shocks, such as a strike. In addition, data
on individual customer-supplier linkages would facil­
itate the study of the geographic extension of supplier
networks and offer new evidence on the ability of
economic development efforts to attract suppliers
to locate in the same state as a large assembly facility.
Lean manufacturing was pioneered by Toyota
Motor Company in Japan during the 1950s. It has
since become the standard for many manufacturing
companies in Japan and around the world. This pro­
duction system tries to improve on the types of mass
production systems that have been prominent in the
postwar period. Instead of organizing production
according to a preset schedule, it operates on the
premise of a so-called pull system, whereby the flow
of materials and products through the various stages
of production is triggered by the customer. In addition,
the production process itself is subject to continuous
improvement efforts.
The 1998 strike at two GM-owned parts plants in
Flint, Michigan, was about issues related to produc­
tion rates and health and safety. Strategically, however,
it centered on issues pertinent to the implementation of
new production methods—more efficient production
processes that would reduce the demand for labor in

18

the assembly plant and efforts by the assembly com­
pany to outsource more of the production of parts.
The strike quickly shut down most of GM’s North
American assembly operation. In turn, it caused pro­
duction adjustments at many of the company’s inde­
pendent suppliers.
In this article, I examine the spatial structure of
the auto supplier industry and how firms in different
locations interact. First, I document the extent to which
plants are concentrated geographically, that is, the de­
gree of spatial agglomeration, in the U.S. auto supplier
industry. My analysis is based on information on the
location of over 3,000 auto supplier plants. I find that
the auto supplier industry is concentrated in five states—
Indiana, Kentucky, Michigan, Ohio, and Tennessee—
that constitute the so-called auto corridor, which is
defined by interstate highways 65 and 75, extending
south from Michigan to Tennessee. These states are
home to 58 percent of the plants in the study. A closer
analysis of plant locations reveals the importance of
access to highway transportation to ensure timely
delivery of production to customers. I find that hav­
ing suppliers located in the immediate vicinity of the
assembly plant is not necessary to maintain a system
of tight linkages and low inventories. Comparing the
spatial structure of individual assembly networks, I
find them to be remarkably similar. The geographic
concentration is highest for assembly plants that are
located near the heart of the auto corridor, with between
70 percent and 80 percent of supplier plants located
within a day’s drive of the assembly plant. This sug­
gests a clustering of economic activity at the regional
rather than local level.
Thomas H. Klier is a senior economist at the Federal
Reserve Bank of Chicago. The author would like to
thank Jim Rubenstein and Bill Testafor their valuable
suggestions; Neil Murphy and George Simler for
excellent research assistance; and seminar participants
at the Federal Reserve Bank of Chicago for helpful
comments.

Economic Perspectives

Second, I investigate the changing nature of the
geographic concentration of this industry over time.
This analysis is limited by the cross-sectional nature
of the data. However, there are a few cases in which
the data allow a comparison of supplier networks of
different vintages. In addition, 1 apply a location rule
to a subset of all the supplier plants that allows me
to use information on the location of all light vehicle
assembly plants inthe U.S. from 1950 to 1997. Consis­
tently, 1 find evidence of increased clustering in the
auto supplier industry relative to 30 or 40 years ago.

Literature review
Geographic concentration in U.S. manufacturing
has received greater attention in recent years. Krugman
(1991) suggests that Silicon Valley-style agglomera­
tions may be more the rule than the exception and
that we can learn from them about the source of the
underlying forces.1 Ellison and Glaeser (1997) address
the question of how to properly measure industry
concentration over and above the general level of
concentration in manufacturing. To that end they
develop a model that captures both random location
effects and those caused by localized industry-specific
spillovers and natural advantages. The authors develop
indexes of localization and find almost all industries
to be somewhat localized. In many industries, however,
the degree of localization is small. The authors report
that almost all of the most extreme cases of concen­
tration are apparently due to natural advantages.2
Hewings et al. (1998) analyze the 1993 commodity flow
statistics, using a detailed econometric input-output
model, to learn about a slightly different issue: To what
extent are the states of a specific region (the Midwest)
linked economically? They find very strong evidence
of industry clusters at the regional level. For example,
in the case of the auto industry, an initial loss of auto­
motive production in Michigan would create secondary
effects that are heavily concentrated in the Midwest.
Specifically, losses in the Midwest would represent 43
percent of the secondary effect outside of Michigan.
Addressing these issues for the U.S. auto indus­
try, several studies suggest that the assembly plants
for light vehicles have reconcentrated in the Midwest
and Southeast since the mid-1970s (Rubenstein, 1992;
McAlindenand Smith, 1993; and Rubenstein, 1997).
Rubenstein (1997) attributes this to the demise of the
branch plant assembly system, whereby identical
models were produced around the country at assembly
plants that were located close to population centers.
Developments in the supplier industry are not as
clear cut. Apparently there has been a migration of
especially labor-intensive parts production to the
southern U.S. and south of the border; however,

Federal Reserve Bank of Chicago

parts requiring highly skilled labor, such as engines,
transmissions, and large stampings, have remained
heavily concentrated in the Midwest. That is espe­
cially true for parts plants operated by the auto as­
semblers themselves (so-called captive suppliers)
(see table 1).
As for the potential location effect of lean manu­
facturing, the prevailing anecdotal evidence suggests
that the application of lean manufacturing techniques
has resulted in a tiering and consolidation of the sup­
plier base of the auto industry, as well as a higher
degree of communication and interaction between
suppliers and assemblers (Helper, 1991). Has this
resulted in tighter geographical linkages between
assembler and supplier plants? Proponents suggest
that close linkages work most effectively when sup­
plying and receiving plants are in reasonably close
proximity (Estall, 1985; Kenney and Florida, 1992;
Mair, 1992; and Dyer, 1994). However, there is also
evidence that spatial clustering is not a necessary
outcome of lean manufacturing applications. What
ultimately matters is the quality of transportation
infrastructure in combination with the capability of
delivery management systems in ensuring predictable
on-time arrival of goods. This might well be achieved
with no significant increase in clustering at the indus­
try level.
A set of studies specifically investigates the exist­
ence of effects of lean manufacturing on the spatial
structure of the auto supplier industry. Rubenstein
and Reid (1987) and Rubenstein (1988) analyze the
changing supplier distribution of U.S. motor vehicle
parts suppliers. Their thorough analysis of supplier
plants located in Ohio cannot establish a clear-cut
effect of lean manufacturing on plant location, yet the
authors find evidence of a change in the locational
pattern after 1970.
TABLE 1

Distribution of captive parts plants
(percent)

Share of captive suppliers
in Ml, IN.andOH
Assembly company

Plants

Employees

General Motors
Chrysler
Ford

69.8
82.3
84.6

73.8
86.9
85.5

Overall

75.6

77.6

Source: ELM International, Inc., 1997, "The ELM GUIDE
supplier database," East Lansing, Ml, database file, and
author's calculations.

19

Most of the existing analysis of the location effects
of lean manufacturing, however, concerns Japaneseowned manufacturing establishments within the U.S.
This is not surprising, as these plants generally apply
lean manufacturing. In addition, most of them repre­
sent new plants established at newly developed,
so-called greenfield sites. As their location decision
usually does not involve a re-location, they are a pre­
ferred object of study. Woodward (1992) investigates
what determines the location of Japanese manufactur­
ing start-up plants in the U.S. The author estimates
location models of the spatial behavior of Japaneseaffiliated manufacturing investments undertaken be­
tween 1980 and 1989. While his observations include
plants from many different manufacturing industries,
he estimates a model specification at the county level
for 250 observations in the Michigan-Tennessee auto­
motive corridor. Woodward finds proximity to urban
areas not to be important for these plants; however, an
interstate connection linking counties to major markets
appears to be crucial. Reid (1994) tests the effect of
just-in-time inventory control on spatial clustering in
observing the level of inputs purchased locally for a
set of 239 Japanese-owned manufacturing plants in
the U.S. The author performs this analysis at three
different levels of aggregation—county, state, and
national. He finds differences in the proportion of
material inputs purchased locally between plants that
use just-in-time inventory control and those that do
not only at the county level. This result suggests
spatial clustering effects on a very local scale. Smith
and Florida (1994) test for the existence of agglomera­
tion effects in the location decisions of over 400
Japanese-affiliated manufacturing establishments in
automotive-related industries. They perform a formal
analysis for all U.S. counties, as well as an automotive
corridor subset, and find that Japanese-affiliated sup­
pliers prefer to locate in close proximity to Japanese
automotive assemblers. On a regional scale, they find
a concentration of Japanese auto suppliers in the
auto corridor.

Spatial characteristics data
In this article, I present evidence on the spatial
characteristics of independent auto supplier plants
located in the U. S., with particular emphasis on link­
ages between supplier and assembly plants. First, I
document the extent to which plants are concentrated
geographically, or the degree of spatial agglomeration,
in the U.S. auto supplier industry. Second, I investigate
the changing nature of this geographic concentration
over time.

20

Publicly available data do not provide this level
of detail. The obvious data source, the Census of
Manufactures, can offer only incomplete information,
because it does not distinguish between original equip­
ment manufacturers and producers of replacement
parts. In addition, because of the large variety of parts
that make up an automobile, supplier plants in the auto
industry are classified among 18 of the 20 two-digit
standard industry classification (SIC) codes. Finally,
Census data cannot establish information about link­
ages between supplier plants and their customers.
The basis for my analysis is the “ELM GUIDE
supplier database,” a set of plant-level data on the
auto supplier industry put together by a private com­
pany in Michigan.3 The data are for 1997 and cover
3,425 independent supplier plants in the U.S.4 As the
database identifies customers for the individual suppli­
er plants, I was able to categorize these plants by sup­
plier tier: 2,008 plants are tier 1 suppliers, that is,
supplier plants that ship their products exclusively
to auto
assembly plants and not to other supplier plants or
other customers; 1,292 are mixed-tier suppliers, that
is, in addition to auto assembly plants, their custom­
ers include other supplier plants and/or nonautomotive assemblers; and 50 observations were excluded
from the analysis because they did not provide infor­
mation on their customers.5
I then added several variables to the database.
For tier 1 plants, I obtained start-up year data from
various state manufacturing directories and phone
calls to individual plants. I added information on foreign
ownership available through industry press reports and
the Japan Auto Parts Industries Association.6 Table 2
shows an ownership breakdown of the industry.
Accounting for incomplete information on start-up
year, I end up with 1,845 individual plant records, rep­
resenting independent tier 1 supplier plants opera­
tional in 1997.7 Next, I analyze data on these 1,845 plants
to test for agglomeration at the industry level, as these
plants arguably represent the subset of supplier plants
that is most closely linked to the auto assembly plants
by way of production and delivery. In addition to the
cross-sectional comparisons, information on the vin­
tage of individual plants allows some comparison of
location patterns of older and recently opened plants.8
The analysis of assembly plant-specific networks draws
on all the 3,137 records of independent supplier plants.9

Industry-level agglomeration
Table 3 presents the distribution of the 3,137
independent supplier plants included in the database.

Economic Perspectives

TABLE 2

TABLE 3

Auto suppliers by ownership, 1997

Distribution of auto suppliers, 1997

(percent)

(percent)

Domestic

Plants

Employment

84.7

81.6

9.6
5.7

11.2
7.2

Foreign-owned
Japanese
Other

Notes: Calculations are based on 3,137 independent
supplier plants open in 1997; numbers do not include
captive supplier plants. Industry employment: 901,343 jobs.
Source: See table 1.

It shows the auto supplier plants and employment to
be highly spatially concentrated, with almost 50 per­
cent of all plants located in just three states—Michi­
gan, Ohio, and Indiana. However, it is important to
keep in mind that this information represents plants
from rather different vintages. For example, the oldest
plants in the sample date from the nineteenth century;
38 plants opened prior to 1900. In order to get abet­
ter read on recent plant location choices, 1 focus on
the subset of supplier plants that have opened since
1980, marking when lean manufacturing arrived in the
U.S.10 As data on the establishment year are available
only for tier 1 supplier plants, 1 focus on the subset
of 820 tier 1 supplier plants that opened in 1980 or
after and were still in operation in 1997. While a pure
cross-sectional data set prevents me from testing for
changes in location patterns over time, concentrating
on plants of recent vintage enables me to present the
location choices in a lean manufacturing environment
in much more detail.
Figure 1 shows the plants that opened between
1980 and 1997 and their concentration among the five
states of the auto corridor. Domestic plants are shown
in black, foreign-owned plants in color. A circle indi­
cates that two or more plants are located within one
zip code. In addition, stars mark the location of light
vehicle assembly plants in operation at any point
during this period. One can clearly see that plant
openings are highly clustered in a north-south direc­
tion (in southern Michigan and the four states to the
south). Figure 2 adds the grid of interstate highways
to the pattern of plant openings. This exercise dem­
onstrates the relevance of the 1-65/1-75 corridor.11
Note, however, that interstate access plays an impor­
tant role for east-west connectivity as well. For
example, Toyota operates a car assembly plant in
Georgetown, Kentucky, a recently opened light truck

Federal Reserve Bank of Chicago

State

Plants

Employment

Illinois
Indiana
Kentucky
Michigan
Ohio
Tennessee
Wisconsin

6.9
9.1
4.0
26.8
13.2
4.7
3.6

6.8
10.1
4.1
19.2
11.2
5.8
3.1

59.6
57.8
100.0

50.4
50.4
100.0

Midwest
Auto corridor
U.S.

Notes: Calculations are based on 3,137 independent
supplier plants open in 1997; numbers do not include
captive supplier plants. Industry employment: 901,343 jobs.
The auto corridor comprises Indiana, Michigan, Ohio,
Kentucky, and Tennessee. The Midwest comprises Illinois,
Indiana, Michigan, Ohio, and Wisconsin.
Source: See table 1.

assembly plant in Princeton, Indiana, and an engine
plant in Buffalo, West Virginia. All three of these are
linked by Interstate 64, highlighting the importance
of highway access to ensure timely delivery of ship­
ments in an environment of just-in-time production.
Looking at the auto corridor locations more
closely, figure 3 (page 23) reveals a different location
pattern for domestic and foreign-owned supplier
plants during 1980-97.12 While they are similarly con­
centrated among three states, foreign-owned suppliers
choose to locate in the southern part of the automo­
tive corridor (that is, Ohio, Kentucky, and Tennessee).
Domestic suppliers, on the other hand, locate in the
northern part, with Ohio being the only state chosen
prominently by both domestic and transplant suppli­
er plants.13 Does this indicate that the auto corridor is
a phenomenon driven by the location of foreignowned plants? What explains the apparent different
spatial pattern in plant locations? Do foreign-owned
suppliers respond differently to lean manufacturing
conditions than domestic suppliers? Figure 3 and ta­
ble 4 (page 23) suggest a different explanation: The
difference in the spatial distribution of domestic and
foreign-owned assembly plants seems to dominate
the location choices of supplier plants.14 As a rule of
thumb, between 1980 and 1993 supplier plants located
close to assembly plants of the same nationality.15
This can be seen in figure 3, which distinguishes
between domestic (gray) and foreign-owned (col­
ored) assembly plants.

21

FIGURE 1

Plant openings by tier 1 suppliers, 1980-97

► 1 foreign-owned supplier
• 2 or more foreign-owned suppliers

► 1 domestic supplier
• 2 or more domestic suppliers

• Assembly plant
Source: See table 1.

22

Economic Perspectives

FIGURE 3

Plant openings by tier 1 suppliers within auto corridor, 1980-97

► 1 foreign-owned supplier
• 2 or more foreign-owned suppliers
► 1 domestic supplier

• 2 or more domestic suppliers

★ Domestic assembly plant
★ Foreign-owned assembly plant

Focusing on relationships to primary customers
only would provide more conclusive evidence. How­
ever, the data do not allow identification of the distri­
bution of output among customers. Instead, 1 present
information on the distribution of supplier plants that
report a particular customer mix. Table 4 shows data
on domestic suppliers that supply only to Big Three

assembly plants, as well as data on Japanese trans­
plant suppliers that do not supply to any Big Three
assembly plants. If the nationality of the assembly
plant customer was important to the location choice
of the supplier plant, one would expect these two
groups to be relatively concentrated in their respective
halves of the auto corridor. Table 4 provides evidence

TABLE 4

Percent of supplier plants opened in auto corridor, 1980-97

Domestic

Japanese-owned

Overall

Notsupplying
to Big Three

20.8
18.5
13.3
11.0
9.2

28.6
21.4
14.3
12.9
0.0

61.6

52.6

64.3

166

173

70

Overall

Supplying only
to Big Three

Michigan
Indiana
Ohio
Tennessee
Kentucky

31.3
10.9
10.4
6.3
4.1

40.0
11.4
10.2
4.2
1.8

Top three

52.6

Numberofplants

607

Ohio
Kentucky
Tennessee
Indiana
Michigan

Source: See table 1.

Federal Reserve Bank of Chicago

23

of just such a “customer” effect, as each group of
supplier plants with a specific customer mix is more
concentrated at one end of the auto corridor.16
This simple comparison between the location
choices of assembly and supplier plants, however,
cannot address the issue of timing. Did assembly or
supplier plants locate first?17 The data allow me to shed
some light on this question for the Japanese-owned
supplier plants. Table 5 shows that 55 percent of
these plants opened between 1987 and 1989, well after
the first Japanese auto assembly plants had started
operating in the U.S.18 That pattern suggests that in
the case of Japanese transplants, the suppliers followed
the assemblers (see also Rubenstein, 1992). However,
the initial location decision of Japanese assembly
plants was undoubtedly influenced by proximity to
the existing, that is, mostly domestic, supplier base.19
Network data
Next, I discuss the extent to which supplier plants
locate near their assembly plant customers. As the
data set includes information on customers of the
individual supplier plants, I am able to construct
supplier networks for specific assembly plants.20 How­
ever, my choice of assembly plants is limited to a set
TABLE 5

Japanese transplant tier 1 supplier plants
Start-up year

Number of
plants

1980

5

1981

1

1

1982

3

2

3

1983

1

1

1984

5

3

1985

8

5

1986

17

10

1987

34

20

1988

36

21

1989

24

14

1990

9

5

1991

5

3

1992

4

2

1993

2

1

1994

4

2

1995

9

5

1996

2

1

1997

4

2

173

100

Note: Column labeled "Percent" may not total
due to rounding.
Source: See table 1.

24

Percent

of essentially single-plant assembly companies as the
supplier plants’ customer information is provided
only at the company level. I can construct networks
for the following assembly plants: Honda of America,
which opened its Marysville, Ohio, plant in 1982 (and
added a second assembly plant in nearby East Liberty,
Ohio, in 1989); Nissan, which opened an assembly
plant in Smyrna, Tennessee, in 1983; NUMMI, ajoint
venture between Toyota and GM, operating in Fremont,
California, since 1984; AutoAlliance, which started as
ajoint venture between Ford and Mazda in 1987 in
Flat Rock, Michigan; Diamond-Star, which started
production as a Mitsubishi-Chrysler joint venture in
Normal, Illinois, in 1988; Saturn, GM’s attempt to cap­
ture the efficiencies of lean manufacturing, which
started production in 1990 in Spring Hill, Tennessee;
BMW, which opened an assembly plant in South
Carolina in 1994; and Mercedes-Benz, which opened
a plant in Alabama in 1997.
Table 6 presents characteristics of the networks
identified from the database.21 Each network includes
all independent supplier plants that list the respective
assembler as a customer. Not surprisingly, the net­
works vary in size, with Honda, the oldest, being the
largest, and Mercedes-Benz, the most recently opened
assembly plant on the list, the smallest. To measure
the networks’ spatial characteristics, I calculate the
median distance between supplier and assembler
and the percentage of suppliers located within both
a 100-mile and a 400-mile radius of the assembly plant
(table 6, column seven, ranks networks by percentage
share of suppliers within 400 miles). The 400-mile radius
roughly defines the boundary for a one-day shipping
distance, while the 100-mile distance captures plants
that locate close enough to allow multiple deliveries
using the same truck.22
According to these statistics, the individual net­
works look more alike than different. In general, the
spatial concentration increases toward the heart of
the automotive corridor. The AutoAlliance and Honda
networks are most concentrated within 100 miles
(column five); for the 400-mile criterion, the disadvan­
tage from being located at the fringe of the automotive
corridor mostly disappears. Two cases in point are
the Diamond-Star and Subaru-Isuzu networks, which
are, for the larger radius, essentially as concentrated
as Honda’s and Toyota’s. The spatial features of sup­
plier networks reported in table 6 seem to be explained
by two factors: where the assembly plant is located
relative to the auto corridor and whether the assembly
plant is domestic or foreign-owned.

Economic Perspectives

TABLE 6

Spatial characteristics of supplier nletworks, 1997

Assembly
company

Network

Industry

Network

Industry

%<100
miles

%<100
miles

%<400
miles

%<400
miles

77
76
76
72
65
45
42
35
34
11

74.8
75.7
71.8
69.3
66.4
36.7
26.6
32.4
17.5
2.4

Start-up
year

Numberof
suppliers

%
Domestic

Median
distance

A.
Honda
Toyota
Subaru-lsuzu
Diamond-Star
AutoAlliance
Nissan
BMW
Saturn
Mercedes-Benz
NUMMI

1982
1988
1987
1988
1987
1983
1994
1990
1997
1984

507
452
292
286
360
460
119
300
77
178

65
69
60
63
71
70
75
81
68
60

251
285
245
309
242
423
477
462
610
1,966

17
10
9
5
29
10
20
8
8
6

B.
Flint (1950)
Ford (1970-80)
Ford (1983-93)

1907
NA.
NA.

126
222
301

72
89
77

192
405
200

28
18
31

9.4
4.2
6.2
1.7
24.7
3.8
3.7
3.4
0.8
0.8

77
55
66

Note: N.A. indicates not applicable.
Sources: G. Rex Henrickson, 1951, Trends in the Geographic Distribution of Suppliers of Some Basically Important
Materials Used at the Buick Motor Division, Flint, Michigan, Ann Arbor, Ml: University of Michigan, Institute for
Human Adjustment; ELM International, Inc., 1997, "The ELM GUIDE supplier database," East Lansing, Ml,
database file; and author's calculations.

For example, figure 4 shows how Honda’s inde­
pendent supplier plants cluster around its two Ohio
assembly plants. The three circles envelop the first
three quartiles of the distance distribution of supplier
plants in the network. The figure shows an assembly
operation that is centrally located in the auto corridor.
It turns out to be the most spatially concentrated net­
work: 17 percent ofHonda’s 507 suppliers are located
within 100 miles and 77 percent within 400 miles of
the assembly plant.
In contrast, Diamond-Star is located at the west­
ern edge of the auto corridor (see figure 5). There­
fore, it is able to attract only 5 percent of its suppliers
to locate within 100 miles. However, that disadvan­
tage disappears at the 400-mile radius, which, for
Diamond-Star as for Honda, includes 77 percent of
its supplier plants.
The case of Saturn presents yet a different picture.
Its suppliers are relatively dispersed (see figure 6).
Notice the large diameter of the first quartile. Only 35
percent of Saturn’s supplier plants are operating within
400 miles of Spring Hill, Tennessee. This reflects the
fact that Saturn most strongly relies on domestic sup­
pliers, which are located at the northern end of the auto

Federal Reserve Bank of Chicago

region. Its assembly plant, however, is located at the
southern end of the corridor.
Alternatively, one can analyze the concentration
of individual supplier networks relative to the distri­
bution of all the supplier plants. In calculating what
share of the entire industry is located within a certain
radius of the assembly plant, one can then assess a
network’s degree of concentration relative to the indus­
try baseline. Table 6, panel A, shows this information
for both the 100-mile and the 400-mile radius. Columns
five and six show that for every single assembly plant
analyzed, a greater share of suppliers is concentrated
within 100 miles than the overall industry distribution
would suggest. At the 400-mile radius (see columns
seven and eight of table 6), one can distinguish two
network groups. The supplier networks of assemblers
located in the northern end of the auto corridor plus
Kentucky represent very closely the industry’s over­
all spatial distribution. However, the five assembly
plants located in Tennessee, Alabama, South Carolina,
and California are far more concentrated than the indus­
try, even at that relatively large radius. What drives
that result is the large number of suppliers operating
at the northern end of the auto corridor. For example,

25

FIGURE 4

Note: The circles around the assembly plant indicate the closest 25 percent, 50 percent, and 75 percent of the supplier network, respectively.
Source: See table 1.

suppliers in Nissan’s network that are located within
400 miles of the Tennessee assembly plant represent
a far greater concentration of auto suppliers in the
region than indicated by the distribution of all sup­
plier plants.
The different spatial distribution of domestic and
foreign-owned supplier plants across the auto corridor
is reflected within the individual networks as well. For­
eign-owned supplier plants are clustered much more
densely around Japanese assembly plants than domes­
tic suppliers (see, for example, Honda, Toyota, and
Nissan in table 7 on page 29). Yet even for that
group, less than one-third of suppliers are located
within a two-hour drive, or 100 miles, of the assembly
plant. This represents a considerably smaller degree of
spatial concentration within lean manufacturing than
previously reported in the literature.23 The case of
Saturn represents a domestic auto assembler whose
network is not very spatially concentrated. This ap­
plies to both its domestic and foreign-owned supplier
plants (quite in contrast to Nissan, which is located not
very far from Saturn). Finally, AutoAlliance shows the

26

effect ofbeing located in the heart ofthe traditional U.S.
auto region. Its network includes by far the largest per­
centage of suppliers within a 100-mile radius. At 31.9
percent, that share is significantly higher for domestic
suppliers than for foreign-owned suppliers (21.9 percent).
The analysis of the regional concentration of
supplier networks at that disaggregate level can again
be complemented by a comparison with the industry
level of spatial concentration. For the 100-mile radius,
table 7 (columns three and four) shows a higher degree
of concentration for both domestic and foreign-owned
suppliers within each network than is indicated by the
overall distribution ofthe industry. At the 400-mile ra­
dius (columns five and six), the differences between
these two measures of spatial concentration disap­
pear in most cases. Noteworthy exceptions are the
most recently opened assembly plants to the south and
east of the auto corridor (Mercedes and BMW) and
NUMMI. Saturn is the only domestic assembly plant
in the study. Its network shows a smaller percentage
of within-network foreign-owned suppliers within 400
miles of the assembly plant than the overall industry

Economic Perspectives

FIGURE 5

Note: The circles around the assembly plant indicate the closest 25 percent, 50 percent, and 75 percent of the supplier network, respectively.
Source: See table 1.

level would suggest. The spatial distribution of Saturn’s
network presents a stark contrast to that of Nissan,
the other assembly plant in Tennessee.

Changing industry structure?
To what extent are these observations indicative
of changes in the spatial pattern of auto supplier
plants? 1 address that question in three different
ways. First, 1 compare the structure of different net­
works overtime. FromHenrickson’s (1951) analysis
of the supplier structure of the Buick city assembly
plant in Flint, Michigan, it is possible to reconstruct
that assembly plant’s supplier network (see table 6,
panel B, page 25) and compare it with a current net­
work (Honda) that operates based on a different
manufacturing system.24 It turns out that the median
distance is statistically different for these two networks;
however, the percentages within 400 miles are not
statistically different. In other words, during the prime
of the manufacturing system perfected by Henry Ford,
one of its showcase plants, GM’s Buick city plant,

Federal Reserve Bank of Chicago

had a supplier structure that was remarkably spatially
concentrated. However, it is important to keep in
mind that such a comparison is not adjusted for dif­
ferent degrees of vertical integration, changes in the
mode and speed of transportation, as well as quality
of the transportation infrastructure since 1950. In oth­
er words, a 400-mile radius in 1950 in all likelihood
represented a smaller degree of spatial concentration
than the same radius in 1997.
Second, 1 test for differences in spatial concen­
tration for one network over time, using data on one
of the Big Three assemblers, Ford. Instead of construct­
ing networks for each ofFord’s individual assembly
plants, 1 use Dearborn, Michigan, as the center of
Ford’s assembly operations. Since 1970 there have
been two decades, 1970-80 and 1983-93, during
which Ford neither opened nor closed an assembly
plant.25 Juxtaposing these two periods allows for an
interesting comparison of the changing spatial pattern
ofFord’s supplier network (see table 6, panel B on
page 25 and figure 7 on page 30). It shows a marked

27

FIGURE 6

Note: The circles around the assembly plant indicate the closest 25 percent, 50 percent, and 75 percent of the supplier network, respectively.
Source: See table 1.

increase in concentration of Ford’s supplier base
around southern Michigan. During the more recent
decade, 31 percent of newly opened supplier plants
located within 100 miles of Dearborn (versus only 17
percent during the earlier decade). Comparing 1970-80
and 1983-93, the closures of two California plants
and a New Jersey plant in the intervening years might
have reduced average distances to Dearborn somewhat
(for example, by reducing the percentage of plants
greater than 400 miles away). However, one would
not expect that alone to contribute to the simulta­
neous increase in plants located within 100 miles of
Dearborn.26 Comparing 1970-80 and 1983-93, the
statistical tests show all three measures of spatial
concentration reported in table 6 to be different at the
99 percent confidence level, providing strong evidence
of increasing spatial concentration within one of the
Big Three supplier networks.
Third, 1 ignore the customer information provided
by the database and employ a simple location algo­
rithm, motivated by a Weberian model of plant location,
to link suppliers with assembly plants.27 By applying

28

a uniform location rule across time for supplier plants,
I can test whether their location decisions changed
over time. To perform this test, I break the sample into
two periods: plants that have opened since 1980, whose
location decisions were presumably influenced by lean
manufacturing constraints, and plants that opened
between 1950 and 1979, when supplier location deci­
sions were influenced by the need to be close to Big
Three operated parts distribution facilities. Comparing
plant locations for these two samples, 1 can test for a
change in location pattern in two directions. That is, 1
can ask if the pattern exhibited by the younger plants
fits that of the older ones and vice versa. Specifically,
for the most recent period 1 apply two versions of a
location rule that minimizes the distance between
supplier and assembly plant.28 This approach repre­
sents the influence of just-in-time production; supplier
plants in that environment want to be located closer
to the assembly plant to minimize production and
transportation costs. It links the supplier to the closest
operational assembly plant. I do not incorporate infor­
mation provided in the database (and used above)

Economic Perspectives

TABLE 7

Spatial characteristics of supplier networks by supplier type, 1997
Assembly
company

Supplier
type

Network

Industry

Network

Industry

Median
distance

%<100
miles

%<100
miles

%<400
miles

%<400
miles

Honda

Domestic
Foreign

280.6
175.2

12.7a
26.3a

8.6a
13.6a

73.1b
83.4b

74.6
76.2

Toyota

Domestic
Foreign

311.3
199.2

5.1a
19.4a

2.9a
11.3a

73.1b
83.5b

74.8b
80.8b

Subaru-lsuzu

Domestic
Foreign

260.4
206.7

6.8a
12.1a

5.8a
8.8a

69.9b
84.5b

71.5
73.2

Diamond-Star

Domestic
Foreign

333.3
280.3

2.8a
8.5a

1,4a
3.3a

69.3
78.3a

69.3
69.5

AutoAlliance

Domestic
Foreign

187.7
286.4

31.9a
21.9a

26.4a
15.5a

66.9
61.9a

67.5b
60.3b

Nissan

Domestic
Foreign

447.1
272.1

7.1a
16.7a

2.9a
8.6a

36.6b
64.5b

32.4b
60.7b

BMW

Domestic
Foreign

494.6
398.0

18.2a
23.3a

3.3a
6.1a

39.7
50.0a

22.6b
48.5b

Saturn

Domestic
Foreign

465.8
435.9

7.4a
8.9a

2.7a
7.3a

32.8
42.9a

28.2b
55.6b

Mercedes-Benz

Domestic
Foreign

638.7
435.4

5.8a
12.0a

0.7a
1.3a

26.9
48.0a

15.2b
30.8b

NUMMI

Domestic
Foreign

1,946.6
1,975.5

6.6a
7.0a

0.6a
1. 9a

11.3
9.9a

2.1
4.0a

indicates too few observations.
blndicates a difference in the percentages of domestic and foreign-owned suppliers at the 99 percent level of confidence.
Source: See table 1.

on actual assembler-supplier linkages. However, in
applying a general location rule 1 am no longer restricted
to the number of assembly plants listed in table 6, but
can consider all light vehicle assembly plants in the
U.S.29 A slightly different version averages the three
shortest distances between a supplier and operation­
al assembly plants. 1 apply the location rule to both
sets of supplier plants, resulting in a distribution of
distances for each sample. 1 then test if the median of
the more recent sample is statistically different from
the median of distances for the older plants.30 If 1 find
no statistical difference, then the just-in-time location
rule describes both time periods equally well, and
there is no evidence for change in location pattern.
However, if there is evidence of a difference in the
pattern, 1 interpret this as a strong signal for a change
in the location pattern, as it is established by apply­
ing the same decision rule for both periods. The test

Federal Reserve Bank of Chicago

results are described in table 8, panel A (page 31). Under
both versions of the just-in-time rule, median distances
decrease over time. In fact, the differences in the me­
dian are significant at the 99 percent level, according
to a Wilcoxon signed-rank test.
In testing for a change in location pattern in the
opposite direction, 1 use the following rule to approx­
imate decisions made by the older supplier plants:
minimize distance to Detroit.31 Prior to the tiering of
the supplier industry, supplier plants would usually
ship their output to a regional parts distribution cen­
ter operated by the Big Three, which in turn directed
the parts to assembly plants around the country. In
recognition of the spatial clustering of auto supplier
plants in southeast Michigan, northern Indiana, and
Ohio, 1 calculate the distance to Detroit for each plant
that opened during the earlier period. These results

29

FIGURE 7

Increase in concentration of Ford’s supplier
netwo r k over time

«. 1 supplier plant
• 2 or more supplier plants
* Assembly plant
Note: The circles around the assembly plant indicate the closest 25
percent, 50 percent, and 75 percent of the supplier network, respectively.
Source: See table 1.

are presented in panel B of table 8. The actual distances
to Detroit increased from 1980 onward, which is not
surprising considering the changing shape of the auto
region in that period. Again, 1 find the median distances
to be statistically different at the 99 percent level, com­
plementing the result of the first part of the test for a
change in location patterns over time. To summarize, 1
find symmetrical evidence for structural change in the
way supplier plants locate around assembler plants.
Both tests suggest an increase in the clustering of

30

suppliers around assembly plants since
1980 relative to 30 or 40 years ago.

Conclusion
By refining a commercially available
database, this article provides a detailed
look at the supplier networks of some
recently opened auto assembly plants
in the U.S. My analysis focuses on a
description of existing spatial relations
between assembly plants and their tier 1
supplier plants. This study supports
earlier findings about regional agglomer­
ation of supplier plants in the 1-65/1-75
automotive corridor. For supplier plants
of recent vintage, the five auto corridor
states, Michigan, Indiana, Ohio, Kentucky
and Tennessee, represent the preferred
location. Within that region, however,
domestic and foreign-owned supplier
plants locate in noticeably different pat­
terns, apparently due to differences in
the location of domestic and foreignowned assembly plants.
The evidence 1 present on the auto
industry supports the view that agglom­
eration economies play out at the re­
gional level (see Hewings et al., 1998). It
does not support the notion that imme­
diate proximity to the assembly plant is
necessary for operating a system based
on tight linkages and low inventories.32
In analyzing the extent of localization of
production around individual assembly
plants, 1 find networks to be remarkably
similar, with about 70 percent to 80 per­
cent of suppliers located within one day’s
drive of the assembly plant. Differences
seem to be explained by the location of
the assembly plant in relation to the heart
of the auto corridor as well as by nation­
ality of the assembly plant. Within indi­
vidual networks, the spatial concentration differs
across domestic and foreign-owned supplier plants.
This evidence on spatial agglomeration has
relevance for economic development (see table 9).
The economic development literature has generally
reported on the effects of locating a new assembly
plant on either its immediate and surrounding coun­
ties (see, for example, Fournier and Isserman, 1993)
or on the host state (see, for example, Marvel and
Shkurti, 1993). However, the analysis presented here

Economic Perspectives

allows us to investigate the extent of the regional
distribution of related upstream plant employment in
Median distances (miles) between supplier
much
greater detail. Take, for example, the case of the
and assembly plant, 1997
Mercedes plant that opened in 1993 in Alabama. The
Supplier plants opened
state provided incentives worth about $250 million to
1950-80
1980-97
attract that plant. However, the evidence presented
on the spatial extension of supplier networks sug­
A. Just-in-time
gests
that suppliers to Mercedes will locate not just
location rule
Shortest distance
in Alabama, but more likely in Tennessee, Kentucky,
All
60.4a(649)
52.2' (806)
and even further north.33 In fact, to date only 35 per­
Domestic
59.6a (605)
47.1' (594)
cent of Mercedes’s suppliers are located within 400
Closest three avg.
miles
of the assembly plant, and only 16.5 percent of
All
1 08.2a (649)
84.3a(806)
its supplier employment resides in Alabama.34 In Mer­
Domestic
105.0a (605)
73.2a (594)
cedes’ case, attractive targets for location
B. Distance to
efforts seem to have been foreign-owned companies
Detroit rule
(see table 7 on page 29). In short, this type of analysis
All
97.2' (649) 296.7a (806)
Domestic
188.6' (605) 203.0a(604)
suggests that subsidies that are offered by a state
not
in the auto corridor are considerably less effective
“Indicates that the median distances are significantly different
at the 99 percent confidence level, according to a Wilcoxon
in terms of attracting a significant portion of the
signed-rank test.
related supplier employment to that state.
Note: Numbers in parentheses indicate number of
observations.
In the case of Toyota’s Kentucky assembly plant,
Source: See table 1.
a comparison of my network data on the distribution
of supplier jobs with forecasts projected by a 1992
study also suggests a greater degree of
TABLE 9
spatial dispersion of supplier employment
than
expected.35
Incentives to attract new auto assembly plants
Finally, several tests address the
1997 employment'
question of structural change in the
Since
spatial pattern of supplier plant loca­
State
Entire
plant
tions.
While limited by the cross-sec­
Company
State
investment
network openedb
tional
nature of the data available, these
($mil.)
(--------- -percent---------)
results suggest that the degree of spatial
Honda
Ohio
21
23.4
27.0
concentration of supplier plants around
Honda
Ohio
67
assembly plants has increased since 1980.
Nissan
Tennessee
33
1 1 .2
21 .1
The timing of that change is consistent
AutoAlliance
Michigan
49
21 .1
1 4.6
with
the application of lean manufactur­
Diamond-Star
Illinois
83
12.9
4.0
ing
techniques
and just-in-time produc­
Toyota
Kentucky
1 50
14.8
28.9
tion
linkages.
However,
the order of
Toyota
Indiana
72
magnitude
of
the
increased
concentra­
Saturn
Tennessee
80
12.1
71 .3
tion does not support the concept of a
Subaru-Isuzu
Indiana
86
1 7.6
22.7
BMW
S. Carolina
1 30
1 8.1
53.6
supplier base that is tightly clustered
Mercedes
Alabama
252
16.5
0.0
around its customers. Within the auto
corridor, the existing infrastructure appar­
aSince I do not have information on the distribution of a supplier's output
ently allows for frequent deliveries to
across its customers, I adjust the reported plant-level employment by dividing
it by the number of customers per supplier plant. In essence, I am treating all
multiple
customers from a single supplier
customers as of equal importance to a supplier. The last two columns report
plant location.
the percentages of supplier employment in the state of the assembly plant
TABLE 8

based on these adjusted employment figures.
Percentages in this column refer to tier 1 suppliers only as I do not have
information on the start-up year of mixed-tier supplier plants.
Sources: R. Perrucci, 1994, Japanese Auto Transplants in the Heartland, New
York, De Gruyter; ELM International, Inc., 1997, "The ELM GUIDE supplier
database," East Lansing, Ml, database file; and author's calculations.

Federal Reserve Bank of Chicago

31

NOTES
Marshall (1920) identified three reasons for localization: An
industrial center allows a pooled labor market for workers with
specialized skills; an industrial center allows provision of nontraded inputs specific to an industry in greater variety and at
lower cost; and an industrial center generates technological
spillovers as information flows more easily locally (Krugman,
1991, pp. 36-37).

2The authors find the largest coagglomeration for the follow­
ing two upstream-downstream industry pairs: motor vehicle
parts and accessories (SIC 3714) and motor vehicles, car
bodies (SIC 3711); and automotive stampings (SIC 3465)
and motor vehicles, car bodies (SIC 3711).
3It identifies for each of these the address, the list of products
produced as well as the production processes used, employ­
ment, and the plants’ customers (at the company level). See
ELM International, Inc. (1997).
4My analysis does not cover the so-called captive supplier
plants. An earlier paper (Klier, 1995) presented a much more
limited analysis of the same issues for a comparatively small set
of data for independent supplier plants operational in 1993.

5It is difficult to accurately assess the coverage of this database,
since the size of the true population is unknown.
6JapanAuto Parts Industries Association (1998).
’About 8.1 percent of the 2,008 tier 1 plant records as provid­
ed by the ELM database could not be tracked down, either in
the manufacturing directories or by phone, and are therefore
not included in the subsequent analysis.

8However, this is not equivalent to a time-series analysis since
the sample only contains plants operating during 1997 and not
those plants that were shut down in earlier years.
9They represent 1,845 tier 1 and 1,292 mixed-tier plants.
10Honda opened its first auto assembly plant in the U.S. in Ohio
in 1982. McAlinden and Smith (1993) refer to the 1980s as a
period of significant structural change for the U.S. automotive
parts industry.

nWoodward (1992) presents empirical evidence of the impor­
tance of highway access at the county level in attracting plant
openings.

12About 63 percent of foreign-owned plants are Japanese;
see table 2.
13Automobile assembly and component plants that are fully or
partly owned by foreign companies are generally referred to as
transplants.
14Smith and Florida (1994) find evidence for such an effect for
their sample of Japanese-owned supplier plants.

15In the case of Japanese assembly plants, this has been well
documented in the context of corporate ties between assembly
and supplier companies (see Reid et al., 1995). For example,
Ohio is perceived by both Japanese assemblers and bankers as
“Honda’s state” (see Rubenstein, 1992).

32

16There were too few observations for the following two
categories to be reported in the table: domestic suppliers not
supplying to Big Three assembly plants (16 plants) and Japanese
suppliers only supplying to Big Three facilities (nine plants).
However, in both cases the evidence is consistent with table 4.
Plants in these two categories are noticeably less concentrated
in the top three states (31.2 percent for the domestic supplier
plants and 33.3 percent for the Japanese-owned plants).
17See Rubenstein (1997) on the reconcentration of auto
assembly plants in the Midwest and Southeast.

18Only two of the Japanese-owned assembly plants in the study
opened after 1987—Diamond-Star and Toyota, both in 1988.
19Reid et al. (1995) suggest that was one of the ways Japanese
assemblers minimized risk and uncertainty related to their
direct foreign investment in the U.S.

20The vast majority of supplier plants (over 90 percent) ship
to multiple customers.
21The tables and maps refer only to supplier plants located in
the U.S. The overwhelming majority of independent suppliers
located in Canada are concentrated in southwest Ontario, be­
tween Windsor and Toronto. These plants are well connected
to assembly plants in Canada and the northern end of the U.S.
auto corridor via route 401.
22All distances are calculated between the respective coordi­
nates of a plant’s zip code; they are not adjusted for actual
travel routes.
23For example, Kenney and Florida (1992) report data on
approximately 70 Japanese-owned auto supplier plants in the
auto corridor and show 41.4 percent of plants within 100 miles
of their respective assembly plants. In contrast, the highest
concentration of Japanese-owned suppliers around Japanese
assemblers I can find applies to the Honda network, with 29.3
percent of Japanese-owned suppliers within 100 miles of the
assembly plant, followed by Toyota (22.9 percent) and Auto
Alliance (22.5 percent).

24The Buick plant in Flint was at the time one of the largest in­
tegrated automotive plants in the world. It employed about
22,000 people and produced about 2,000 cars a day. Henrickson’s
data include both independent and captive suppliers of metal
auto parts, tire and tube supplies, and mechanical rubber goods.
25Between 1980 and 1983, three Ford assembly plants closed
(two in California and one in New Jersey). In 1992, a body plant
for Ford’s large vans in Avon Lake, Ohio, added an assembly line
for the production of the Mercury Villager/Nissan Quest.
26As the opening of the Avon Lake assembly line in 1992 could
possibly explain some of the increase in supplier plants locat­
ing close to Dearborn, Michigan, I checked for robustness of my
results by shortening the second time period to end in 1991. The
exercise leaves the spatial distribution of Ford suppliers that
opened during the 1980s essentially unchanged. This strongly
suggests that the opening of the Avon Lake assembly line is
not driving the reported reconcentration.

27I would like to thank David Marshall, who suggested using
this technique.

Economic Perspectives

28In calculating these distances, I consider only assembly plants
that were operational when the supplier plant opened.
29From 1950 to 1979, that corresponds to 77 light vehicle
assembly plants; for the later time period, there are 76 plants.

30The Wilcoxon signed-rank test is a nonparametric test that
can be used to test whether the median of a set of observations
equals some prespecified value. The test is based on calculating,
ranking, and signing the differences between the actual obser­
vations and the constant. In panel A of table 8, I report the
results from testing whether the median of the more recent
distances between assembly and supplier plants (52.2 miles
in the case of all observations) is different from the median
of the distribution of distances for the older set of observa­
tions (60.4 miles). The test statistic T, which is distributed
approximately normally, is obtained by taking the differences
D = x. - median 9, where x represents the actual distances
observed in the older data set. These differences are then ranked
and signed; the test statistic T represents the sum of the signed
ranks. The null hypothesis states that the median difference D.
equals zero. If it cannot be rejected, it follows that the median
distances for both data sets are equal.
31I would like to thank Jim Rubenstein, who suggested this
approach.
32See Reid (1994), Mair (1993), and Kenney and Florida (1992),
who seem to suggest the need for very close proximity between
assembler and suppliers.

They take this as evidence to suggest that the benefits of incen­
tive packages intended to attract large manufacturing plants will
not remain within the communities or states providing such
incentives.
34As table 9 shows, the percentage of supplier employment
residing within the state of the assembly plant tends to increase
if calculated for the set of supplier plants that opened after the
respective assembly plant. In a couple of cases the percentages
increase dramatically, but it is important to point out that
these changes are in reference to only a small number of sup­
plier plant openings. In Mercedes’ case, no supplier plant
opened during 1997.

35The Center for Business and Economic Research’s (1992)
analysis of the economic impact of Toyota’s assembly plant
on the other auto corridor states plus Illinois employs a specific
input-output model (RIMS II) and its multipliers. Comparing
the distribution of jobs associated with the production of inputs
for Toyota’s assembly plant, information from my network
data shows the overall network employment at about 36 per­
cent of that estimated in the earlier study. However, one needs
to point out that the numbers are not directly comparable, as
the ELM database does not include purchases of raw materials
and production equipment. With that caveat, a comparison of
the distribution of employment by state suggests that Toyota’s
supplier network might actually be more dispersed than origi­
nally estimated. Specifically, based on information presented in
this article, the share of jobs in Michigan and Indiana is lower
than estimated, while Illinois, Ohio, and Tennessee report a
relatively higher share.

33Elhance and Chapman (1992) find similar evidence in ana­
lyzing the labor market of the Diamond-Star assembly plant in
central Illinois. They find that the labor market for that plant
covers a large geographical area, stretching over 15 states.

REFERENCES

Center for Business and Economic Research, 1992,
“The economic significance of Toyota Motor manu­
facturing, U.S. A. Inc., in Kentucky,” Review and
Perspective, University of Kentucky, December.

Estall, R. C., 1985, “Stock control in manufacturing:

The just-in-time system and its locational implications,”
Hrea, Vol. 17, No. 2, pp. 129-132.
Fournier, Stephen F., and Andrew M. Isserman, 1993,

Database Publishing Co., 1997, Arizona and California

Dyer, Jeffrey H., 1994, “Dedicated assets: Japan’s

“Putting it all together: The effects of the Honda
plant on its host county and rural hinterland,” West
Virginia University, Regional Research Institute,
research paper, No. 9305.

manufacturing edge,” Harvard Business Review,
November-December, pp. 174-178.

Glasmeier, Amy, and Richard McCluskey, 1987, “U.S.

Manufacturers Register, Phoenix, AZ.

Elhance, Arun P., and Margaret Chapman, 1992,

“Labor market of a U.S.-Japanese automobile joint
venture,” Growth and Change, Vol. 23, No. 2, pp.
160-182.
Ellison, Glenn, and Edward L. Glaeser, 1997, “Geo­

graphic concentration in U.S. manufacturing industries:
A dartboard approach,” Journal of Political Econo­
my, Vol. 105, No. 5, pp. 889-927.
ELM International, Inc., 1997, “The ELM GUIDE

supplier database,” East Lansing, MI, database file.

Federal Reserve Bank of Chicago

auto parts production: An analysis of the organization
and location of a changing industry,” Economic
Geography, Vol. 63, No.2, pp. 142-159.
Helper, Susan, 1991, “How much has really changed
between U.S. automakers and their suppliers?,”
Sloan Management Review, Vol. 15, No. 3, pp. 15-28.

Henrickson, G. Rex, 1951, Trends in the Geographic
Distribution of Suppliers of Some Basically Important
Materials Used a! the Buick Motor Division, Flint,
Michigan, Ann Arbor, ML University of Michigan,
Institute for Human Adjustment.

33

Hewings, Geoffrey J. D., Philip R. Israilevich, and
Graham R. Schindler, 1998, “Interstate trade among

Midwest economies,” Chicago Fed Letter, Federal
Reserve Bank of Chicago, May, No. 129.
Japan Auto Parts Industries Association, 1998,

Perrucci, R., 1994, Japanese Auto Transplants in the
Heartland: Corporatism and Community, New York:
De Gruyter.

Pick Publications, 1997, Michigan Manufacturers
Directory, Southfield, ML

Membership Directory.
Kenney, Martin, and Richard Florida, 1992, “The

Japanese transplants—Production organization and
regional development,” Journal of the American
Planning Association, Vol. 58, No. 1, pp. 21-38.
Klier, Thomas, 1995, “The geography of lean manu­

facturing: Recent evidence from the U.S. auto industry,”
Economic Perspectives, Federal Reserve Bank of
Chicago, Vol. 19, No. 6, pp. 2-17.

Reid, Neil, 1994, “Just-in-time inventory control and
the economic integration of Japanese-owned manu­
facturing plants with the county, state, and national
economies of the United States,” Regional Studies,
Vol. 29, No. 4, pp. 345-355.

Reid, Neil, Andrew Solocha, and Breandan
O hUallachain, 1995, “Japanese corporate groups

Krugman, Paul, 1991, Geography and Trade, Cam­
bridge, MA: MIT Press.

and the locational strategy of Japanese auto and
component parts makers in the United States,” in The
Location of Foreign Direct Investment, Milford B.
Green and Rod B. McNaughton (eds.), Aldershot,
UK: Avebury, pp. 107-120.

Mair, Andrew, 1994, Honda’s Global Local Corpora­

Rubenstein, James M., 1997, “The evolving geography

tion, New York: St. Martin’s Press.

of production—Is manufacturing activity moving out of
the Midwest? Evidence from the auto industry,”
Assessing the Midwest Economy, Federal Reserve
Bank of Chicago, working paper, No. SP-3.

________ , 1993, “New growth poles? Just-in-time
manufacturing and local economic development strat­
egy,” Regional Studies, Vol. 27, No. 3, pp. 207-221.
________ , 1992, “Just-in-time manufacturing and
the spatial structure ofthe automobile industry: Les­
sons from Japan,” Tijdschrift voor Econ. en Soc.
Geografie, Vol. 82, No. 2, pp. 82-92.
Mair, Andrew, Richard Florida, and Martin Kenney,

1988, “The new geography of automobile production:
Japanese transplants in North America,” Economic
Geography, Vol. 20, October, pp. 352-373.
Manufacturers’ News Inc., \991, Alabama, Georgia,
Florida, Illinois, Indiana, Iowa, Kentucky, Missouri,
Nebraska, North Carolina, Ohio, Oklahoma, Penn­
sylvania, South Carolina, Tennessee, Texas, Virginia,
and Wisconsin Manufacturers Register, Evanston, IL.

Marshall, Alfred, 1920, Principles ofEconomics,
London: Macmillan.
Marvel, Mary K., and William J. Shkurti, 1993,

“The economic impact of development: Honda in
Ohio,” Economic Development Quarterly, Vol. 7,
No. 1, pp. 50-62.

_________ , 1992, The Changing U.S. Auto Industry—
A Geographical Analysis, London: Routledge.
________ , 1988, “Changing distribution ofAmerican
motor vehicle parts suppliers,” Geographical Review,
Vol. 18, No. 3, pp. 288-298.
Rubenstein, James M., and Neil Reid, 1987, “Ohio’s

motor vehicle industry—Industrial change and geo­
graphical implications,” Miami University, geographi­
cal research paper, No. 1.
Smith, Donald, and Richard Florida, 1994, “Agglomer­
ation and industrial location: An econometric analysis
of Japanese-affiliated manufacturing establishments
in automotive-related industries,” Journal of Urban
Economics, Vol. 36, No. 1, pp. 23^11.

Tower Publishing, 1997..Massachusetts, Maine, New
Hampshire, and Vermont Directory ofManufacturers
1997, Standish, ME.
Ward’s Communications, 1950-98, Ward’s Automo­

tive Yearbook, Overland Park, KS.

McAlinden, Sean, and Brett Smith, 1993, “The

Woodward, Douglas P., 1992, “Locational determi­

changing structure ofthe U.S. automotive parts
industry,” University of Michigan, Transportation
Research Institute, research paper, No. UMTRI93-6.

nants of Japanese manufacturing start-ups in the
United States,” Southern Economic Journal, Vol. 58,
No. 3, pp. 690-708.

34

Economic Perspectives

The new view of growth and business cycles

Jonas D. M. Fisher

Introduction and summary
Two central concerns of economic policy are growth
and business cycle stabilization. There is considerable
interest in devising government policies and institu­
tions to influence prospects for economic growth and
mitigate the distress associated with economic down­
turns. Proper evaluation of the benefits and costs of
a given policy proposal requires knowledge of the
determinants of growth and business cycles. This is
one reason for the considerable body of research
aimed at understanding these phenomena.
The last two decades have seen considerable
advances in this research. Recent empirical evidence,
however, brings into question two of its basic assump­
tions—first, that technological change is homoge­
neous in nature, in that it affects our ability to produce
all goods symmetrically, including consumption and
investment goods; and second, that business cycles
are driven by shocks which affect the demand for
investment goods.
In this article, I document the key evidence that
challenges the conventional views of growth and
business cycles. I then discuss the plausibility of alter­
native theories that have been advanced to meet the
challenge. To date, the evidence seems to support a
new view of growth and business cycles, one that is
based on technical change biased toward new invest­
ment goods like capital equipment.
The key evidence involves two observations on
the behavior of the relative price of business equip­
ment over the last 40 years. First, in almost every year
since the end of the 1950s, business equipment has
become cheaper than the previous year in terms of
its value in consumption goods. This means that if
one had to trade restaurant meals for a piece of equip­
ment that makes the same number and quality of, say,
bicycles, one would forgo fewer meals in 1998 than in
1958. Second, this relative price tends to fall the most
when the economy, and investment expenditures in

Federal Reserve Bank of Chicago

particular, are growing at relatively high rates, that is,
it is countercyclical.
The first piece of evidence is striking because it
suggests that much of post-WWII economic growth
can be attributed to technological change embodied
in new capital equipment. This conflicts with conven­
tional views on what drives economic growth. A piece
of capital equipment is a good that is used to produce
another good, such as a crane or a computer. An im­
provement in capital-embodied technology is the
invention of equipment that takes the same amount
of labor and preexisting equipment to produce as the
old equipment but that produces more goods when
combined with the same amount of labor as before.
If a new production process yields the same units of
capital equipment with less factor inputs, then this
has the same economic implications as if the capital
equipment produced were itself more efficient. Hence,
an equivalent interpretation of what constitutes
capital-embodied technical change is that it involves
an improvement in the technology that produces
capital equipment.
To understand the relationship between capitalembodied technical change and the trend in the equip­
ment price, suppose the technology for producing
consumption goods is fixed. With improvements in
technology embodied in equipment, the supply of
(quality-adjusted) investment goods increases rela­
tive to consumption goods, so the equipment price
falls. Greenwood et al. (1997) build on this insight to
show that a large fraction of economic growth can be
attributed to capital-embodied technical change.
This conflicts with the conventional view that most

Jonas D. M. Fisher is a senior economist at the Federal
Reserve Bank of Chicago. This article has benefited
from conversations with Larry Christiano. The author
thanks Judy Yoo for excellent research assistance.

35

growth is due to disembodied technical change, or
multifactor productivity. Improvements in disembod­
ied technology, usually measured as the Solow (1957)
residual, make it possible to produce all kinds of
goods, not just capital goods, with less capital and
labor.1 If this were the dominant source of growth,
then we should not have seen such a large drop in
the price of equipment over the last 40 years.
The second piece of evidence runs counter to
standard views of the business cycle. Standard theo­
ries hold that the business cycle is driven by shocks
which affect the demand for investment goods. For
example, consider the IS-LMmodel, which summarizes
much of what is often called Keynesian macroeconom­
ics. This model is the focus of most textbooks on mac­
roeconomics and underlies much of the discussion of
macroeconomic policy in the media.2 In this model, busi­
ness cycles are due to shocks to aggregate demand,
such as monetary and fiscal disturbances. For example,
expansionary monetary policy stimulates demand for
investment goods through lower interest rates. If there
is an upward sloping supply schedule for investment
goods, we would expect the relative price of invest­
ment goods to rise. The same holds for expansionary
fiscal policy, if government spending does not fully
crowd out investment. Another view of business
cycles, often attributed to Keynes, is that they are
primarily investment cycles driven by variation in an­
imal spirits, that is, changes in confidence about fu­
ture growth prospects.3 With the same assumptions on
investment supply, we would expect investment prices
to be high when investment is high. In summary, tra­
ditional Keynesian views of business cycles imply
that investment good prices should be procyclical,
that is, be high when overall economic activity is
relatively high.
In recent years, an alternative view of business
cycles, based on “fundamentals” that influence aggre­
gate supply, has gained credence. This real business
cycle view says that business cycles are driven in
large part by disturbances to multifactor productivity.
Just as the shocks to aggregate demand which are
central to Keynesian theories, these disturbances
influence business cycles through their effect on the
demand for investment goods.4 Hence, if there are
costs in terms of forgone consumption of expanding
investment good production, that is, if the supply
schedule of capital is upward sloping, these models
also predict the relative price of investment goods to
be procyclical (Greenwood and Hercowitz, 1988).
Since the relative price evidence contradicts the
major schools of business cycle thought, it poses a
challenge to our understanding of business cycles.

36

There are two leading hypotheses that could reconcile
the theory and evidence. One, the embodied technol­
ogy view, is built from the real business cycle tradition
and takes into account the trend evidence on equip­
ment prices. Falling equipment prices are compelling
evidence of capital-embodied technological progress
over long horizons. Perhaps changes in the rate of
such technological progress occur over shorter hori­
zons as well. Suppose the business cycle were driven,
to a large extent, by these disturbances. An increase in
the rate of capital-embodied technical change would
lead to an outward shift in the supply schedule for
investment goods. With stable investment demand,
investment would rise and equipment prices would
fall. This new view of business cycles, which comple­
ments the new view of growth suggested by the longrun evidence on equipment prices, has been explored
by Christiano and Fisher (1998), Fisher (1997), and
Greenwood et al. (1998).
The other leading theory is more easily under­
stood in the context of traditional Keynesian views
of the business cycle. If shocks to aggregate demand
occur with a downward sloping investment supply
curve, then the price of equipment could fall in a boom.
A downward sloping investment supply curve would
arise if increasing returns to scale played an important
part in the production of capital equipment, so this is
called the increasing returns view. This view has been
advanced by Murphy, Shleifer, and Vishny (1989).
Below, I document the trend and business cycle
evidence on equipment prices. There is no reason to
expect that capital-embodied technological change is
unique to equipment. Equipment is one of many invest­
ment good aggregates, that is, types of capital. More­
over, for simplicity most economic models assume
only one or two types of capital. Therefore, in addition
to equipment prices, I analyze other investment good
aggregates. Next, I discuss research that sheds light
on the plausibility of the alternative views, including
some new evidence. To date, the evidence seems to
support the new view of growth and business cycles
based on capital-embodied technical change.
If growth and business cycles are originating
from changes in capital-embodied technology, then
the models we use for policy analysis have to incor­
porate this and, consequently, policy recommenda­
tions could change. For example, to the extent that
technological change is embodied in capital equipment,
government policies that affect equipment investment
could have a dual impact on growth via the quality
and quantity of capital goods. This could mean, for
example, that investment tax credits directed toward
improvements in the efficiency of capital equipment
could have a significant impact on growth.

Economic Perspectives

Measuringprices and quantities
This section describes how relative prices and
real quantities of investment goods are measured. My
measures of prices and quantities are based on mea­
sures published in the “National income and product
accounts” (NIPA) of the U.S. Bureau of Economic
Analysis (BEA).
The basis of the BEA procedure is to construct a
price deflator. To be concrete, a given nominal quan­
tity of expenditures on some good /', X't, is decomposed
into a price deflator, P't, (which measures the nominal
price of the good) multiplied by a quality-adjusted
index of the real quantity of the good, q't.
The BEA measures P't and q't for different goods
using a so-called chain-weighting procedure, which
is summarized in box 1. My measure of quantity is
simply q't, measured in units of 1992 dollars. My mea­
sure of the real price, alternatively the relative price,
of good / at date /, p‘, is the real quantity of consump­
tion goods that would need to be sold in order to pur­
chase one unit of good / at time /. It is defined as the
price deflator for good / divided by the price deflator
for consumption of nondurables and services. The
rationale for this measure is described in box 1.

business cycle. Below, I provide a brief description of
how I measure the business cycle component of the
data. A detailed discussion of the procedure is given
in Christiano and Fitzgerald (1998).
Figure 1 illustrates the basic idea behind the pro­
cedure. The colored line in panel A of figure 1 displays
real 1992 dollar chain-weighted gross domestic prod­
uct (GDP). The reported data are the logarithm of the
raw data. The advantage of using the logarithm is
that the resulting movements correspond to percent
changes in the underlying data. The deviations be­
tween the data and the trend line (graphed in panel B)
contain the rapidly varying, erratic component, inher­
ited from the choppy portion of the data that is evident
in panel A. The colored line in panel B is my measure
of the business cycle component of real GDP. This
measure excludes both the trend part of the data and
the rapidly varying, erratic component. It includes only
the component of the data that contains fluctuations
in the range of two to eight years. According to this
approach, the economy is in recession when the
business cycle measure is negative and in prosperity
when it is positive.
Figure 1 also compares this measure of the busi­
ness cycle with the one produced by the National
Bureau ofEconomic Research (NBER). This organiza­
tion decides, based on an informal examination of many
data series by a panel of experts, when the economy
has reached a business cycle peak or trough. The
start of each shaded area indicates the date when,
according to the NBER, the economy reached a busi­
ness cycle peak. The end of each shaded area indi­
cates a business cycle trough. Note how real GDP falls
from peak to trough and then generally grows from
trough to peak. An obvious difference in the two busi­
ness cycle measures is that the measure used in this
article is a continuous variable, while the NBER’s takes
the form of peak and trough dates. As a result, my
measure not only indicates when a recession occurs,
but also the intensity of the recession. Apart from
these differences, the two measures appear reasonably
consistent. For example, near the trough of every NBER
recession, my measure of the business cycle is always
negative. However, the two measures do not always
agree. According to my measure, the economy was in
recessionin 1967 and 1987, while the NBER did not de­
clare a recession then. In part, this is because there
must be several quarters of negative GDP growth before
the NBER declares a recession. The procedure I use
only requires a temporary slowdown.

Measuring the business cycle component ofthe data5
In the introduction I described how the price of
producer durable equipment (PDE) varies over the

The data
I consider a broad variety of investment goods, as
outlined in table 1. The broadest measure of investment

The implications for stabilization policy of the em­
bodied technology view are less obvious. The fact
that it seems to supplant the increasing returns view
means that the arguments for interventionist stabili­
zation policy that this view lends support to are less
compelling. For example, increasing returns could
provide scope for policy intervention, as it either
involves externalities or is inconsistent with perfect
competition. Moreover, it makes models based on
animal spirits more plausible, which also has implica­
tions for stabilization policy (see Christiano and
Harrison, 1999). The embodied technology view is
more in line with the real business cycle tradition, in
which policy interventions are counterproductive.

Evidence on investment good prices
To study the trend and business cycle properties
of investment good prices, we need two things—a
way to extract real prices and quantities from data on
nominal investment expenditures; and a precise defi­
nition of what we mean by the business cycle compo­
nent of the data. Below, I address these issues. Then,
I introduce the data and present the results character­
izing the trend and cycle behavior of investment
good prices.

Federal Reserve Bank of Chicago

37

BOX 1

Measuring real quantities and prices from nominal expenditure data
The U.S. Bureau of Economic Analysis (BEA)
uses the chain-type Fisher index to measure real
output and prices. For a thorough discussion of
the procedures the BEA uses, see Landefeld and
Parker (1997), which this box draw s on. This index,
developed bv Irving Fisher, is a geometric mean of
the conventional fixed-w eighted Laspevres index
(w hich uses w eights of the first period in a twoperiod example) and a Paassche index (which uses
the w eights of the second period). The Laspevres
price index for period t constructed using base
year t - l,i( is given by

The Paassche price index for period t constructed
using base year /, St is given by-

series that allow s for the effects of changes in rel­
ative prices and in the composition of output over
time. Notice that a quantity index can be comput­
ed in a manner analogous to the price index. A nice
feature of the Fisher index is that the product of
these tw o indexes equals nominal expenditures.
Landefeld and Parker (1997) discuss several ad­
vantages of this index over previously used fixed
weight indexes.
To measure relative prices we need to choose
a numeraire. In the introduction the term “value in
consumption goods” w as used. Implicit in this
statement is the assumption that consumption
goods, specifically nondurable and services con­
sumption, is the numeraire. Define the price defla­
tor for nondurable and services consumption as
P‘. Then the relative price of the good i at time 1, p‘
is defined as
time t dollars / good i

time t dollars/consumption good
consumption goods

Here N is the number of goods w hose prices are
being summarized by the index, P't is the date t dollar
price of the zth quality-adjusted good, and t// is the
quality-adjusted quantity of good i purchased at
date t. The Fisher price index at date 1, Ft is
Ft = yl L,X S, .
From this definition we see that changes in Ft
are calculated using the “w eights” of adjacent
years. These period to period changes are
“chained” (multiplied) together to form a time

is total private investment (TP1). This measure in­
cludes all private expenditures on capital goods and
consumer goods designed to last more than three
years.6 This is a broader measure of investment than
the conventional NIPA measure of investment, pri­
vate fixed investment (PF1), which excludes expendi­
tures on consumer goods. Within TP1, 1 define two
main components, nonresidential and residential.
Nonresidential has two main subcomponents, struc­
tures (NRS, for example, factory buildings and office
buildings) and producer durable equipment (PDE, for
example, auto-assembly robots and personal comput­
ers). Similarly, residential is broken down into residen­
tial structures and equipment (RSE, for example, single
family homes and refrigerators) and consumer
durables (CD, for example, televisions and vacuum

38

good i

Notice that the units of the price are w hat we
require. The BEA does not provide a measure of
price deflator for nondurable and services consump­
tion. To construct the consumption deflator used
in this article, I applied the chain-weighting meth­
odology outlined above, treating the NIPA quantity
and price indexes for nondurable consumption
and service consumption as the prices and quanti­
ties in the formulas.

cleaners). These four major subcomponents of TP1 are
then broken down further.7
The “Nominal share” and “Real share” data pro­
vide information on the relative magnitudes of expen­
ditures on the different measures of investment, as
well as a preliminary indication of interesting trends
in relative prices. The nominal and real shares for TP1
are calculated as the ratio of nominal and real TP1 rel­
ative to nominal and real GDP, respectively. For exam­
ple, in 1958 nominal TP1 expenditures were 22 percent
of nominal GDP and real TP1 expenditures were 16
percent of real GDP. The remaining shares are calcu­
lated using TP1 as the base for the share calculations.
For example, PDE expenditures accounted for 24 per­
cent of nominal TP1 and 20 percent of real TP1 in 1958.8

Economic Perspectives

real and nominal shares for many of the
components of investment listed in table
1, suggesting that trends in relative pric­
es are exhibited by many ofthe subcom­
ponents of TP1. Second, the difference
between the real shares of TP1 and PF1
(the former is a fraction of GDP, while the
latter is a fraction of the former) is seen
to be due to the increasing quantities of
consumer durables being purchased.
Third, the much talked about “informa­
tion age” manifests itself here as the
huge increase in the fraction of TP1 that
has been due to expenditures on informa­
tion and related equipment since 1960. In
1960 this type of investment accounted
for less than 1 percent of real TP1. By
1995, its share had grown to 13 percent.
Finally, note that both residential and
nonresidential structures account for
less ofTPl in 1998 than in 1958.

(I explain the last two columns in table 1 in the section
on prices of investment goods over the business cy­
cle, which begins on page 40.)
Table 1 reveals several interesting facts about
how expenditures on investment have changed since
1958 and underlying trends in relative prices. First,
nominal TP1 expenditures have been roughly stable
(abstracting from short-run movements) as a fraction
of nominal GDP since the late 1950s. Yet, the real quan­
tity of this broadest measure has been growing as a
fraction of real GDP. In 1958, TP1 was 16 percent of
1992 chain-weighted GDP, compared with 26 percent
in 1998. The fact that nominal and real shares behave
in this way is an indication that the relative price of
this bundle of investment goods fell between 1958
and 1998. Notice that there are differences between

Federal Reserve Bank of Chicago

Trends in investment goodprices
In this section, 1 explain two main
findings relating to the long-run behavior
of relative prices for the various compo­
nents of investment listed in table 1. First,
the relative price of TP1 has fallen consis­
tently since the mid-1950s. Second, there
is considerable heterogeneity in the longrun behavior of the prices of the subcom­
ponents ofTPl. Generally, the behavior
ofthe price ofTPl is dominated by dra­
matic drops in the prices of PDE and CD,
which are also evident in the prices of most
of the main subcomponents of these in­
vestment aggregates. The prices ofRSE
and NRS and their subcomponents, while exhibiting
trends over subsamples ofthe period studied, have
not fallen as consistently and their changes over time
are much smaller than those of PDE and CD.
Figure 2 displays the relative price trend evidence.
The black lines in figure 2 are measures ofthe (natural
logarithm of the) relative price of each of the invest­
ment components listed in table 1 over the period for
which data are available.9 The colored lines are the
trends calculated in the same way as the trend of real
GDP displayed in figure 1. The first column of panels
in figure 2 displays prices and trend lines for the main
aggregates. The remaining columns display prices
and trends for the four broad categories of TP1 and
their main subcomponents.
Figure 2 shows that the relative prices of differ­
ent components of investment have behaved quite

39

TABLE 1

Measures of investment used in the analysis

Total private investment
Nonresidential
Structures
Nonresidential buildings
Utilities
Mining exploration, shafts, & wells
Producer durable equipment
Information & related
Industrial
Transportation & related
Residential
Residential structures & equipment
Single family structures
Multifamily structures
Other structures
Consumerdurables
Motorvehicles& parts
Furniture & household equipment
Other
Private fixed investment

Business cycle
volatility

Real share

Nominal share

1958

1978

1998

1958

1978

1998

0.2184
0.4165
0.1730
0.0975
0.0418“
0.0233
0.2438
0.0355
0.0796
0.0598
0.5830
0.2188
0.1289
0.0228
0.0627
0.3643
0.1453
0.1659
0.0534
0.6357

0.2641
0.4496
0.1510
0.0821
0.0399
0.0255
0.2985
0.0690
0.0783
0.0782
0.5504
0.2175
0.1203
0.0212
0.0715
0.3329
0.1539
0.1223
0.0567
0.6671

0.2378
0.4656
0.1226
0.0906
0.0238b
0.0115
0.3430
0.1153
0.0734
0.0871
0.5344
0.1782
0.0897
0.0122
0.0721
0.3562
0.1414
0.1443
0.0705
0.6438

0.1558
0.4254
0.2613
0.1435“
0.0552“
0.0302“
0.1981
0.0080“
0.1134“
0.0797“
0.5730
0.3259
0.2118“
0.0406“
0.1028“
0.2835
0.1361
0.0918
0.0551
0.7294

0.2154
0.4356
0.1674
0.0982
0.0412
0.0207
0.2676
0.0327
0.0958
0.0900
0.5646
0.2526
0.1350
0.0257
0.0860
0.3184
0.1629
0.0933
0.0670
0.6812

0.2627
0.4815
0.1032
0.0764
0.0226b
0.0089
0.3841
0.1300b
0.0747b
0.0762b
0.5200
0.1546
0.0754
0.0109
0.0645
0.3663
0.1290
0.1717
0.0697
0.6333

CTq'/

2.97
2.83
2.66
3.63
2.67
5.49
3.18
3.05
3.63
5.25
3.98
6.24
8.89
10.80
3.18
2.99
5.16
1.94
1.52
3.09

®qy

0.55
0.98
0.90
0.49
0.64
3.19
0.85
0.95
0.91
0.63
0.43
0.57
0.81
0.81
0.34
0.61
0.94
0.54
0.59
0.58

a1959 data.
b1995 data.
c1960 data.
Notes: See box 1 for a description of the notation. For total private investment and gross domestic product, Y, nominal shares
in the first row are P™q™/(PYqY). Nominal and real shares for investment good / in the other rows are given by P‘ q‘/(PT^'qT^).
Real shares for total private investment and gross domestic product are q™/qY. Real shares for investment good /' in the other
rows are given by q‘/qTPtl.
Source: U.S. Department of Commerce, Bureau of Economic Analysis, 1947-98, "National income and product accounts,"
Survey of Current Business, and author's calculations of the business cycle component of the data.

differently in the postwar era. The price of the broadest
investment measure, TPI, has been falling consistently
since the early 1950s. Since the plot of the relative
price of TPI is in natural logarithms, one can take the
difference between the prices for two years to calcu­
late the percentage change. This procedure indicates
that the price of TPI in terms of consumption goods
fell about 42 percent between 1958 and 1998.
Studying the other plots in figure 2, we see that
this large drop in the price of TPI can be attributed to
strong downward trends in PDE (particularly informa­
tion and related and transportation equipment) and
CD (all three types). The drop in the relative price of
information equipment is particularly dramatic, at almost
200 percent since 1961. The prices ofNRS and its com­
ponents were generally rising until the late 1970s, were
falling for most of the rest of the sample period, and
have started to rise again in the 1990s. RSE and its
components display a similar pattern. Generally, the
long-run changes in structures prices have been much
smaller than in PDE and CD prices. When the invest­
ment components are aggregated into nonresidential

40

and residential, the strong downward trends in PDE
and CD prices dominate the changing trends in
structures.10

Prices ofinvestment goods over the business cycle
My objective here is to determine the extent to
which investment good prices are generally procyclical,
countercyclical, or acyclical (do not display any dis­
tinctive pattern over the business cycle). I find that,
generally speaking, prices of PDE, NRS, and their com­
ponents are countercyclical, prices of RSE and its
components are procyclical, and prices of CD and its
components are acyclical. There is some sample period
sensitivity, as outlined below.
In table 1, the column headed o ,/a indicates
the relative volatility of the different investment com­
ponents over the business cycle. This is the standard
deviation of the business cycle component of the in­
dicated real quantity series divided by the standard
deviation of the business cycle component of real
GDP. We see that TPI varies almost three times as
much as GDP. The most volatile components of

Economic Perspectives

Fede ral Res erv e Bank of Chicago

FIGURE 2

Trends in investment good prices, 1947-98 (logarithm)
A. Total private investment

E. Nonresidential structures

M. Residential structures

B. Private fixed investment

F. Nonresidential buildings

N. Single family

0. Multifamily

H. Mining

L. Transportation

Notes: Relative price (black line) is a measure of the (natural logarithm of the) relative price of each of the investment components listed in table 1 over the period for which data
are available. The trend (colored line) is calculated in the same way as the trend of real GDP displayed in figure 1. Panels A through D show prices and trends for the main
aggregates. Panels E through T show prices and trends for the four broad categories of total private investment, along with their main subcomponents.
Source: See figure 1.

Q. Consumer durables

investment are single family structures, multifamily
structures, and consumer expenditures on motor vehi­
cles and parts. The least volatile components are
NRS, furniture and household equipment, and the
“other” component of CD. The column headed /a
indicates the relative volatility of the prices of differ­
ent investment components over the business cycle.
This is the standard deviation of the business cycle
component of the indicated relative price series divid­
ed by the standard deviation of the business cycle
component of real GDP. The prices are much less vola­
tile than the quantities. With one exception (mining ex­
ploration, shafts, and wells), all the prices are less
volatile than real GDP over the business cycle.
As a preliminary look at the cyclicality of invest­
ment good prices, figure 3 displays the business cycle
components of the prices (colored lines) and quanti­
ties (black lines) of seven of the broadest measures
listed in table 1, along with the business cycle compo­
nent of the deflator for consumption of nondurables
and services. The latter price is used in the denomi­
nator of all the investment relative prices, so its busi­
ness cycle dynamics will influence all the relative
price measures discussed here.11
Notice first that the consumption deflator rises in
all but one recession, 1981 :Q3-82:Q4 (see shaded areas
in figure 3). This is a force for procyclicality of invest­
ment good prices. For example, if the price deflator for
an investment good were constant, then the real price
of that good would be procyclical. As expected, the
quantities are generally procyclical, although the peaks
and troughs do not exactly coincide with the NBER
dates. The prices do not display as consistent a pat­
tern as the quantities. For example, sometimes the
price of TPI moves with the quantity of TPI (1950s,
1960s, and 1990s) and sometimes it moves in the oppo­
site direction (1970s and 1980s). More distinct patterns
emerge when TPI is decomposed into nonresidential
and residential. In the 1950s and 1990s, the prices and
quantities of nonresidential appear to move closely
together. In the 1960s, 1970s, and 1980s, prices and
quantities of this investment measure generally move
in opposite directions. Prices and quantities of resi­
dential show more evidence of moving together. The
most striking pattern to emerge among the subcom­
ponents of nonresidential and residential is in PDE.
With the exception of the 1950s, almost every time the
quantity of PDE moves up, the price of PDE moves
down. This suggests countercyclical behavior in the
real price of PDE.
For a more formal examination of how the prices
of investment goods vary with the business cycle,
I use a cross-correlogram. A cross-correlogram is a

42

diagrammatic device for describing how two variables
are related dynamically. For example, it provides a mea­
sure of whether, say, movements in one variable tend
to occur at the same time and in the same direction as
movements in another variable. It can also be used to
measure whether, for example, positive movements in
a variable tend to occur several quarters ahead of posi­
tive movements in another variable.
The basis for the cross-correlogram is the corre­
lation coefficient, or correlation. A correlation is a mea­
sure of the degree to which two variables move together
and always takes on values between -1 and 1. If a cor­
relation is positive, then the two variables are said to
be positively correlated. Similarly, if a correlation is
negative, the variables are said to be negatively cor­
related. Larger absolute values in a correlation indicate
a stronger pattern of moving together. A correlation
for two variables measured contemporaneously is a
measure of how much two variables move together at
the same time. A correlation can be computed for two
variables measured at different times. For example,
we can measure the correlation between variable x at
time / and variable y at time / - k, where k is a positive
integer. This would measure the degree to which vari­
ations in v occur before movements in x. A cross-corre­
logram plots these correlations for various values of k.
Figure 4 displays cross-correlograms (along with
a two-standard-deviation confidence interval, a mea­
sure of how precisely the correlations are estimated)
for various business cycle components of real invest­
ment and GDP, -6 < k < 6. For example, panel A of
figure 4 displays the correlations of real nonresidential
investment at date / and real GDP at date t - k for the
various values of k. The fact that the correlation for
k = 0 is positive and close to 1 for all the plots in fig­
ure 4 shows that all the components of investment
displayed are strongly positively correlated with GDP
contemporaneously. This confirms the impression
given by figure 3 that real expenditures on these invest­
ment goods are strongly procyclical. Notice that the
largest correlations for nonresidential and its two main
subcomponents, NRS and PDE, are for k> 0. This says
that these components of investment tend to lag GDP
over the business cycle. Another way of saying this
is that movements above trend in GDP tend to occur
before movements above trend in these measures of
investment. On the other hand, the largest correlations
for residential and its main subcomponents, RSE and
CD, are all for k < 0. This says that these components
of investment lead output over the business cycle.
Because the correlations in figure 4 are mostly positive,
this figure shows that the main components of invest­
ment are generally procyclical. (If they had been mostly

Economic Perspectives

FIGURE 3

Business cycle components of investment good prices and quantities
E. Total private investment

logarithm

B. Nonresidential

F. Residential

logarithm

logarithm

C. Nonresidential structures

G. Residential structures

logarithm

logarithm

D. Durable equipment

H. Consumer durables

logarithm

Notes: Each business cycle component has been scaled by its standard deviation, and all data are quarterly.
The colored lines represent the business cycle component of the price series for the indicated variable and the black lines
represent the business cycle component of the quantity series for the indicated variable. Shaded areas indicate recessions
as determined by the National Bureau of Economic Research.
Source: See figure 1.

Federal Reserve Bank of Chicago

43

negative, then this would have been evidence of coun­
tercyclicality. If the correlations were mostly close to
zero, this would have been evidence of acyclicality.)
Figure 5 displays cross-correlograms (with stan­
dard errors) for the prices of the broadest measures
of investment and real GDP. The plots in figure 3 indi­
cate that there may be some sample period sensitivity
in the estimation of the underlying correlations, so
figure 5 displays cross-correlograms based on two
sample periods. The first column of panels in figure 5
is based on the sample period 1947: Q1-98 :Q3 and the

44

second columnisbased on 1959:Q1-98:Q3. Notice
that none of the correlations for the TP1 price based
on the longer sample are significantly different from
zero. This means that the price of the broadest mea­
sure of investment is essentially acyclical. There is
some evidence of countercyclical movements in this
price for the shorter sample, although the correlations
in this case are generally not very large in absolute
value or statistically significant.
The cyclical behavior of prices for the narrower
investment aggregates displayed in figure 5 reveals

Economic Perspectives

that the lack of any distinct cyclical pattern for the
price of TP1 masks interesting differences between
the prices of nonresidential and residential goods.
Over the longer sample, the nonresidential price is es­
timated to be essentially acyclical, but the residential
price is clearly procyclical. Over the shorter sample
the nonresidential investment price is clearly counter­
cyclical and the residential price remains procyclical.
The difference in the estimated cross-correlogram for
nonresidential over the two sample periods turns out
to be due to differences in the behavior of the price

Federal Reserve Bank of Chicago

of PDE in the 1950s compared with the later sample
period (see figure 3).
The evidence in figure 5 suggests two things.
First, the cyclical behavior of investment good prices
depends to some extent on the sample period exam­
ined. Second, considering a broad investment aggregate
masks potentially interesting cyclical characteristics of
more narrowly defined investment good prices. Figures
6 and 7 try to uncover whether the cyclical behavior
of nonresidential and residential prices also masks
different cyclical behavior among the subcomponents

45

of these broad investment aggregates. These figures
display price-output cross-correlograms for the main
subcomponents of nonresidential and residential. Due
to data availability, the sample period for estimating
the correlations is 1959:Q1-98:Q3.
The first column in figure 6 pertains to NRS and
its main subcomponents, nonresidential buildings,
utilities, and mining. The price of NRS is significantly
countercyclical. This appears to be mainly driven by
the price of utilities and mining. The second column
of figure 6 pertains to PDE and its main subcomponents,
information and related equipment, industrial equip­
ment, and transportation equipment. There are two
observations to make here. First, the price of PDE is
strongly and significantly countercyclical. The con­
temporaneous (k = 0) correlation is -0.63 with a stan­
dard error of 0.03. The largest correlation in absolute
value is for k = 2, indicating that this price lags output
by about two quarters, about the same as the quantity
of PDE (see figure 4). The second observation is that
the prices of the main components of PDE behave al­
most identically: They are strongly and significantly
negatively correlated with output and lag output by
about two quarters. The behavior of the industrial
equipment price is particularly striking, given that the
long-run behavior of this price is so different from that
of the other two subcomponents of PDE (see figure 2).
Figure 7 is constructed similarly to figure 6, with
RSE and its subcomponents in the first column and
CD and its subcomponents in the second column.
This figure shows that prices of RSE are generally
procyclical and prices of CD goods are mostly acycli­
cal. The behavior of RSE is driven mostly by the cycli­
cality of single and multifamily structures. Interestingly,
despite the fact that investment in RSE tends to lead
output over the business cycle, the real price of RSE
and its components lags output. The real price of CD
is driven mostly by motor vehicles and other. Of the
subcomponents of CD, only the furniture price dis­
plays significant countercyclicality.
Summary ofthe evidence
The key features of the evidence presented in
this section can be summarized as follows. First, there
is strong evidence of a downward trend in the price
of investment goods in terms of consumption goods.
This downward trend is concentrated among compo­
nents of PDE and CD. Second, the broadest category
of investment, TPI, displays little distinct cyclical
variation over the sample period 1947:Q1-98:Q3, but
is moderately countercyclical in the later period,
195 9: Q1 -9 8: Q 3. If we are willing to ab st ract from the
1950s, say because of the dominating influence of the
Korean war, then it seems reasonable to say that the

46

price of the broadest component of investment is
weakly countercyclical. Certainly it is difficult to make
the case that this price is procyclical, regardless of the
sample period considered.
Many components of TPI display distinct cyclical
characteristics, even if we include the 1950s. The prices
of the two main components, nonresidential and resi­
dential, behave differently. The former is significantly
countercyclical and the latter is significantly procycli­
cal. The behavior of the nonresidential price is domi­
nated by the PDE price. The PDE price is strongly
countercyclical, as are the prices of all its subcompo­
nents. The price of NRS is mildly countercyclical, but
this pattern is not shared by all its subcomponents.
The behavior of the residential price is dominated by
RSE prices, which are strongly procyclical. CD prices
are acyclical or weakly countercyclical.

Implications for growth and the
business cycle
How does the trend and cycle behavior of invest­
ment goods prices presented above challenge con­
ventional views about growth and business cycles?
Next, I discuss various attempts to reconcile theory
with the evidence and some empirical work that sheds
light on the plausibility of competing theories.

Growth theory
Recent years have seen an explosion of theoreti­
cal and empirical research into economic growth.12 On
the theoretical side, two leading classes of models of
the determinants of economic growth have emerged.
The first is based on the accumulation of human
capital and follows from the work of Lucas (1988).
Human capital consists of the abilities, skills, and
knowledge of particular workers. The basic idea behind
this view of economic growth is that it is fundamen­
tally based on improvements in the stock of human
capital of workers over time. This view of growth holds
that, other things being equal, the larger is the stock
of human capital of workers, the more productive they
are. This means that one expects an improvement in
the stock of human capital to increase the amount
of output of any good that can be produced for a
fixed quantity of workers and capital In this sense,
growth due to the accumulation of human capital
has a homogeneous impact on the economy’s ability
to produce goods.
The second leading class of models focuses on
research and development. Pioneering work along
these lines includes Romer (1990), Grossman and
Helpman(1991), and Aghion and Howitt (1992). One
of the key insights of this literature is that growth can
emerge if there are nondecreasing returns to produced

Economic Perspectives

FIGURE 6

Business cycle correlations between nonresidential prices (t) and output (/-A)
A. Nonresidential structures

E. Producer durable equipment

k
B. Nonresidential buildings

k
F. Information

k

k
G. Industrial

C. Utilities

k
D. Mining

k
Note: Black lines are point estimates of correlations for the indicated series;
Source: See figure 1.

Federal Reserve Bank of Chicago

k
H. Transportation

k
lines are a two-standard-error confidence band.

47

FIGURE 7

Business cycle correlations between residential prices (t) and output (/-A)
A. Residential structures

E. Consumer durables

k
B. Single family structures

k
F. Motor vehicles

k
C. Multifamily structures

k
G. Furniture

k
D. Other structures

k
Note: Black lines are point estimates of correlations for the indicated series;
Source: See figure 1.

48

k
H. Other

k
lines are a two-standard-error confidence band.

Economic Perspectives

factors of production (such as knowledge or capital,
but not labor).13 The bottom line of this theory is sim­
ilar to that of the human capital models. Improvements
in technology due to research and development usually
increase the productivity of all factors of production.
Consequently, if there is such an improvement in
technology, more of all goods can be produced with
a fixed quantity of capital and labor. Again, techno­
logical change is assumed to have a homogeneous
impact on produced goods.
The evidence on trends in investment good prices,
particularly the trend in the price of PDE, challenges
these views of growth, because it strongly suggests
that there have been substantial improvements in tech­
nology that have affected one kind of good but not
another. Specifically, the data suggest that the quality
and technology of capital goods production have
advanced almost nonstop since the end of World
War II. Why do the data suggest this? Assuming
that the prices and quantities of PDE are correctly
measured, the real price of PDE measures how many
(constant quality) consumption goods need to be
sold in order to raise the funds to purchase one (con­
stant quality) unit of PDE. If this price has been fall­
ing, then fewer and fewer consumption goods are
needed to buy a unit of PDE. This suggests that the
supply of PDE has grown relative to the supply of
consumption goods. One way the supply of PDE can
rise in this way is if the technology for producing
capital goods improves at a faster rate than that for
producing consumption goods. In this case, the same
amount of capital and labor applied to producing PDE
or consumption goods will yield more PDE than con­
sumption as time passes. That is, the supply of PDE
will grow relative to consumption goods. The basic
logic of supply and demand then dictates that the
price of PDE in terms of consumption goods must fall.
Greenwood et al. (1997) build on this intuition to
show how the trend in the relative price of PDE and
the associated increase in the share of PDE in aggre­
gate output (see table 1) can be accounted for in a
growth model in which most growth is due to capitalembodied technical change. In addition, the authors
argue that other potential explanations for the price
and quantity trends are implausible or boil down to
essentially the same explanation.14
Greenwood et al. (1997) apply their model of
growth to reevaluate conventional estimates of the
importance of technological change in improving
standards of living. This line of research is called
growth accounting. The effects of technical change
using standard models, like the ones briefly described
above, can be summarized by multifactor productivity,

Federal Reserve Bank of Chicago

which is also called the Solow residual. Multifactor
productivity is an index of the quantity of aggregate
output that can be produced using a fixed quantity of
(quality-adjusted) capital and labor. The higher the
multifactor productivity, the more output can be pro­
duced. Traditionally, most of growth is viewed as be­
ing due to improvements in multifactor productivity.
Greenwood et al. (1997) use their model to show that
approximately 60 percent of all improvements in pro­
ductivity can be attributed to capital-embodied tech­
nical change, while the multifactor productivity index
accounts for the rest. This says that capital-embodied
technical change is a fundamental part of growth.

Business cycle theory
To assess the cyclical evidence on relative prices,
we need to understand how various shocks to the
economy might influence the cost of investment goods
compared with consumption goods. Figure 8 displays
a production possibilities frontier (PPF) for consump­
tion and investment goods. The PPF depicts the vari­
ous quantities of consumption and investment goods
that can be produced if capital and labor are fully
employed and used efficiently. The shape of the fron­
tier reflects the fact that, holding fixed the quantity of
labor and capital employed in producing goods, it is
costly to shift production toward either producing
more consumption goods or more investment goods.15
This is reflected in the figure by the increase in the
(absolute value of the) slope of the frontier as one
moves from the upper left to the lower right. In a com­
petitive equilibrium, the slope of the frontier equals the
relative price of the goods. Hence, as more investment
goods are produced, the relative price of investment
goods rises.
The PPF summarizes the supply side of the econ­
omy. The actual price in a competitive equilibrium is
determined by the interaction of the demand for con­
sumption and investment goods with the supply. Sup­
pose that the demand for consumption and investment
goods dictates that the quantity of consumption goods
and investment goods actually produced is given by
Co and IQ in figure 8. Now, suppose a Keynesian de­
mand shock—for example, an increase in the money
supply which lowers interest rates—increases the
demand for investment goods relative to consumption
goods. Since this is a demand shock, the PPF in figure
8 does not change. The change in demand leads to a
movement down the frontier, say to a point where
consumption and investment are given by Ct and .
Since the slope of the frontier is steeper at this point,
the relative price of investment goods must rise. If
aggregate output is driven by shocks to investment

49

demand, then the price of investment goods is predicted
to be procyclical.
An aggregate supply shock has a similar implica­
tion. The conventional assumption about these kinds
of shocks is that they raise multifactor productivity
and influence all produced goods symmetrically. This
is shown in figure 9 as a proportional shift out in the
solid line PPF to the dashed line PPF. The dashed line
PPF has been drawn so that its slope is identical to
the slope of the solid line PPF along a straight line
from the origin. This means that if the ratio of consump­
tion to investment goods produced before and after
the technology shock is constant, then the relative
price of investment goods will be unchanged. How­
ever, this is not what is predicted in standard models.
These models say that when a good technology shock
arrives, which raises the productivity of all factors of
production, the optimal response of individuals is to
smooth consumption. That is, not have consumption
change too much in the short run. The result of this
is that investment rises more than consumption. In
figure 9, this is represented by consumption and invest­
ment changing from Co and 70 before the productivity
shock to Cj and Ix after the shock. It follows that the
price of investment goods must rise in this case as well.
Since output also rises with a positive technology
shock, the price of investment goods is predicted
to be procyclical.16
In view of the cyclical evidence presented earlier,
these model predictions are problematic. They are con­
sistent with the behavior of residential investment, but
inconsistent with the behavior of the other major com­
ponents of investment and the broadest measure, TPI.
Why are investment goods prices not procyclical?
The two leading explanations involve assumptions
about the technology for producing investment goods.
One is based on increasing returns to scale in the
production of investment goods (but not consumption
goods). The other is based on a variation in the rate
of capital-embodied technical change. The increasing
returns view assumes that the more investment goods
that are produced, the less costly it is to produce a
unit of investment goods. One way to represent this
is shown in figure 10, which displays a pseudo-PPF.17
Notice that the shape is different from figures 8 and
9. Now when more investment goods are produced
relative to consumption goods, the price of investment
goods falls. In this case, both aggregate technology
shocks and Keynesian demand shocks can lead to
countercyclical relative prices.
To understand the embodied technology view,
consider an increase in the productivity of producing
investment goods that has no direct impact on the

50

production of consumption goods. This could take
the form of improvements in the efficiency of produc­
ing investment goods. It could also take the form of
an improvement in the quality of investment goods
produced so that a given quantity of capital and labor
can produce a higher quantity of quality-adjusted
goods. Either way, we can represent the change in
technology as in figure 11. The improvement in tech­
nology is shown by the shift from the solid to the
dashed frontier. Along the dashed frontier, for each
quantity of consumption goods produced, more invest­
ment goods can be produced. Moreover, along any
straight line from the origin, the slope of the dashed
frontier is flatter than the solid frontier. That is, for
any fixed ratio of consumption to investment goods,
the investment goods are cheaper in terms of consump­
tion goods after the change in technology. Now, after
the increase in technology, there will be a shift in favor

Economic Perspectives

of the production of investment goods. If this shift is
strong enough, the movement along the dashed fron­
tier could in principle raise the investment good price.
In practice, this does not happen. Since aggregate
output rises after this kind of technology shock, if
business cycles are in part driven by this kind of
disturbance, then investment good prices could be
countercyclical.

Evaluating the theories
Beyond the work of Greenwood et al. (1997), little
has been done to evaluate the plausibility of the capitalembodied technological change theory of the trend evi­
dence on investment prices. However, more work has
been done to evaluate the differing views on the cycli­
cality of investment good prices.
Generally, the empirical evidence seems to go
against the increasing returns interpretation of the cycli­
cal evidence on prices. Harrison (1998) examines annual
data on capital, labor, and value added in various indus­
tries in the consumption good sector and the invest­
ment good sector. She finds some empirical support
for increasing returns associated with capital and la­
bor in the production of investment goods. However,
she does not find a sufficient degree of increasing re­
turns to generate increasing returns in the factor of
production, labor, that is variable in the short run. Con­
sequently, the work does not support the increasing
returns view. Other research on measuring increasing
returns focuses on the manufacturing sector. Basu and
Fernald (1997), Burnside (1996), and Burnside, Eichenbaum, andRebelo (1995) have overturned previous
empirical claims of increasing returns in the manufac­
turing sector, including capital equipment industries.
Other empirical work attempts to address a key
implication of the increasing returns view—that the
supply curve for investment goods slopes down.
That is, holding other things constant, the cost of
investment goods is diminishing in the quantity of
investment goods produced. Shea (1993), in a study
of many sectors of the economy, uses instrumental
variables econometric techniques to distinguish supply
shocks from demand shocks to trace out the slope of
supply curves. The author’s main conclusion is that,
broadly speaking, supply curves slope up. Goolsbee
(1998) focuses specifically on the supply of capital
goods and uses a series of “natural experiments”
(involving periodic changes in federal laws providing
for investment tax credits) to identify a disturbance
that affects the demand for investment goods but
not the supply. He finds clear evidence of an upward
sloping investment supply curve. To summarize,
empirical work on the sign of the slope of the invest­
ment good supply schedule finds that it is positive.

Federal Reserve Bank of Chicago

Other research assesses the plausibility of the
embodied technology view. Christiano and Fisher
(1998) and Greenwood et al. (1998) evaluate business
cycle models in which a major driving force for fluc­
tuations is variations in capital-embodied technical
change. They test the embodied technology view by
examining the ability of their models to account for
various business cycle phenomena. Both studies find
that their models do about as well as other business
cycle models in accounting for business cycle phenom­
ena. As a measure of the importance of capital-embod­
ied technical change as a driving force for business
cycles, Greenwood et al. (1998) find that about 30 per­
cent of business cycle variation in output can be attrib­
uted to this kind of shock. Christiano and Fisher (1998),
in a very different model, find that about three-quarters
of output fluctuations are due to this shock. Either
way, the evidence suggests that variation in the rate

51

of technical change embodied in capital equipment
accounts for a significant proportion of business
cycle variation in output.

New evidence
Some new research attempts to distinguish the
increasing returns view from the embodied technology
view of the cyclical behavior of investment good prices.
This evidence is based on two econometric procedures
designed to identify disturbances to the aggregate
economy that influence the demand for investment
goods, but leave supply unchanged. The specific
shocks considered are an exogenous increase in gov­
ernment purchases (that is an increase in government
purchases that is unrelated to developments in the
economy) and an exogenous monetary contraction.
In the government spending case, the idea is to
investigate how particular investment quantities and
prices respond to an exogenous increase in government
purchases. The exogenous increase in government
spending takes the form of a large military buildup
(specifically the Korean war, the Vietnam war, and the
Carter-Reagan buildup.) The methodology is identical

52

to that employed by Eichenbaum and Fisher (1998).18
Figure 12 displays the estimates, which are based on
quarterly data for 1947:Q1-98:Q3. The first row offigure 12 plots the response to an exogenous increase in
government purchases of real investment in PDE and
RSE (solid lines) along with a 68 percent confidence
band (colored lines). The second row plots the corre­
sponding relative price responses. Interestingly, PDE
investment rises and RSE investment falls.19 Under
the increasing returns view, we would expect the PDE
price to fall and the RSE price to rise. The second row
of plots indicates that the RSE price response is incon­
sistent with the increasing returns view, while the PDE
price response seems to confirm it.
The monetary shocks case examines how quanti­
ties and prices of PDE and RSE respond to an estimate
of a contractionary monetary disturbance. The meth­
odology is standard20 and has been summarized by
Christiano (1996) (see also Christiano, Eichenbaum,
and Evans, 1999). The estimated responses (along
with a 95 percent confidence interval) are presented
in figure 13. Looking at the quantities in the first row
of plots, notice that both PDE and RSE fall after an

Economic Perspectives

exogenous monetary contraction. Under the increasing
returns view, one would expect the prices of both
investment goods to rise. Studying the second row
of plots, we see that the PDE price response is not
significantly different from zero and the RSE price
drops significantly.
Taken together, the evidence on the responses
of RSE prices and quantities to government spending
and monetary shocks goes against the increasing
returns view. It conforms to a standard neoclassical
view of investment, in the sense that it is consistent
with the discussion of the production possibilities
frontier in figure 8. Of course, the increasing returns
view is really intended to apply to PDE investment.
The responses of PDE prices and quantities provide
mixed signals. The responses to a monetary shock
provide evidence neither for nor against increasing
returns, since the quantity falls but the price response
is not very precisely estimated and could be either
positive, negative, or zero. The responses to a gov­
ernment spending shock might be viewed as evidence
in favor of increasing returns. However, one interpre­
tation of the PDE price response in this case is that it

Federal Reserve Bank of Chicago

is dominated by the Korean war military buildup. This
occurred just after World War 11, when military spend­
ing had fallen from very high levels. The increasing
returns that could support a lower price with higher
investment might conceivably be due to the resump­
tion of large-scale production at facilities that had
been operating far below minimum efficient scale. If
this is true, it seems more like a special case than an
enduring feature of the U.S. economy.

Conclusion
In this article, 1 have presented evidence on trends
and business cycle variation in the prices of invest­
ment goods relative to nondurables and services
consumption. This evidence seems to go against
conventional views of both business cycles and
growth. How can one reconcile theory with the evi­
dence? The leading views include one based on
increasing returns to scale in the production of invest­
ment goods and another based on capital-embodied
technical change. While some of the evidence 1 pre­
sented could be viewed as supporting the increasing
returns view, generally, there is little empirical support

53

for increasing returns. At this point, then, the leading
candidate to reconcile theory with the data appears
to be the one based on capital-embodied technical
change, that is, the embodied technology view.
This conclusion has implications for our under­
standing of growth and business cycles, future research
on these subjects, and policy. The prospect of a com­
prehensive theory of growth and business cycles is
appealing because of its simplicity. Disembodied tech­
nical change has gained credence for its supposed
ability to account for growth and business cycles. Yet,
the theory of business cycles based on disembodied
technology has always been problematic because the
shocks are hard to interpret. The growth accounting
results of Greenwood et al. (1997) bring into question
the growth implications of this theory as well. In the
search for a comprehensive theory of growth and
business cycles, then, advances in capital-embodied
technology seem to offer a promising alternative. In
addition, they provide a much more tangible notion of
growth. These considerations suggest that future
research on growth and business cycles that empha­
sizes capital-embodied technical change may be fruitful.
If growth and business cycles are originating from
changes in capital-embodied technology, then the mod­
els we use for policy analysis have to incorporate this
and, consequently, policy recommendations could
change. To the extent that technological change is em­
bodied in capital equipment, government policies that
affect equipment investment could have a dual impact
on growth via the quality and the quantity of capital
goods. This could mean, for example, that investment

tax credits directed toward improvements in the effi­
ciency of capital equipment could have a significant
impact on growth. More research is required to uncover
the full implications of this.
The implications for stabilization policy ofthe
embodied technology view are less obvious. The fact
that it seems to supplant the increasing returns view
means that the arguments for interventionist stabili­
zation policy that this view supports are less compel­
ling. For example, increasing returns could provide
scope for policy intervention, because it either involves
externalities or is inconsistent with perfect competition.
Moreover, it makes animal spirits models more plausi­
ble, which also has implications for stabilization policy
(see, for example, Christiano and Harrison, 1999). The
embodied technology view is more in line with the real
business cycle tradition, in which policy interventions
are counterproductive. Real business cycle theory says
that the business cycle is largely the result of optimal
behavior by individuals in the economy interacting,
for the most part, in perfectly competitive markets. Any
policy interventions in such an environment tend to
reduce overall welfare. To the extent that the embodied
technology view is more compelling than previous in­
carnations of real business cycle models, it lends
greater support to the argument that interventionist
stabilization policy cannot improve the well-being of
any individual in the U.S. economy without hurting
some other individual. Of course, this still leaves
open the possibility that equity considerations might
be used to defend interventionist stabilization policy.

NOTES
Equivalently, higher quality goods of all kinds can be produced with
the same amount of capital and labor. As described in more detail below,
new models of endogenous growth have reduced forms, which have sim­
ilar implications for growth accounting to those of models written in
terms ofexogenous disembodied technical change.

Examples of textbooks that emphasize the IS LM model are Abel and
Bernanke (1997), Gordon (1998), Hall and Taylor (1997), and Mankiw
(1997).
3For a survey of theories based on animal spirits, see Farmer (1993).

4A good summary of this view is Prescott (1986). For a discussion of
how this view can be used to explain the 1990-91 recession, see Hansen
and Prescott (1993).

5This section relies heavily on Christiano and Fitzgerald (1998,
pp. 58-59).
6This is the empirical counterpart to investment as it is usually defined
in the real business cycle literature.
7The aggregation in this table is identical to the aggregation used by the
BEA, except for “residential,” which is calculated as the chain-weighted
aggregate of “residential structures and equipment” and “consumer du­
rable.” See box 1 for the chain-weighting procedure.

54

Eor TPI and GDP,y, the nominal shares in the first row are
PPPIqPFI/(Pytqy} and the real shares are qPPIlqy. Nominal and real shares
for investment good i in the other rows are given by Pt'q‘ I(PtTPIqtTPP) and
the real shares areqlt/q^PI-

9In the notation used above, the black lines are (the natural logarithm of)
p‘ for z corresponding to the 20 types of investment listed in table 1 over
the period for which data are available.
10Many ofthe trends evident in figure 2 are not apparent in theNIPA
fixed-weighted constant 1982 dollar and earlierNIPA data. In a very
influential book, Gordon (1989) argued that the conventional BEA treat­
ment ofinvestment good quality severely underestimated the degree of
quality change in investment goods. His analysis was the first to show
that there is a substantial downward trend in the prices of PDE and CD.
The BEA now incorporates many ofthe adjustments for quality change
advocated by Gordon (1989).

11 The procedure used to extract the business cycle component ofthe rela­
tive price data involves the application of a linear filter. This, combined
with the fact that this filter is applied to the natural logarithm of the rela­
tive prices, implies that the business cycle component of each relative
price is the business cycle component ofthe relevant investment deflator
minus the business cycle component ofthe consumption deflator.

Economic Perspectives

12For a comprehensive review of this literature, see Barro and
Sala-i-Martin (199 5).
13The assumption of constant returns to scale is usually based on a repli­
cation argument. A fixed quantity ofcapital and labor applied to produce
x amount of some good can always be applied again to produce another x
of the good. That is, increasing the quantity of factors of production by
some proportion changes the amount produced by the same proportion.
This argument seems harder to apply in the case of technology. For exam­
ple, suppose a group ofresearchers have discovered a newprocess for mak­
ing steel. Ifanother group ofresearchers make the same di scovery, there
is no net improvement in knowledge. In this case, there would be de­
creasing returns. On the other hand, fixed costs or advantages to having
many researchers working on similar projects may mean that increasing
returns to scale are important in the process of knowledge creation.

14Greenwood et al. (1997) show how the research and development and
human capital classes of models can be used to account for the evidence,
if these activities have a disproportionate impact on the production of
equipment compared with consumption goods. Two explanations they
consider differ fundamentally from their basic story. They both involve a
two-sector interpretation of the evidence, in which equipment and con­
sumption goods are produced in separate sectors (using separate produc­
tion functions). In one case, the production functions have different factor
shares, that is, the different goods require capital and labor in different
proportions to produce a unit of the good. The authors conclude that the
“prospect for explaining the relative price decline with a two-sector
model based on differences in share parameters looks bleak, given the
implausibly large differences required in the structure ofproduction
across sectors (p. 358).” The other explanation involves an externality
in the production of investment goods. Specifically, the productivity of
factors in the investment good sector is increasing in the quantity of in­
vestment goods along the lines described in Romer (1986). Greenwood
et al. (1997) show that this explanation can, in principle, account for the
trend evidence. However, this theory relies on an externality which is
difficult to identify empirically. Some evidence on increasing returns to
scale, which the production externality implies, is discussed below.
Generally, there is little empirical support for this view.
15 The shape ofthe frontier can be justified by standard neoclassical assump­
tions about how goods are produced, in particular that they are produced
using constant returns to scale production functions in labor and capital
and that it is costly to transfer labor and/or capital across sectors produc­
ing consumption goods and sectors producing investment goods. Note
that adjustment costs in the installation of investment goods affect the rel­
ative price ofinstalled capacity, not the relative price of investment goods.
16This discussion assumes that the shares of factors in production are
identical in producing consumption and investment goods and/or that
there are costs ofadjusting factors ofproduction across sectors. It is pos­
sible for the price of investment goods to be countercyclical in this type

of model if the share of labor in production is greater in the consumption
sector than in the investment goods sector. As long as factors of produc­
tion are perfectly mobile across sectors (that is, there are no costs to
shifting factors across sectors), an increase in technology lowers the
price of investment goods in this case. Factor shares are difficult to mea­
sure, so assessing the plausibility of this possibility is difficult. Howev­
er, the Greenwood et al. (1997) results for long-run trends suggest that
the differences in factor shares required to reconcile the empirical evi­
dence on prices with this explanation may be implausible. Also, it is
implausible to assume that there are no costs of shifting factors of pro­
duction across sectors.
17This frontier does not necessarily reflect true technological possibili­
ties, but takes into account the restrictions on individual decisionmak­
ing, such as individuals not internalizing a production externality, such
that the points on the frontier are consistent with optimizing behavior
ofproducers.
18The methodology is identical to that employed by Eichenbaum and
Fisher (1998). This methodology uses four variables, in addition to the
investment good quantity and price variables, in a vector autoregres­
sion, along with a dummy variable which takes on the value zero at all
dates except 1950:Q3,1965:Q1 and 198O:Q1, in which cases the vari­
able equals unity. These dates correspond to the beginning of three large
military buildups. The key identifying assumption is that these build­
ups were exogenous events. For further discussion, see Edelberg,
Eichenbaum, and Fisher (1999). The four variables are the log level of
time t real GDP, the net three-month Treasury bill rate, the log of the
Producer Price Index ofcrude fuel, and the log level of real defense pur­
chases, g Six lags were used. The plotted responses in figure 12 corre­
spond to the average response of the indicated variable across the three
military buildup episodes, taking into account the endogenous varia­
tion in the variable.
19See Edelberg, Eichenbaum, and Fisher (1999) for a discussion of how
this evidence can be explained within the context ofa standard neoclas­
sical model.

20Technically, I estimate a vector autoregression in the deflator for non­
durables and services, real GDP, an index of changes in sensitive materi­
als prices, the federal funds rate, plus the investment price and quantity I
am interested in. All variables except the federal funds rate are first
logged. The impulse response functions in figure 13 correspond to an
orthoganalized innovation in the federal funds rate. The orthoganalization procedure assumes the order of the vector autoregression is the same
as listed in the text and a triangular decomposition. Ordering is not im­
portant for the investment responses as long as standard assumptions are
made about the variables that precede the federal funds rate in the order­
ing (see Christiano, Eichenbaum, and Evans, 1999). Finally, the stan­
dard errors are computed using the procedure described by Christiano,
Eichenbaum, and Evans (1999).

REFERENCES

Abel, Andrew, and Ben Bernanke, 1997, Macroeco­

Burnside, Craig, 1996, “Production function regres­

nomics, 3rd edition, New York: Addison-Wesley.

sions, returns to scale, and externalities,” Journal of
Monetary Economics, Vol. 37, No. 2, April, pp. 177-201.

Aghion, Philippe, and Peter Howitt, 1992, “A model
of growth through creative destruction,” Economet­
rica, Vol. 60, No. 2, March, pp. 323-351.

Burnside, Craig, Martin Eichenbaum, and Sergio
Rebelo, 1995, “Capacity utilization and returns to

Barro, Robert, and Xavier Sala-i-Martin, 1995,
Economic Growth, New-' York: McGraw-Hill.

scale,” in NBER Macroeconomics Annual 1995, Vol.
10, Ben Bernanke and Julio Rotemberg (eds.), Cam­
bridge, MA: MIT Press.

Basu, Susanto, and John Fernald, 1997, “Returns to

Christiano, Lawrence J., 1996, “Identification

scale in U.S. production: Estimates and implications,”
Journal of Political Economy, Vol. 105, No. 2, April,
pp. 249-283.

Federal Reserve Bank of Chicago

and the liquidity effect: A case study,” EconomicPerspectives, Federal Reserve Bank of Chicago, Vol.
20, No. 3, May/June, pp. 2-13.

55

Christiano, Lawrence J., Martin Eichenbaum, and
Charles Evans, 1999, “Monetary policy shocks: What
have we learned and to what end?,"' in Handbook of
Macroeconomics. John Taylor and Michael Woodford
(eds.), Amsterdam: Elsevier Science, forthcoming.

Christiano, Lawrence J., and Jonas D. M. Fisher,
1998, “Stock market and investment good prices:
Implications for macroeconomics, ' Federal Reserve
Bank of Chicago, working paper.

Christiano, Lawrence J., and Terry Fitzgerald, 1998,
“The business cycle: It’s still a puzzle," EconomicPerspectives. Federal Reserve Bank of Chicago, Vol.
22, No. 4, Fourth Quarter, pp. 56-83.

Christiano, Lawrence J., and Sharon Harrison,
1999, “Chaos, sunspots, and automatic stabilizers,"
Journal of Monetary Economics, forthcoming.

Edelberg, Wendy, Marty Eichenbaum, and Jonas D.
M. Fisher, 1999, “Understanding the effects of a
shock to government purchases," Review of Eco­
nomic Dynamics. Vol. 2, forthcoming.

Eichenbaum, Marty, and Jonas D. M. Fisher, 1998,
“How- does an increase in government purchases
effect the economy?, ' Economic Perspectives.
Federal Reserve Bank of Chicago, Vol. 22, No. 3,
Third Quarter, pp. 29-43.

Farmer, Roger, 1993, The Macroeconomics of SelfFulfilling Prophecies. Cambridge: MIT Press.

Hall, Robert, and John Taylor, 1997, Macroeconom­
ics, 5th edition, New-' York: W. W. Norton.

Hansen, Gary, and Edward Prescott, 1993, “Did tech­
nology shocks cause the 1990-91 recession?," Amer­
ican Economic Review, Vol. 83, No. 2, May, pp. 280-286.
Harrison, Sharon G., 1998, “Evidence on the empiri­
cal plausibility of externalities and indeterminacy in a
two sector model," Barnard College, Columbia Uni­
versity, unpublished manuscript.

Hornstein, Andreas, and J. Praschnik, 1997, “Inter­
mediate inputs and sectoral comovement in the busi­
ness cycle," Journal of Monetary Economics, Vol. 40,
No. 3, December, pp. 573-595.

Huffman, Gregory, and Mark Wynne, 1999, “The
role of intratemporal adjustment costs in a multisec­
tor economy, ' Journal of Monetary Economics,
forthcoming.

Krusell, Per, Lee Ohanian, Victor Rios-Rull, and
Giovanni Violante, 1995, “Capital-skill complementar­
ity and inequality,' University of Rochester, manu­
script.

Landefeld, J. Steven, and Robert P. Parker, 1997,
“BEA’s chain indexes, time series, and measures of
long-term economic growth," Survey of Current
Business, pp. 58-68.

Lucas, Robert E., 1988, “On the mechanics of econom­

Fisher, Jonas D. M., 1997, “Relative prices, comple­

ic development," Journal of Monetary Economics,
Vol. 22, No. I. July pp. 3-42.

mentarities, and comovement among components of
aggregate expenditures," Journal of Monetary Eco­
nomics, Vol. 39, No. 3, August, pp. 449-474.

New York: Worth.

Goolsbee, Austan, 1998, “Investment tax incentives,
prices, and the supply of capital goods," Quarterly
Journal of Economics, Vol. 113, No. 1, February, pp.
121-148.
Gordon, Robert, 1998, Macroeconomics, 7th edition,
New York: Addison-Weslev.
___________ , 1989, The Measurement ofDurable
Goods Prices, Chicago: University of Chicago Press.

Greenwood, Jeremy, and Zvi Hercowitz, 1988,
“Investment, capacity utilization, and the real business
cycle," American Economic Review, Vol. 78, No. 3,
June, pp. 402-417.

Greenwood, Jeremy, Zvi Hercowitz, and Per Krusell,

Mankiw, Gregory, 1997, Macroeconomics, 3rd edition,

Murphy, Kevin M., Andrei Shleifer, and Robert W.
Vishny, 1989, “Building blocks of market clearing
business cycle models," in NBER Macroeconomics
Annual 1989, Olivier J. Blanchard and Stanley Fischer
(eds.), Cambridge, MA: MIT Press.

Prescott, Edward, 1986, “Theory ahead of business
cycle measurement," Quarterly Review, Federal Re­
serve Bank of Minneapolis, Fall, pp. 9-22.

Romer, Paul, 1990, “Endogenous technical change,"
Journal of Political Economy, Vol. 98, No. 5, Part 2,
October, pp. S71-S102.
____________ , 1986, “Increasing returns and long-run
growth," Journal of Political Economy, Vol. 94, No.
5, October, pp. 1002-1037.

1998, “The role of investment-specific technological
change in the business cycle," University of Roches­
ter, manuscript, January.

Quarterly Journal of Economics, Vol. 108, February,
pp. 1-32.

___________ , 1997, “Long-run implications of invest­
ment-specific technological change," American
Economic Review, Vol. 87, No. 3, June, pp. 342-362.

Solow, Robert, 1957, “Technical change and the ag­
gregate production function," Review of Economics
and Statistics, Vol. 39, pp. 312-320.

Shea, John, 1993, “Do supply curves slope up?,"

Grossman, Gene, and Elhanan Helpman, 1991, Inno­
vation and Growth in the Global Economy, Cam­
bridge, MA: MIT Press.
56

Economic Perspectives