Cementing Relationships: Vertical Integration, Foreclosure, Productivity, and Prices
Author(s): Ali Hortaçsu and Chad Syverson
Source: Journal of Political Economy , April 2007, Vol. 115, No. 2 (April 2007), pp. 250-
301
Published by: The University of Chicago Press
Stable URL: http://www.jstor.com/stable/10.1086/514347
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms
The University of Chicago Press is collaborating with JSTOR to digitize, preserve and extend
access to Journal of Political Economy
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
250
[ Journal of Political Economy, 2007, vol. 115, no. 2]
᭧ 2007 by The University of Chicago. All rights reserved. 0022-3808/2007/11502-0003$10.00
Cementing Relationships: Vertical Integration,
Foreclosure, Productivity, and Prices
Ali Hortac¸su and Chad Syverson
University of Chicago and National Bureau of Economic Research
This paper empirically investigates the possible market power effects
of vertical integration proposed in the theoretical literature on vertical
foreclosure. It uses a rich data set of cement and ready-mixed concrete
plants that spans several decades to perform a detailed case study.
There is little evidence that foreclosure is quantitatively important in
these industries. Instead, prices fall, quantities rise, and entry rates
remain unchanged when markets become more integrated. These
patterns are consistent, however, with an alternative efficiency-based
mechanism. Namely, higher-productivity producers are more likely to
vertically integrate and are also larger, more likely to survive, and more
likely to charge lower prices. We find evidence that integrated producers’
productivity advantage is tied to improved logistics coordination
afforded by large local concrete operations. Interestingly, this
benefit is not due to firms’ vertical structures per se: nonvertical firms
with large local concrete operations have similarly high productivity
levels.
We thank Daron Acemoglu, Tim Bresnahan, Jeremy Fox, John Haltiwanger, Justine
Hastings, Tom Holmes, Tom Hubbard, Eddie Lazear, Canice Prendergast, Lynn Riggs,
Mike Riordan, Mark Roberts, Nancy Rose, Kathryn Shaw, Mike Whinston, numerous seminar
participants, and anonymous referees for helpful comments. Jeremy Shapiro provided
valuable research assistance. Hortac¸su thanks the Center for Industrial Organization at
Northwestern University for its hospitality, and Syverson is grateful for financial support
from the National Science Foundation (award SES-0519062), the Olin Foundation, and
the Stigler Center. The research in this paper was conducted while the authors were
Special Sworn Status researchers of the U.S. Census Bureau at the Chicago CensusResearch
Data Center. Research results and conclusions expressed are those of the authors and do
not necessarily reflect the views of the Census Bureau. This paper has been screened to
ensure that no confidential data are revealed. Support for this research at the Chicago
Research Data Center from NSF (awards SES-0004335 and ITR-0427889) is also gratefully
acknowledged.
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
cementing relationships 251
I. Introduction
Is vertical integration a device for firms to create and harness market
power, or does it enhance efficiency and improve social welfare? This
paper looks at this issue in two vertically linked industries, using a rich
data set on their producers, by investigating patterns of prices, productivity,
entry, exit, and scale across integrated and unintegrated firms.
The reasons for, and results of, vertical integration have been a topic
of considerable attention since Coase’s (1937) landmark paper. Since
then, economic theories of vertical mergers have evolved both in response
to and as drivers of antitrust policy. Along with theoretical developments
in efficiency-based theories of vertical integration, recent
years have also seen the introduction of new theoretical arguments for
the possible harm that can arise through vertical integration’s foreclosure,
or market power, effect.1
This study utilizes integration episodes in the cement (Standard Industrial
Classification [SIC] 3241) and ready-mixed concrete (SIC 3273)
industries between 1963 and 1997 as an empirical laboratory to investigate
the causes and consequences of vertical mergers.2
A first look at
simple patterns in the data is instructive. Table 1 shows how the prices
and quantities sold in downstream industry (ready-mixed concrete) markets
are related to the extent of vertical integration in those markets.
The table reports coefficients obtained by regressing local average
(logged) concrete prices or total (logged) quantities, depending on the
specification, on either the market share or the number of vertically
integrated firms in the downstream market. Specifications are reported
both excluding and including market fixed effects.3
(Year fixed effects
are included in all specifications to account for aggregate movements.
We will further detail our sample, construction of the variables, and the
market definition below.)
As the table makes apparent, the greater the presence of vertically
integrated firms in a market, whether measured by their market share
or number, the lower the prices and the higher the quantities in the
1
Efficiency reasons for integration are espoused in, e.g., Williamson (1971) and Grossman
and Hart (1986). Work on foreclosure includes Salinger (1988), Hart and Tirole
(1990), Ordover, Saloner, and Salop (1990), Bolton and Whinston (1993), Riordan (1998),
and Nocke and White (forthcoming). Snyder (1995), Bernheim and Whinston (2004),
and Rey and Tirole (2007) offer comprehensive surveys.
2
While often interchanged colloquially, cement is not concrete. Cement—made by baking
limestone and clay or shale together in a kiln and grinding the result into a powder—
is a single but important ingredient in the production of concrete. Ready-mixed concrete
is produced by mixing cement with sand, gravel, water, and chemical admixtures and is
what is contained in the familiar trucks with the spinning barrels on their backs. Thus
cement is the upstream industry and ready-mixed the downstream industry.
3
Coefficients in the latter case thus reflect how concrete prices and quantities change
within a market when that market becomes more vertically integrated.
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
252 journal of political economy
TABLE 1
Market-Level Relationships between Average Prices, Total Quantities, and the
Extent of Integration
Quantity-Weighted Average
Price Total Quantity
(1) (2) (3) (4) (5) (6) (7) (8)
2
R .065 .070 .430 .429 .072 .185 .890 .891
Market share of vertically
integrated
firms
Ϫ.143*
(.029)
Ϫ.083*
(.041)
1.532*
(.290)
.331*
(.154)
Number of vertically
integrated
firms
Ϫ.037*
(.006)
Ϫ.014
(.011)
.637*
(.043)
.153*
(.034)
Market fixed
effects? No No Yes Yes No No Yes Yes
Note.—The table shows the coefficients obtained by regressing quantity-weighted average concrete prices and total
concrete quantities sold in a market on either the market share or the number of vertically integrated firms operating
in the market. The sample consists of 1,873 market-year observations. Standard errors are clustered by market.
* Significant at the 5 percent level.
local ready-mixed concrete market. These specifications are very simple
and subject to deeper investigation below, but they do suggest at a first
glance that factors other than foreclosure may be driving patterns in
the data. We will see that these results are in fact robust, with foreclosure
having little role in explaining what happens when firms in these industries
vertically integrate.
While foreclosure effects do not appear to explain the data, an efficiency-based
explanation may. Indeed, as we will show, once we explicitly
account for efficiency differences, the impact of vertical integration on
market-level outcomes ceases to be statistically and economically significant
in most of our specifications. Plant-level evidence also reveals
a strong relationship between vertical integration and productivity, especially
downstream. Integrated ready-mixed concrete plants are more
productive than unintegrated plants, even those in the same market.
What, then, is the efficiency benefit of vertical integration? We argue
in Section V that in these particular industries, the benefit might not
arise because of vertical integration per se. Instead, efficiency gains appear
to be sourced in firms’ abilities to operate multiple ready-mixed
plants in the same market. This allows firms to harness scale economies
tied to logistical coordination if they have the necessary managerial
talent: deliveries of the industry’s highly perishable product, typically
ordered on very short notice by consumers at multiple locations, can
be made more efficiently by having a central dispatch office that substitutes
production and delivery among the firm’s several local plants.
We show how this efficiency gain is consistent with several features of
our data as well as with the stated intentions and operational practices
of firms with multiple concrete plants in the same market. While these
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
cementing relationships 253
are in many cases vertically integrated firms, the evidence suggests that
they need not have a vertical structure to benefit from such gains. This
is reflected most tellingly in the evidence we document that vertically
integrated concrete plants are no more efficient than those owned by
non–vertically integrated multiunit firms with total local concrete sales
of equal size.4
The cement and ready-mixed concrete industries have several practical
features that make them favorable for a case study, and we will
discuss these further below. Perhaps the most salient motivation for
investigating these industries is that vertical integration among their
producers has been the focus of substantial policy attention. U.S. antitrust
authorities challenged the legality of several mergers in the late
1960s and early 1970s (no other industry saw as many vertical merger
cases brought), and more recent cases have been considered by competition
authorities in Europe, New Zealand, and Brazil. Thus our findings
may be interesting from a policy evaluation perspective.5
This paper builds on an empirical literature that, compared to the
resurgent theoretical literature surrounding the vertical foreclosure debate,
has been relatively thin. Grimm, Winston, and Evans (1992) study
railroad destination pairs served by a single firm and find that increased
interline competition—competing railroads that connect intermediate
points on the otherwise monopolized route—reduces welfare distortion.
Waterman and Weiss (1996) and Chipty (2001) show that U.S. cable
television systems integrated with content providers are more likely to
include their own suppliers’ paid content. Chipty, however, argues that
higher-quality programming is offered in integrated markets, resulting
in higher consumer surplus. Hastings and Gilbert (2005) utilize both
within- and across-market variation in integration status among producers
of wholesale gasoline. They examine the effect of vertical mergers
on the wholesale price paid by competing independent gasoline retailers,
and they find that the closer the competitor, the higher the wholesale
price it has to pay.6
Asker (2004) tests for foreclosure due to exclusive
dealing relationships in beer distribution and finds no significant
evidence that exclusive dealing increases market power. Rosengren and
4
As explained further in Secs. II and III, our data allow us to link plants at the firm
level. A firm that owns multiple production plants in any industry (or industries) is a
multiunit firm. Vertically integrated firms, which own concrete and cement plants specifically,
are a subset of multiunit firms.
5
It should be made clear that we are not trying to evaluate the appropriateness of the
antitrust authorities’ actions in any particular case. We are investigating all mergers between
U.S. producers in the two industries, not just those that were the subject of legal action.
6
Hastings (2004) analyzes the impact of a multimarket vertical merger on retail gasoline
prices in California. She finds that reducing the market share of unintegrated retailers
led to higher retail prices. However, she does not draw a connection between her work
and the foreclosure literature, instead ascribing the finding to product differentiation
between branded and unbranded gasoline.
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
254 journal of political economy
Meehan (1994) also find no support for foreclosure theory in event
studies of the effects of vertical merger announcements and antitrust
challenges on the stock prices of merging and rival firms. Snyder (1995),
however, uses similar methods on a different set of industries and finds
support for foreclosure.7
Section II documents cement’s and ready-mixed concrete’s patterns
of integration over the sample period. It also discusses the evolution of
economic theories of vertical integration over the same time frame,
because this academic debate was mirrored by adjustments in policies
aimed at vertical integration in the two industries. Section III describes
the data used in the analysis. In Section IV we test the implications of
foreclosure in the industry and compare foreclosure’s predictions to
those of an alternative efficiency-based explanation. We follow with a
section examining this alternative in detail (Sec. V). A short discussion
section (Sec. VI) presents conclusions.
II. Vertical Integration in Cement and Ready-Mixed Concrete:
History, Policy, and Theory
Our data span 34 years, from 1963 to 1997. Over this time the cement
and ready-mixed industries experienced two distinct periods of integration,
separated by an over decade-long period of initial disintegration
and then stability. Table 2 shows this evolution. Its first four rows report
the fraction of cement and concrete industry plants and sales accounted
for by vertically integrated producers (i.e., plants in firms owning both
cement and ready-mixed plants). Notice that these increases in vertical
integration coincide with growth in concrete plants in all multiunit firms
(those owning two or more plants, regardless of industry). While all
vertically integrated plants by our definition are in multiunit firms, other
organizational structures meet this condition as well: firms that own
several ready-mixed plants or firms owning a concrete plant(s) plus
7
The aforementioned policy actions against cement and concrete producers in the
1960s and 1970s also spurred at the time a small literature on the impact of forward
integration by cement producers. These papers preceded the recent theoretical literature
on foreclosure and as such do not directly test for such effects. They did, however, center
on the issue of whether and how integration might enhance market power. Allen (1971)
reviews evidence brought forth in the Federal Trade Commission’s 1966 report on these
industries (FTC 1966). While institutionally instructive, this study does not conduct formal
statistical tests of economic hypotheses. McBride (1983) attempts to do so, finding a
negative correlation between average cement prices in 17 markets and the cumulative
number of market ready-mixed plants acquired by cement firms. He explains these results
as resulting from the fact that forward integration into concrete makes it easier for cement
plants to “adjust” their prices to accommodate demand swings, presumably as a result of
(undocumented) collusive agreements in cement pricing that are easier to monitor than
concrete prices. However, Johnson and Parkman (1987) argue that accounting for preexisting
price trends makes this result statistically insignificant.
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
cementing relationships 255
TABLE 2
Evolution of Vertical Integration in the Cement and Ready-Mixed Concrete
Industries
1963 1967 1972 1977 1982 1987 1992 1997
Cement plants that are vertically
integrated (%) 21.9 47.4 41.9 34.8 32.5 35.2 49.5 30.5
Cement sales from vertically
integrated producers (%) 25.2 51.2 48.4 41.0 49.5 51.3 75.1 55.4
Ready-mixed plants that are
vertically integrated (%) 1.8 3.2 3.8 3.1 3.0 5.5 11.1 10.6
Ready-mixed sales from vertically
integrated producers
(%) 6.1 8.9 10.0 8.7 8.5 11.3 14.4 14.2
Ready-mixed plants in multiunit
firms (%) 24.8 26.4 32.2 34.3 35.4 41.7 49.6 55.6
Ready-mixed sales from
plants in multiunit firms
(%) 40.1 46.3 52.4 54.0 50.9 57.5 61.3 65.0
Note.—The table shows the fraction of plants (or sales) accounted for by firms of various organizational types in
the cement and ready-mixed concrete industries.
plants in another industry or industries. Thus the overall trend in vertical
integration in concrete was matched by similar horizontal consolidation.
This is a point we shall return to below when we discuss the productivity
advantage of integrated producers.
The first vertical merger wave, driven by forward integration by cement
producers, occurred in the early and mid-1960s. It received substantial
attention from antitrust authorities. The FTC brought 15 antitrust
cases during the 1960s against cement companies that had
purchased concrete firms. Each case ended in divestiture of ready-mixed
plants. In its report on these industries (FTC 1966), the commission
cited several likely anticompetitive effects of vertical mergers, including
limitation of unintegrated cement firms’ downstream market access,
increased entry costs for unintegrated concrete firms, and reduced bargaining
power for these unintegrated concrete producers with their
suppliers.
The vigorous enforcement action accompanied a chilling of merger
activity in the sector throughout the 1970s. At the same time, however,
the economic foundations of the so-called naive foreclosure theory exemplified
by the FTC report were under attack by “Chicago School”
critiques. Allen (1971), Posner (1976), and Bork (1978) pointed out
that in cases of fixed-proportions technology (as cement is for readymixed
concrete), a monopolist upstream producer cannot raise its profits
by monopolizing its downstream market.8
Thus vertical mergers
would occur only if there are efficiency gains.
8
Vernon and Graham (1971) show that if the downstream firm can substitute away
from the monopolist supplier’s input, a vertical acquisition may increase the monopolist’s
profits, though with ambiguous welfare effects.
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
256 journal of political economy
These and similar arguments influenced antitrust authorities in the
Reagan and first Bush administrations and softened official views toward
vertical mergers considerably. In 1985, the FTC explicitly eased its enforcement
policy regarding cement and ready-mixed vertical mergers.9
Industry firms responded by reintegrating. The fraction of cement plants
in vertically integrated firms rose from 32.5 percent to 49.5 percent
between 1982 and 1992, and the fraction of sales grew from 49.5 to 75.1
percent. For ready-mixed the corresponding growth for plants went
from 3.0 percent to 11.1 percent and the share of industry sales from
8.5 to 14.4 percent.10
The theoretical debate pendulum swung back in the other direction
in the late 1980s, as several authors formulated game-theoretic models
to formalize certain conditions, robust to Chicago School criticisms,
under which vertical mergers would have anticompetitive effects. An
influential example of these newer foreclosure models is the “ex post
monopolization” model of Hart and Tirole (1990).11
Their framework
has upstream homogeneous-good producers competing in prices. Importantly,
for reasons that will become clear in a moment, they are
assumed to have different marginal costs. In the cement-concrete context,
these cost differences might arise because of productivity differences
or because in a particular market a local cement producer has a
transport cost advantage over more distant producers. Without loss of
generality, one can think of the local upstream industry comprising a
monopolist U that is disciplined by the potential entry of higher-cost
competitors.
Suppose that this monopolist supplies two local downstream producers,
D1 and D2. The structure of the game is as follows: U offers each
a (possibly nonlinear) tariff . Producers then order quantitiesD T(7) Di i i
9
The announcement cites “developments in economic thinking” as being in part behind
the decision. It also states that cement mergers after 1977 were no longer subject to special
consideration (FTC 1985, 21507).
10
The decline in integration seen from 1992 to 1997 was due to a demand-driven bout
of unprecedented entry into the cement industry. The year 1997 saw a net gain of 61
plants over 1992’s 218 plants, which was roughly what had been the industry’s long-run
average. These new plants were primarily small, unintegrated plants specializing in grinding
clinker (an intermediate material in the cement-making process) that was kilned
elsewhere, sometimes overseas.
11
We thank Michael Whinston for an insightful discussion that led us to this model.
Incidentally, in a section on applications, Hart and Tirole explicitly cite cement and readymixed
mergers as involving the anticompetitive effects they highlight. The model is also
described in detail in Snyder (1995), Bernheim and Whinston (2004), and Rey and Tirole
(2007) and is used in the experimental study of Martin, Normann, and Snyder (2001).
O’Brien and Shaffer (1992) show that the Hart-Tirole model’s assumption of Cournot
competition downstream is not necessary to obtain foreclosure (we thank a referee for
pointing us to this paper). They derive a result similar to the one we sketch here with a
downstream industry comprising differentiated-product Bertrand competitors. Given the
spatial nature of ready-mixed concrete competition, this may also be a reasonable modeling
framework for our purposes.
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
cementing relationships 257
and pay . The then produce , observe each other’s outputs,q T(q ) D qi i i i i
and set prices. Downstream competition is modeled as Bertrand with
capacity constraints, which yields the Cournot outcome under conditions
described in Kreps and Scheinkman (1983).
Let ( , ) be the monopoly price and quantity if U also had mo-m m
Q p
nopoly power in the downstream market. Monopolist U would like to
offer the quantity and tariff schedule to them m m
(q , T) p (Q /2, p Q /2)i i
downstream producers so as to achieve monopoly profits. However, Hart
and Tirole point out that offering this contract may not be credible if
contracts are secret or can be secretly renegotiated ex post. If one downstream
firm did agree to the above contract terms, the monopolist would
then have an incentive to sell more than to the other downstreamm
Q /2
firm.12
But this action would lower the profits of the first downstream
firm, making it unwilling to sign such a contract in the first place. Hart
and Tirole show that the equilibrium with secretly renegotiable contracts
has U selling Cournot quantities to the downstream firms and earning
less than the monopoly profit.
However, U can commit to limiting total supply to the monopoly
quantity by vertically integrating. An integrated U could sell throughm
Q
its subsidiary and sell nothing to the other downstream firm. This extreme
outcome is unlikely in reality because the outside downstream
firm could try to buy input from another upstream supplier. But since
U has a cost advantage upstream—and this is where the asymmetric cost
assumption is important—the integrated firm can (and will) also sell
the input to the outside downstream firm. However, it will be at a price
that just undercuts the higher-cost upstream competitor. This leads to
an asymmetric Cournot outcome downstream, with the unintegrated
downstream firm facing higher costs than its integrated competitor. The
greater the upstream cost asymmetry, the closer equilibrium gets to the
monopoly outcome.
The end results of integration, therefore, are higher average prices
and lower quantities in the downstream output market. Moreover, if the
downstream technology has a fixed operating cost (or if no other upstream
supplier is available), exit of unintegrated downstream firms is
more likely. While the above version of the model is silent about what
happens to (potential) competitors in the upstream market, Hart and
Tirole also provide a variant—the “scarce needs” version, in which two
upstream competitors have the same marginal costs but different fixed
entry costs—in which vertical integration leads to the nonentry of one
of the upstream firms.
12
The monopolist takes as given for the contracted firm and reoptimizes quantitym
Q /2
for the other firm using the “right-shifted” version of the residual demand curve. This
yields an optimum quantity that is greater than .m
Q /2
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
258 journal of political economy
Thus if foreclosure effects are dominant, we should expect that vertical
integration will be associated with higher prices, lower quantities
of the downstream good, higher exit rates among unintegrated producers,
and lower entry rates downstream and upstream.
Note that if vertical integration increases productive efficiency, we
would instead expect lower prices and higher quantities of concrete.
However, one may still expect higher exit rates among unintegrated
producers if they do not benefit from the efficiency enhancements experienced
by integrated rivals. Entry rates may also be lower if the potential
entrants are not efficient enough to survive in the market, though
the opposite may be true if acquisition of existing downstream plants
provides an avenue of entry and expansion by low-cost producers in
other markets (including cement producers that want to enter into the
concrete business).
It is also reasonable to expect that foreclosure and efficiency effects
could simultaneously operate in many market settings. As detailed in
the next two sections, our empirical analysis will address this by including
explicit controls for plant-level productivity to look at the impact of
integration on prices, quantities, entry, and exit, net of efficiency improvements.
We suspend until Section V the investigation into the logistics
coordination mechanism that ties vertical integration to produc-
tivity.
III. Data
A. Plant-Level Ownership, Productivity, and Prices
The core of our analysis uses plant-level microdata from the 1963, 1967,
1972, 1977, 1982, 1987, 1992, and 1997 Census of Manufactures. The
census is comprehensive; we observe every U.S. cement and ready-mixed
concrete plant operating in the respective census years. A typical census
has 220 cement and 5,200 ready-mixed plants, though these relatively
stable levels hide substantial gross plant entry and exit in both industries.
The census microdata contain a wealth of information on plants’ production
activities that we exploit below. Crucially here, they also contain
firm identification numbers for each plant. (Plants, or “establishments”
in Census Bureau terminology, are unique physical locations in which
products are made. A firm can own one or more plants.) Thus we are
able to observe when a single firm owns plants in both industries, in
which case we consider the firm and its component plants vertically
integrated.
The comprehensiveness of the Census of Manufactures is extremely
useful. It allows us to observe each plant’s integration status in each
census year, and because the data contain permanent plant identifiers
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
cementing relationships 259
that are invariant to ownership changes, we can track changes in this
status over time. Entry and exit of plants between census years are completely
observable as well, allowing us to look at vertical integration’s
long-run impact on markets.
Besides ownership information, the census contains data on plant
revenues, several labor input measures, book values of equipment and
structures capital stocks, expenditures on inputs, and state and county
codes. We use these data directly and also to calculate other plant-level
technology measures such as labor and total factor productivity (TFP),
measures of overhead labor, and capital-labor ratios. Details of their
construction are in the Data Appendix. Further, for some ready-mixed
concrete plants, we have access to auxiliary files containing the total
value and the physical quantity of product shipments at a highly detailed
(seven-digit SIC) level. We can therefore measure plant output in either
dollars or physical units (cubic yards, in this case), as well as calculate
the plant’s average unit price (measured on a free-on-board basis and
deflated to 1987 dollars using the industry-level price index from the
NBER Productivity Database).13
B. Local Markets in the Cement and Ready-Mixed Concrete Industries
A useful attribute of the cement and ready-mixed industries as a forum
for testing foreclosure theory is that, because of the high transport costs
inherent to the industries, they contain many local markets. This naturally
raises the empirical issue of defining these markets. We use different
market definitions for cement and concrete because while transport
costs are high in both industries, they are higher for concrete, and
shipment patterns reflect this.
For ready-mixed concrete, we use the component economic areas
(CEAs) created by the Bureau of Economic Analysis (BEA). CEAs are
usually collections of counties, but not always, centered on metropolitan
statistical areas (MSAs), and they are mutually exclusive and exhaustive
of the land area of the United States. The roughly 3,200 U.S. counties
are grouped into 348 such markets. The procedures used to create CEAs
ensure that counties in a given CEA are economically intertwined; see
Johnson (1995) for detailed information on their construction. We be-
13
Conveniently, most plants in the ready-mixed concrete industry are highly specialized;
virtually all their revenue—roughly 95 percent on average—comes from sales of readymixed,
which is itself a seven-digit product. Other concrete products such as block, prefabricated
structural members, and pipe are typically made by producers in concrete
industries other than SIC 3273. Likewise, the amount of ready-mixed produced at these
other industries’ plants is minuscule. Unlike the ready-mixed industry, cement plants
produce a number of seven-digit products (different cement types based on their chemical
composition). This makes cement plants’ average unit output prices more difficult to
construct and less comparable across plants.
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
260 journal of political economy
lieve that CEAs are well-founded, albeit imperfect, market area measures
for an industry in which maximum ideal delivery distances are 30–45minute
drives from the plant. See Syverson (2004) for an extended
discussion of the trade-offs involved in defining ready-mixed markets.
Since most, though not all, of the empirical exercises below deal with
downstream markets, CEAs are the primary geographic demarcation
used in the paper.
For cement, with its lower transport costs, we use larger market areas.
Our primary market definition is that used by the U.S. Geological Survey
in its Minerals Yearbook annual surveys of the industry (see, e.g., USGS
2004). These 26 markets are essentially states or collections of states,
with some states with particularly large numbers of cement plants (California,
Pennsylvania, and Texas) split into separate markets. This reflects
an apparent USGS objective of roughly equating the number of
plants across markets. Because of our concern that this objective may
sometimes get in the way of defining the true geographic markets present
in the industry, in some specifications we use economic areas (EAs),
which are aggregations of CEAs.14
There are considerably more EAs
(172) than USGS cement markets, and they are not beholden to the
latter’s state boundary or producer equalization conditions.
Table 3 shows some summary measures of market structure for our
geographic markets in both industries. The average number of readymixed
plants in a CEA market-year observation in our sample is 14.6,
and the median is 10. The largest market has 125 ready-mixed plants.
The respective values for firm counts are 11.8, 8, and 110. As suggested
by the above discussion about the fraction of integrated ready-mixed
plants, most markets are not integrated: the average number of integrated
firms in a market is 0.3, and the median is zero (though certainly
in the sample’s later years these values are higher). If we condition on
at least one integrated firm operating in the concrete market, however,
the average market share of integrated firms is substantial: 31.5 percent.
The median Herfindahl-Hirschman index (HHI) for concrete markets
in our sample is 2,464.15
For the USGS cement markets, the average number of plants is 8.3,
14
While there are on average two CEAs per EA, larger and denser EAs typically have
more CEAs than those in less populated areas. For example, the Kansas City EA comprises
the Kansas City (KS-MO), Lawrence (KS), and St. Joseph (MO) CEAs, whereas the Bangor
(ME) CEA is the only one in the Bangor EA.
15
We estimated some descriptive multivariate regressions to gauge the relationship between
market structure and the extent of integration. The market share of integrated
firms is higher in more concentrated markets (as measured by HHI) as well as in those
with high demand (i.e., larger markets; see below how we construct this demand measure
from local construction activity data). However, there is no clear relationship between
integrated firms’ share of the local concrete market and concentration in the cement market
in which the CEA is contained.
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
cementing relationships 261
TABLE 3
Summary Statistics of Market Structure in the Cement and Ready-Mixed
Industries
Average
First
Quartile
Second
Quartile
Third
Quartile Maximum
A. Ready-Mixed Concrete (2,767 CEA-Year Markets)
Number of plants 14.6 5 10 17 125
Number of firms 11.8 5 8 14 110
Number of vertically
integrated firms .3 0 0 0 6
Number of vertically
integrated firms (if
10) 1.4 1 1 2 6
Market share of vertically
integrated
firms (if 10) .314 .128 .270 .430 1.000
HHI 3,001 1,585 2,464 3,832 10,000
B. Cement (208 USGS-Year Markets)
Number of plants 8.3 5 7 11 24
Number of firms 7.1 5 6.5 9 23
Number of vertically
integrated firms 2.5 1 2 4 7
HHI 2,795 1,990 2,362 3,348 10,000
Note.—The table shows moments of the distribution of market structure measures for the cement and ready-mixed
concrete industries. See the text for details and market definition.
the median 7, and the maximum 24. A larger fraction of firms are
vertically integrated: an average of 2.5 per market (of 7.1 firms total).
The median HHI is 2,362. The counts are of course lower (and the
concentration numbers higher) for EAs, since these are notably smaller
than USGS markets.
In most of the empirical tests below, we control for activity in the
local construction sector (SICs 15–17), which buys the vast majority of
the cement and concrete industries’ outputs. These demand measures
are built from County Business Patterns employment data for the sector
that we aggregate to the market level. Total market demand (size) is
measured simply as logged total construction employment, and market
density is the (log of the) number of construction-sector workers per
square mile in the market.16
16
As discussed in Syverson (2004, forthcoming), construction activity is likely exogenous
to the specifics of local concrete competition because while the construction sector accounts
for virtually all ready-mixed sales, concrete’s cost share among the sector’s intermediate
inputs is small. Hence any changes in concrete prices should not by themselves
create notable changes in the level of construction activity. We expect that similar effects
operate on a broader geographic scale in cement.
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
262 journal of political economy
IV. The Extent of Integration and Market Power
Our empirical approach is straightforward. We compare patterns in the
data to the predictions of foreclosure theory as well as the alternative
explanation that increases in integration reflect the expansion of larger
and more efficient producers. Our ability to distinguish foreclosure from
the alternative rests on two key sets of implications. First, foreclosure
and productivity/efficiency theories have opposing predictions regarding
the response of downstream prices and quantities to integration.
Second, while the two stories’ predictions for firm turnover are similar,
foreclosure implies an effect of vertical integration per se. Productivity/
efficiency explanations, on the other hand, imply that any observed
effect of integration on turnover acts only inasmuch as productivity and
integration are connected; hence integration’s effects should disappear
once productivity is controlled for.17
A. Does Vertical Integration Raise Prices?
Foreclosure implies a positive relationship between downstream prices
and the extent of vertical integration in a market, whereas an efficiencybased
explanation for integration predicts a negative relationship. As
seen in table 1 above, the basic patterns in the data are not consistent
with foreclosure. We now investigate whether these market-level results
hold up to deeper scrutiny. We first take advantage of our detailed plantlevel
data to see if there is a relation between vertical integration and
prices among producers in the same market. Next, we examine the
efficiency and vertical integration link directly in plant-level productivity
data. Finally, we revisit the market-level evidence displayed in table 1,
this time using a richer set of controls and paying more attention to
potential endogeneity problems.
17
A third distinguishing implication of foreclosure models is that unintegrated downstream
producers face higher input costs. We can in principle test this in our data, because
a subset of plants in the Census of Manufactures answer questions regarding their total
expenditures on and purchased quantities of detailed intermediate inputs, allowing us to
compute unit cement prices paid by concrete producers. When we do so, we find that,
rather than unintegrated concrete producers reporting higher cement input prices, there
are no significant differences across integrated and unintegrated plants. While on its face
this is not what is suggested by foreclosure, it is far from clear what the reported cement
prices from integrated concrete makers reflect. The Census Bureau instructs establishments
to report the value of internal materials transfers “at their full economic value (the
value assigned by the shipping plant, plus the cost of freight and other handling charges).”
It is possible that these results simply reflect integrated plants “marking to the market”
their internal transfer prices.
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
cementing relationships 263
TABLE 4
Vertical Integration and Ready-Mixed Concrete Prices: Plant-Level Results
Within-Market Difference Change for Continuers
(1) (2) (3) (4) (5) (6) (7) (8)
Observations 12,553 12,553 8,555 8,555 4,025 4,025 2,439 2,439
2
R .424 .425 .501 .501 .456 .460 .575 .576
Vertical inte-
gration
indicator
Ϫ.022*
(.006)
Ϫ.017*
(.006)
Ϫ.006
(.007)
Ϫ.006
(.007)
.011
(.035)
.011
(.035)
.003
(.034)
.005
(.034)
Multiunit
indicator
Ϫ.012*
(.004)
Ϫ.001
(.005)
Ϫ.037
(.020)
Ϫ.033
(.028)
TFP Ϫ.214*
(.015)
Ϫ.215*
(.015)
Ϫ.237*
(.028)
Ϫ.237*
(.028)
Integrated vs. Unintegrated
Entrants
Integrated Entrants vs. Unintegrated
Incumbents
(9) (10) (11) (12) (13) (14) (15) (16)
Observations 2,771 2,771 2,025 2,025 7,490 7,490 6,104 6,104
2
R .563 .566 .655 .655 .430 .430 .523 .523
Vertical inte-
gration
indicator
Ϫ.037*
(.018)
Ϫ.025
(.019)
Ϫ.020
(.020)
Ϫ.012
(.020)
Ϫ.023
(.012)
Ϫ.023
(.012)
Ϫ.012
(.012)
Ϫ.012
(.012)
Multiunit
indicator
Ϫ.032*
(.012)
Ϫ.025
(.014)
NA NA
TFP Ϫ.218*
(.031)
Ϫ.215*
(.031)
Ϫ.221*
(.019)
Ϫ.221*
(.019)
Note.—The table shows the results from regressing plant-level concrete prices on a number of variables. Vertical
integration (multiunit) indicator is equal to one if the plant is in a vertically integrated (multiunit/multiplant) firm
and zero otherwise; TFP is the plant’s quantity-based total factor productivity. See the text for details. In the vertically
integrated entrants vs. non–vertically integrated incumbents comparison, there are no observations of new multiunit
entrants that are not also vertically integrated entrants. All regressions include market-year fixed effects.
* Significant at the 5 percent level.
1. Plant-Level Evidence
We start our look at the plant-level relationship between downstream
prices and vertical integration by regressing ready-mixed concrete
plants’ (logged) prices on an indicator for the plant’s vertical integration
status and a full set of market-year fixed effects. The vertical integration
dummy coefficient therefore captures the mean price difference across
integrated and unintegrated producers in the same market and time
period. This specification is useful in that it compares producers that
are facing the same market-level demand and supply conditions, which
the market-level regressions in table 1 do not fully control for.
The results of this exercise are shown in column 1 of table 4. On
average, integrated producers charge prices 2.2 percent lower than their
unintegrated competitors in the same market. Thus the negative relationship
between downstream prices and integration seen across markets
also holds even among producers within them.
We further investigate these plant-level patterns with several additional
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
264 journal of political economy
comparisons, again all within the same market-year. Column 5 of the
table compares price growth rates among continuing plants (i.e., those
that operated in both the previous and current censuses) that were
unintegrated in the previous census. The vertical integration indicator
in this case takes a value of one if a plant became integrated between
the previous and current censuses. The estimated coefficient implies
that these newly integrated plants have somewhat—1.1 percent—more
positive price changes than their cohorts that remain unintegrated, but
the difference is not statistically significant. Columns 9 and 13 contrast
the prices of integrated entrants (entrants are defined as plants appearing
for the first time in the current census) to those of unintegrated
entrants and unintegrated incumbents, respectively. As with the comparison
in column 1, integrated producers charge lower prices than
those in the unintegrated comparison groups. Integrated firms’ new
plants price 3.7 percent below new plants of unintegrated firms and 2.3
percent below integrated incumbents (though this last difference is only
marginally significant, with a p-value of 0.07). Integrated producers’
lower prices thus arise not because formerly unintegrated existing plants
see large price drops upon becoming integrated, but instead because
newly built integrated plants tend to charge lower prices, and this difference
persists over time.
To get at the source of integrated plants’ lower prices, we introduce
two additional controls into each of the four plant-level price comparisons
above. These are a dummy indicating if a plant is owned by a
multiunit firm and the plant’s TFP level. The multiunit dummy captures,
in a rough measure, the size of the firm that owns the plant. Recall
from the discussion in Section II that all vertically integrated plants are
in multiplant firms, but nonvertical organizational structures meet this
condition as well. So when we include the multiunit indicator in a
specification, the vertical integration coefficient reflects differences between
integrated plants and other plants in multiunit firms without
cement operations. The TFP measure reflects a producer’s efficiency,
in other words, the inverse of its cost level. We use a physical TFP
measure—that is, the output measure in the TFP index is the physical
output of the plant—rather than the literature standard revenue-based
output measure. As Foster, Haltiwanger, and Syverson (2005) discuss,
the latter attributes price differences across industry plants to output
differences. This is something one would like to avoid, and we can do
so since we can measure the actual cubic yards of concrete that plants
ship.
The remaining columns of table 4 show each of the four plant-level
price comparisons with the multiunit status dummy and physical TFP
measure included in the specifications, first separately and then toThis
content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
cementing relationships 265
TABLE 5
Plant-Level TFP Comparisons
Within-
Market
Difference
(1)
Change for
Continuers
(2)
Integrated vs.
Unintegrated
Entrants
(3)
Integrated
Entrants vs.
Unintegrated
Incumbents
(4)
Observations 8,555 2,439 2,025 6,104
2
R .308 .419 .573 .352
Vertical integration
indicator
.043*
(.014)
.102
(.055)
.054
(.045)
.046*
(.028)
Note.—The table shows the results from regressing ready-mixed concrete plants’ TFP levels on an indicator for the
plant being in a vertically integrated firm. All regressions include market-year fixed effects.
* Significant at the 5 percent level.
gether.18
Controlling for a plant’s multiunit status explains some of the
price difference between plants within a market, to the point of making
the vertical integration indicator insignificant in the comparison of entering
plants’ prices; but prices of vertically integrated plants remain
significantly lower than those of other multiunit plants in the broadest
comparison. More starkly, TFP always enters negatively into the regressions
and reduces the magnitude and eliminates the statistical significance
of the vertical integration indicator. These results suggest that
vertically integrated producers charge lower prices because to some
degree all plants in multiunit firms do, but especially because they have
higher productivity levels.19
2. Are Integrated Plants More Productive?
To look at the productivity-integration relationship more closely, we
repeat the four sets of comparisons above, except we replace prices with
plants’ TFP levels.20
(Again all specifications include CEA-year fixed
effects, so results reflect variation among plants in the same marketyear.)
The results are shown in table 5. Column 1 indicates that vertically
18
The number of observations in the specifications including plant TFP is smaller because
capital stocks, which we need to construct a TFP measure, were not available in the
1967 Census of Manufactures, whereas price data were.
19
The fact that we construct plant-level prices as the ratio of reported revenues to
physical quantities means that if there is measurement error in quantities, this will create
a negative correlation between measured prices and quantities (and, by implication, TFP
as well). This could potentially lead to these specifications overstating the actual correlation
between prices and productivity levels. However, work in Foster et al. (2005) and Syverson
(forthcoming) indicates that the negative relationship is not a spurious result driven by
measurement error. Further, as we show below, the negative price-productivity correlation
also holds for market-level averages, where the averaging process should lessen any influence
of measurement error.
20
We will explore the connection between vertical integration and firm size in detail
below.
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
266 journal of political economy
integrated plants have TFP levels that are 4.3 percent higher on average
than their unintegrated competitors. In other words, they produce 104.3
cubic yards of ready-mixed concrete from the same inputs that unintegrated
producers in the same market use to produce 100 cubic yards.
Comparing TFP changes among formerly unintegrated plants in column
2 shows that plants that become integrated see roughly 10 percent faster
productivity growth, though this difference is only marginally statistically
significant. TFP gaps similar to those in the entire sample are observed
when we restrict the comparison to between integrated entrants and
either unintegrated entrants or incumbents. The significance of these
differences is weaker, however, likely because of in part the reduced
sample sizes.
These patterns are consistent with the alternative explanation to foreclosure
forwarded above: namely, that increases in integration reflect
expansion of more efficient producers. Their lower costs drive down
average market prices directly, because they have lower costs and can
charge less, but also indirectly, by reducing their competitors’ optimal
prices in equilibrium.
3. Market-Level Evidence Revisited
With these plant-level findings in hand, we now revisit market-level analyses
of the types shown in table 1. We check the robustness of these
earlier results by adding more controls. We include the HHI concentration
measure of the CEA-year observation to control for the horizontal
structure of the local concrete market. As discussed above, the
extent of vertical integration in the downstream market is related to
downstream (though not upstream) concentration. Controlling for this
correlation lets us see the vertical integration–price relationship that is
orthogonal to any changes in horizontal concentration. We also control
for demand density (described above) in the local market; Syverson
(forthcoming) shows how it affects average prices through competitiondriven
selection and markups. We cluster standard errors by CEA market
to account for any temporal correlation in unobservables.
The results are shown in table 6. Columns 1, 2, 6, and 7 repeat the
market-level price specifications from table 1 (for both extent of integration
measures) while including the additional controls. Just as in
table 1, more integrated concrete markets have lower average prices.21
21
For space and clarity reasons, we do not report the HHI or demand density coefficients.
HHI typically enters positively, indicating that, all else equal, higher concentration
coincides with higher prices. However, it is usually not statistically significant,
and its economic magnitude is modest: a 2,250-point increase in the HHI—the interquartile
range—is associated with a roughly 1 percent increase in prices. Density too is
typically insignificant, except in the specification excluding market fixed effects, in
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
TABLE6
VerticalIntegrationandReady-MixedConcretePrices:Market-LevelResults
BenchmarkSpecification
(1)(2)(3)(4)(5)(6)(7)(8)(9)(10)
Observations1,8701,8701,8701,5501,5501,8701,8701,8701,5501,550
2
R.087.433.434.573.573.087.432.432.573.573
Marketshareofverticallyinte-
gratedfirms
Ϫ.125*
(.028)
Ϫ.090*
(.041)
Ϫ.086*
(.041)
Ϫ.043
(.039)
Ϫ.043
(.039)
MarketshareofmultiunitfirmsϪ.015
(.022)
.001
(.024)
Numberofverticallyintegrated
firms
Ϫ.028*
(.007)
Ϫ.015
(.011)
Ϫ.013
(.011)
Ϫ.009
(.009)
Ϫ.007
(.009)
NumberofmultiunitfirmsϪ.003
(.004)
Ϫ.004
(.004)
Quantity-weightedaverageTFPϪ.293*
(.054)
Ϫ.293*
(.054)
Ϫ.294*
(.054)
Ϫ.294*
(.054)
Marketfixedeffects?NoYesYesYesYesNoYesYesYesYes
Note.—Thetableshowsthecoefficientsobtainedbyregressingquantity-weightedaverageconcretepricesinamarketoneitherthemarketshareorthenumberofverticallyintegratedfirms
operatinginthemarket.Themarketshareornumberofmultiunitfirmsandthequantity-weightedaverageTFPinthemarketarealsoincludedinsomespecifications.Allregressionscontrolfor
theHHIanddensityofdemandinthemarketaswellasyeareffects(coefficientsnotreported).Standarderrorsareclusteredbymarket.
*Significantatthe5percentlevel.
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
268 journal of political economy
The magnitudes of the coefficients track the earlier values, though the
specifications that include market fixed effects have smaller estimates
that in one case become insignificant.22
The implied average price differences
are substantial. For example, column 1 implies that going from
a market with no integrated producers to one in which integrated firms
hold a market share of 0.315 (the average share conditional on at least
one vertically integrated producer being in a market) corresponds to a
4 percent decline in the average price of ready-mixed. Likewise, the
coefficient in column 6 indicates that an additional integrated firm
operating in a market corresponds to a roughly 3 percent drop in average
prices. These estimated magnitudes are roughly 25–35 percent of
the between-market standard deviation in average prices of 10.2 percent.
As suggested by the plant-level analyses above, we investigate how
controlling for multiunit status and TFP affects the price-integration
relationship. For market-level controls analogous to those above, we use
the market share (or number) of multiplant firms and the quantityweighted
average TFP in the market. Again we add each control separately
to the regressions and then control for them simultaneously. The
results are in columns 3–5 and 8–10 (again the sample size falls because
we cannot measure TFP in the 1967 Census of Manufactures).
The results are similar to those in the plant-level analyses above. The
market share or number of multiunit firms enters negatively and slightly
reduces the estimated impact of vertical integration, but the greatest
impact is found when average TFP in the market is included in the
specification. The average TFP coefficients are negative; intuitively,
growth in the average productivity of a market’s producers is associated
with declining average prices. Our TFP control therefore does seem to
capture efficiency effects on market outcomes. Moreover, the coefficients
on the extent of integration measures remain negative but shrink
in size and become insignificant. Hence if changes in average TFP sufficiently
capture the efficiency gains that accompany vertical integration,
the results suggest that the remaining price effects of integration are
essentially zero. These results suggest two inferences. First, any impact
of the extent of vertical integration in a market appears to be driven
by the increase in average productivity levels among market producers
rather than something tied to integration per se. Second, it does not
which case it enters negatively. It is positive and insignificant otherwise. The sign switches
because within-market changes in density actually reflect only changes in the level of
demand, since market areas are fixed. Ceteris paribus, higher demand will raise prices
in a market.
22
The smaller coefficients with market fixed effects could result either because there
are market unobservables correlated positively with vertical integration and negatively with
prices (more on this below) or because attenuation bias from measurement error (say if
our market definitions imperfectly capture the true geographic markets) is exacerbated
when we identify vertical integration effects only from within-market changes.
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
cementing relationships 269
appear that efficiency gains from vertical mergers or other unobservables
are simply hiding foreclosure effects in the raw data. Once we
control for the change in average efficiency, we still do not see a positive
impact of integration on prices.
These results are not on their face consistent with the positive relationship
between prices and integration implied by foreclosure. Of
course, the decision to vertically integrate is a choice made by firms,
and the extent of integration is not random across our sample. Perhaps
integration does facilitate foreclosure, but there are market-specific
unobservables correlated with both higher integration intensity and
lower prices. We make a few observations on this point. First, any unobservables
would have to have particularly strong correlations with higher
integration intensity and lower prices to swamp the positive price effects
of foreclosure, since the observed net integration-price correlations are
substantially negative. Second, it is unclear what the candidates for such
unobservables might be. Suppose that the price drops are driven by
unobservable inward shifts in market demand (and recall that we are
already controlling for demand density, which is positively correlated
with demand across markets and whose growth within markets equals
demand growth since areas are fixed). If so, these unobservable demand
contractions would have to induce large, expanding firms—which, as
we will discuss later, aptly describes integrated firms in our sample—to
prefer integrating into such markets. But one typically expects entry
and expansion to follow positive demand shocks, not negative ones.
Nonetheless, possible endogeneity concerns should be minimized to
the greatest extent possible given the data. Including market fixed effects
goes some distance in this direction, since it removes any marketlevel
influences on prices and the extent of integration that are fixed
over time. However, we take additional approaches toward this end.
These are discussed below.23
Our first approach uses variation in antitrust authorities’ enforcement
regimes (some of which we discussed briefly above in Sec. II) and spatial
demand patterns within markets to instrument for market structure. For
the sake of brevity, we detail the construction of the instrument set in
the Data Appendix, but their logic can be summarized succinctly. The
antitrust enforcement instruments use the stated policy rules of the U.S.
antitrust agencies (both the FTC and the Department of Justice) to
identify instances in which the mechanical nature of the rules creates
plausibly exogenous variation in the extent of antitrust scrutiny across
our sample markets, either in cross section or over time. These policies
23
Vertical integration is not the only market structure attribute that is endogenous: the
market share of multiunit firms and concentration reflected in the HHI are as well. Our
methods also attempt to address this endogeneity.
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
270 journal of political economy
include rules applied both economywide and specifically to the cement
and ready-mixed industries. The spatial demand instruments work off
the facts that construction activity should be exogenous to outcomes of
the ready-mixed market because of concrete’s small share in construction
costs (as discussed above), and that demand variation across areas
within a given market plausibly affects the returns to a firm owning
multiple concrete plants. Both the antitrust policy and spatial demand
instruments should have explanatory power over firms’ preferred organizational
structures within and across markets—that is, whether they
are multiunit or vertically integrated, and how large a market share they
can have within a given market—and be orthogonal to other local demand
or supply shocks. The first-stage results, not reported here, indicate
that interactions of firm and market structures, and the extent
of vertical integration in a market in particular, are related to the instruments
in the expected ways. For example, the extent of integration
in concrete markets with a larger share of firms specifically marked by
the FTC for merger scrutiny is lower during the years in which this
policy was in force.
The results of the instrumented specification are shown in colums 1–
4 of table 7. (All specifications control for demand, downstream HHI,
and market fixed effects.) The implied impact of vertical integration
on prices is negative, as with the benchmark results above. Moreover,
the magnitude of the implied relationship between integration and
prices is larger than in table 6 (in the fully saturated specification,
moving a market from integrated firms having zero market share to
those having a 31.5 percent share implies a 25 percent drop in average
price). We note, however, that the standard errors are much larger here
than in the benchmark specification. This lack of precision, which is
even more pronounced in our market-level quantity regressions discussed
in subsection B below, may arise because our instruments have
difficulty predicting changes in vertical integration that are orthogonal
to changes in horizontal structure (as captured by the share of multiunit
firms and market HHI).24
Another approach to address endogeneity concerns estimates price
effects using the subsample of CEA-years involved in multimarket mergers.
The idea, as with Hastings and Gilbert (2005), is to identify inte-
24
As pointed out by Shea (1997), in the presence of multiple endogenous variables in
a specification, instruments need not just be relevant to each endogenous variable; they
must have linearly independent relevance to each. We constructed the R2
relevance statistics
Shea suggests to measure this and found, indeed, that after we accounted for the market
share of multiunit firms and market HHI (two other endogenous variables), the instruments
did not have a high degree of explanatory power over the market share of vertically
integrated producers. This suggests caution in interpreting the results from this specification,
particularly with regard to the validity of the standard errors, even though the
instruments pass the commonly used F-test in the first stage.
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
TABLE7
VerticalIntegrationandReady-MixedConcretePrices:Market-LevelResults
RobustnessChecks
AntitrustandSpatialDe-
mandInstrumentalVariablesMultiple-MarketMergersAddingState-YearEffects
(1)(2)(3)(4)(5)(6)(7)(8)(9)(10)(11)(12)
Observations1,8701,8701,5501,5501421421331331,8701,8701,5501,550
2
RNANANANA.453.472.584.602.586.586.681.681
Marketshareofverticallyintegrated
firms
Ϫ.468*
(.217)
Ϫ.722*
(.308)
Ϫ.707*
(.220)
Ϫ.914*
(.331)
Ϫ.164*
(.049)
Ϫ.132*
(.054)
Ϫ.098
(.051)
Ϫ.068
(.051)
Ϫ.071
(.044)
Ϫ.065
(.045)
Ϫ.022
(.044)
Ϫ.019
(.044)
Marketshareofmultiunitfirms.285
(.220)
.343
(.348)
Ϫ.141
(.105)
Ϫ.136
(.089)
Ϫ.022
(.025)
Ϫ.008
(.027)
Quantity-weightedaverageTFPϪ.287*
(.019)
Ϫ.296*
(.043)
Ϫ.363*
(.067)
Ϫ.364*
(.066)
Ϫ.269*
(.056)
Ϫ.269*
(.056)
Note.—Thetableshowsthecoefficientsobtainedbyregressingquantity-weightedaverageconcretepricesinamarketonthemarketshareofverticallyintegratedfirmsoperatinginthemarket.
Themarketshareofmultiunitfirmsandthequantity-weightedaverageTFPinthemarketarealsoincludedinsomespecifications.AllregressionscontrolfortheHHIanddensityofdemandin
themarketaswellasyeareffects(coefficientsnotreported).Thevariousidentificationapproaches,listedintheheaderrow,aredetailedinthetext.Standarderrorsareclusteredbymarket.
*Significantatthe5percentlevel.
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
272 journal of political economy
gration’s impacts using arguably random variation in the extent of integration
across the several markets that experience a single-firm
merger. For example, consider the simple case of a cement firm integrating
by purchasing a concrete firm that owns two plants, each operating
in a different market. One plant has a market share of 20 percent
in its local market, the other a 30 percent share. If the cement firm’s
decision to integrate with this particular downstream firm was made on
the basis of considerations of the target firm as a whole rather than in
its specific markets, differences in the extent of integration due to the
merger—integrated firms’ market share will grow 10 percentage points
more in the second market than in the first—will be exogenous to
market-specific outcomes. Hastings and Gilbert use this type of variation
from a particular merger in the wholesale gasoline industry that affected
14 markets. While most of the multimarket mergers in our sample involve
fewer markets, we observe not one but many such mergers during
our observation period.
We define a multiple-market merger as having occurred if a single
firm newly integrates plants in two or more markets during the same
intercensus period. This happened 29 times in our sample, each merger
involving an average of just under five markets, giving us a sample of
142 market-years with which to estimate vertical integration’s effects.
We run the price regressions as before on this subsample but now include
merger fixed effects in the regressions. Thus all the variation in
vertical integration intensity used to identify integration’s impact comes
from across-market variation within particular merger episodes. As in
the example above, the specification uses the fact that the market share
vertical integration measure was 10 percentage points higher in the
second market than in the first and compares this to the difference in
average prices between the two markets.
The results using the sample of multimarket merger episodes are
shown in columns 5–8 of table 7. Again, all specifications control for
the market demand and HHI, though these coefficients are not reported.
A higher market share of vertically integrated firms corresponds
to lower prices. (Similar results, not reported here, are obtained when
the number of vertically integrated firms is used to measure the extent
of integration in the market.) The magnitude of the vertical integration
coefficient is notably similar to that in the benchmark specification.25
When we add controls for the share of multiunit firms and the average
25
Recall that the exogeneity argument here is that firms make merger decisions on a
whole-firm basis, not on the expected market structure change in any particular market
they are merging into. Obviously we cannot test this proposition directly. However, arguments
that particular market outcomes are the primary concern would have to explain,
as is implied by the results, why the merging firms choose to most intensively enter markets
with idiosyncratically low prices.
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
cementing relationships 273
TFP in the market, we also find results similar to those above. The
multiunit share is negatively (albeit weakly) related to average prices
and explains part of the correlation between vertical integration and
prices. Average TFP is also associated with lower prices (we lose nine
CEA-year markets and four merger episodes in the smaller sample), and
including it makes the extent of integration measure insignificant. When
we control for both multiunit share and average TFP, integration is
negatively but insignificantly connected to average prices.
Our third approach to address possible endogeneity is to add stateyear
fixed effects to our benchmark specification. These control for any
state-specific unobserved demand or supply shocks that might be correlated
with the extent of integration and average prices. One example
of such shocks would be local government construction spending that
for one reason or another is not reflected in our construction employment
(demand) measure. Note that we still include CEA effects as well;
in effect, the specification identifies vertical integration’s impact from
intertemporal price and integration variations across different CEAs in
the same state.26
The results are in columns 9–12 of table 7. Not surprisingly, the
reduced variation we have to estimate the link between integration and
prices leads to less statistical significance; in no case does the vertical
integration coefficient have a p-value below 0.11. However, the qualitative
features of the results echo those seen in the benchmark. Vertical
integration is negatively related to average concrete prices, as are the
share of multiunit plants and the average TFP level in the market,
though only the latter is significantly so. This is again consistent with
efficiency effects driving the negative relationship between prices and
the extent of integration.
On the whole, the many results presented in this section do not point
to foreclosure-driven price increases. Instead, vertical integration is associated
with lower prices at both the plant and market levels. We do
find evidence that this negative relationship is not driven by vertical
integration per se, but rather because vertical integration is tied to
larger, more productive firms. This is consistent with the alternative
efficiency-based explanation of the observed patterns in the data. In
26
A measurement issue arises here because states do not completely nest every CEA
(recall that CEA boundaries need not coincide with state boundaries). To implement this
empirical specification, we have to assign a “primary” state to each CEA that is not wholly
contained within one state. We do so by counting the number of counties of each state
in every multistate CEA and assigning the primary state designation to the state with the
most counties. In the few cases in which the maximum counts were equal, the state with
the primary influence was apparent because it had a large metropolitan area in the CEA.
The state-year dummies were constructed by interacting year effects with either this “modalstate”
designation for multistate CEAs or the unique state containing the other CEAs.
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
274 journal of political economy
TABLE 8
Vertical Integration and Ready-Mixed Concrete Quantities: Plant-Level
Results
Within-Market Difference Change for Continuers
(1) (2) (3) (4) (5) (6) (7) (8)
Observations 12,553 12,553 8,555 8,555 4,025 4,025 2,439 2,439
2
R .359 .362 .376 .377 .457 .457 .539 .540
Vertical inte-
gration
indicator
.486*
(.046)
.426*
(.047)
.399*
(.046)
.370*
(.047)
Ϫ.191*
(.096)
Ϫ.192*
(.096)
Ϫ.208*
(.098)
Ϫ.204*
(.097)
Multiunit
indicator
.160*
(.025)
.088*
(.028)
Ϫ.002
(.064)
Ϫ.087
(.082)
TFP .548*
(.052)
.538*
(.052)
.487*
(.076)
.488*
(.077)
Integrated vs. Unintegrated
Entrants
Integrated Entrants vs. Unintegrated
Incumbents
(9) (10) (11) (12) (13) (14) (15) (16)
Observations 2,771 2,771 2,025 2,025 7,490 7,490 6,104 6,104
2
R .543 .544 .601 .602 .367 .367 .392 .392
Vertical inte-
gration
indicator
.324*
(.119)
.276*
(.122)
.240
(.134)
.212
(.136)
.042
(.086)
.042
(.086)
Ϫ.070
(.012)
Ϫ.070
(.012)
Multiunit
indicator
.127
(.077)
.082
(.089)
NA NA
TFP .592*
(.132)
.583*
(.132)
.533*
(.068)
.533*
(.068)
Note.—The table shows the results from regressing plant-level physical output on a number of variables. Vertical
integration (multiunit) indicator is equal to one if the plant is in a vertically integrated (multiunit/multiplant) firm
and zero otherwise; TFP is the plant’s quantity-based total factor productivity. See the text for details. In the vertically
integrated entrants vs. non–vertically integrated incumbents comparison, there are no observations of new multiunit
entrants that are not also vertically integrated entrants. All regressions include market-year fixed effects.
* Significant at the 5 percent level.
Section V we will discuss why vertical integration, productivity, and firm
size might be linked in these industries.
B. Does Vertical Integration Reduce Quantities?
We use specifications similar to those above to investigate how vertical
integration affects market quantities.
We start again by documenting the plant-level patterns. Table 8 repeats
the comparisons of table 4, except using plants’ physical outputs (logged
cubic yards of ready-mixed concrete). Column 1 shows that integrated
ready-mixed producers make considerably more output than their unintegrated
competitors in the same market. However, as seen in column
5, the output of ready-mixed plants that become integrated actually
shrinks (some 18 percent) relative to that of unintegrated continuers.
This is counterbalanced, though, by the building of new plants by the
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC56 UTC
All use subject to https://about.jstor.org/terms
cementing relationships 275
integrated firm; as can be seen in the estimates in column 9, new integrated
plants are larger than plants of unintegrated producers.27
When we add controls for multiunit status and the plant’s TFP, both
are associated with higher plant output, with multiunit status sometimes
statistically significant and TFP always so. Further, in both the overall
and entrant-only comparisons, their inclusion reduces the implied quantity
impact of vertical integration. This is qualitatively similar to their
effect in the price comparisons above. In the broadest comparison (col.
4), however, there is still a statistically distinct quantity difference between
vertically integrated plants and other multiunit plants, even when
we control for TFP. Including the controls has little impact on how
much newly integrated firms’ outputs fall relative to plants that remain
unintegrated, even though productivity gains are strongly correlated
with output growth (col. 8).
For the market-level analysis, we replace the weighted-average price
as the dependent variable with the total (logged) physical quantity of
ready-mixed concrete sold in the market.28
Just as with the market-level
price regressions, we control for year effects, the market’s HHI, and
local demand as measured by logged construction employment (levels
rather than density here). Again we do not report these coefficients for
space reasons; HHI is significantly negative and demand, not surprisingly,
significantly positive in every specification. As with the price regressions,
we estimate several specifications that include our benchmark
estimations and those that address possible endogeneity of firm and
market structure.
The results are presented in tables 9 and 10. In the benchmark specifications
in table 9, greater vertical integration measured by either the
market share or the number of vertically integrated firms is associated
with higher output levels in the local concrete market. This again reflects
27
Note that there is a less direct connection between plant- and market-level outcomes
in quantity terms than with prices. Firms’ total quantities produced in a market can change
either intensively, through the amount produced by individual plants, or extensively, with
the number of plants the firm owns. The expected output change due to a new integrated
firm entering a market depends on the size of the output reductions at its newly acquired
plants relative to the output added by its newly built plants. In our sample, the net revenue
changes when a new vertically integrated firm comes into a market are on average positive.
This is consistent with the market-level evidence below. We discuss further in Sec. V why
an integrated producer in the same market over the same period might purchase and
shrink existing plants while also building new ones.
28
Here we must be particularly careful with regard to measurement issues. Because the
product supplement of the Census of Manufactures does not cover all plants, we cannot
directly calculate total market quantity simply by aggregating plants’ reported quantities.
We therefore use plant revenue instead. Yet revenue data alone are problematic because
they incorporate price variation across plants: because our price deflator is not market
specific, unadjusted revenue overstates (understates) output in markets with higher
(lower) than average prices. Our solution is to measure total quantity in a market as
market revenue divided by the market’s quantity-weighted average price that we use above.
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
276
TABLE9
VerticalIntegrationandReady-MixedConcreteQuantities:Market-LevelResults
BenchmarkSpecifications
(1)(2)(3)(4)(5)(6)(7)(8)(9)(10)
Observations1,8701,8701,8701,5501,5501,8701,8701,8701,5501,550
2
R.781.904.907.923.926.785.905.908.924.927
Marketshareofverticallyinte-
gratedfirms
.688*
(.133)
.335*
(.143)
.238
(.137)
.068
(.150)
Ϫ.011
(.143)
Marketshareofmultiunitfirms.332*
(.080)
.324*
(.081)
Numberofverticallyintegrated
firms
.190*
(.025)
.121*
(.033)
.085*
(.031)
.060
(.032)
.024
(.029)
Numberofmultiunitfirms.057*
(.010)
.062*
(.011)
Quantity-weightedaverageTFP.380*
(.088)
.367*
(.084)
.381*
(.087)
.383*
(.084)
Marketfixedeffects?NoYesYesYesYesNoYesYesYesYes
Note.—Thetableshowsthecoefficientsobtainedbyregressingthetotal(logged)quantityofready-mixedconcretesoldinamarketoneitherthemarketshareorthenumberofverticallyintegrated
firmsoperatinginthemarket.Themarketshareornumberofmultiunitfirmsandthequantity-weightedaverageTFPinthemarketarealsoincludedinsomespecifications.Allregressionscontrol
fortheHHIanddensityofdemandinthemarketaswellasyeareffects(coefficientsnotreported).Standarderrorsareclusteredbymarket.
*Significantatthe5percentlevel.
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
277
TABLE10
VerticalIntegrationandReady-MixedConcreteQuantities:Market-LevelResults
RobustnessChecks
AntitrustandSpatial
DemandInstrumental
VariablesMultiple-MarketMergersAddingState-YearEffects
(1)(2)(3)(4)(5)(6)(7)(8)(9)(10)(11)(12)
Observations1,8701,8701,5501,5501421421331331,8701,8701,5501,550
2
RNANANANA.829.836.858.863.923.929.942.944
Marketshareofverticallyinte-
gratedfirms
.710
(.812)
Ϫ.643
(1.063)
.228
(.744)
.416
(.952)
.490
(.425)
.348
(.406)
.407
(.400)
.292
(.375)
.403*
(.179)
.273
(.172)
.168
(.210)
.056
(.203)
Marketshareofmultiunitfirms1.527*
(.759)
Ϫ.311
(.914)
.634
(.379)
.507
(.353)
.366*
(.089)
.362*
(.090)
Quantity-weightedaverageTFP.331*
(.065)
.339*
(.072)
.564*
(.232)
.572*
(.230)
.350*
(.102)
.339*
(.097)
Note.—Thetableshowsthecoefficientsobtainedbyregressingthetotal(logged)quantityofready-mixedconcretesoldinamarketoneitherthemarketshareorthenumberofvertically
integratedfirmsoperatinginthemarket.Themarketshareofmultiunitfirmsandthequantity-weightedaverageTFPinthemarketarealsoincludedinsomespecifications.Allregressionscontrol
fortheHHIanddensityofdemandinthemarketaswellasyeareffects(coefficientsnotreported).Thevariousidentificationapproaches,listedintheheaderrow,aredetailedinthetext.Standard
errorsareclusteredbymarket.
*Significantatthe5percentlevel.
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
278 journal of political economy
the patterns seen in table 1. This is true in the cross section, as seen in
columns 1 and 6, and within markets over time, as in columns 2 and
7. The estimates imply that going from a completely unintegrated market
to one in which integrated firms have a 31.5 percent share implies
a rise in total quantities sold of roughly 10 percent. Alternatively, going
from zero to one vertically integrated firm in the concrete market corresponds
to a quantity increase of about 12 percent.29
Similar patterns
are seen in response to changes in integration within markets. Columns
3 and 4 of the table show the results when market fixed effects are
included. Output increases significantly when a market becomes more
integrated.
When we control for the market share (or number) of multiunit firms
and the quantity-weighted average TFP in the market, again we see, as
with the market-level price tests above, that these measures explain the
link between market outcome and the extent of vertical integration.
The multiunit measure always enters as significantly positive and substantially
reduces the magnitude of the coefficient on the extent of
integration measure. Average TFP follows the same pattern. When included
together, these controls drive the coefficient on the vertical integration
measure essentially to zero. Unlike the average price results,
where average TFP explained most of the relationship between integration
and prices, both multiunit firms and average TFP have independent,
significant correlations with total market quantities.
In the robustness checks in table 10, we obtain a positive point estimate
of the effect of vertical integration on total market quantities in
all but one of the specifications. As in table 9, we also find that adding
the TFP and multiunit controls reduces the point estimate on vertical
integration. Statistical significance, however, is obtained less readily. The
antitrust enforcement and spatial demand instrumentation strategy
yields imprecise point estimates (see the discussion above). In the multimarket
merger specification, while the coefficients on the extent of
integration measure are all positive and with a size similar to that of
the benchmark results, they are not significant. (The multimarket
merger specifications that used the number of vertically integrated firms
29
We note that the above argument that market-level construction activity is likely exogenous
to concrete prices is still consistent with a nonzero quantity response to changes
in concrete prices. To see why, note that the exogeneity argument implies that the derivative
of construction output with respect to concrete prices is roughly zero. This derivative
can be restated as the product of two derivatives: construction output with respect to
concrete output and concrete output with respect to concrete prices. Thus even when
the latter is nonzero (i.e., concrete quantities are responsive to concrete prices), the total
derivative is roughly zero if the former derivative is small enough. This former derivative
is determined by the construction sector’s production function. The very small cost share
of concrete in total construction costs (around 2 percent in the data tables of the Benchmark
Input-Output Accounts published by the BEA) suggests that this derivative is in fact
quite small.
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
cementing relationships 279
to measure the extent of integration, not reported here for brevity, did
yield significantly positive estimated impacts in the regression without
multiunit and TFP controls. The estimates also become smaller and
insignificant once we added these controls.) Finally, the specifications
that control for state-year fixed effects follow the benchmark results
quite closely, qualitatively and quantitatively.
In sum, these quantity tests, as with the price tests above, give no
indication that vertical integration in the cement and ready-mixed industries
facilitates the exercise of market power. Vertically integrated
plants sell more and charge lower prices on average than their unintegrated
competitors in the same markets. The net quantity impact of
the entry of a vertically integrated firm into a market is positive because
the decline in output from acquired plants is compensated by additional
output from newly built plants. Furthermore, at the market level, increases
in the extent of integration in a market are associated with
greater total quantities sold. As with the price results, any such vertical
integration effect does not seem attributable to vertical integration itself,
but rather to concomitant increases in the presence of larger, more
efficient firms in the market. (As we will detail below, this appears to
arise because such firms are able to harness logistical efficiencies.) There
is also no evidence that foreclosure impacts exist but are simply being
empirically swamped by efficiency gains.
C. Do Integration’s Effects Differ across Markets with Different Ex Ante
Foreclosure Potential?
The tests above assume that vertical integration has common price and
quantity implications for all markets. However, the model exposited
earlier suggests that foreclosure potential may be linked to upstream
and downstream market structure. The upstream firm’s ability to leverage
market power gains by integrating depends in part on the extent
of its preintegration market power. Further, the heightened vertical antitrust
enforcement regime during part of our sample may also have
limited upstream producers’ foreclosure abilities during that period. We
test in this subsection whether there are noticeable differences in price
and quantity effects across four types of markets that theory suggests
offer greater ex ante foreclosure potential.
The first specification takes the model’s polar case quite literally. It
tests whether price and quantity impacts are systematically different in
ready-mixed markets that both (a) are served by a single, vertically integrated
cement firm and (b) have a small number of ready-mixed
firms—specifically, fewer than five. Market power gains are presumably
largest in these markets since unintegrated ready-mixed producers are
more vulnerable to being foreclosed. The second specification is a looser
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
280 journal of political economy
version of this cement monopoly test. We allow the integration coefficients
to differ depending on whether the local cement industry HHI
is above or below its mean value in our sample.30
The third variation
allows differential impacts in markets in which cement imports, the
availability of which could limit foreclosure power, are less likely to be
a viable alternative for cement buyers. We consider imports to be most
easily available in EAs containing a port that was one of the 25 largest
reception points for imported cement (using 2002 data from the U.S.
Customs Service) and if the year was 1977 or later, since before that
time cement imports were a trivial share of national consumption.31
Our
fourth and final market breakdown takes advantage of the antitrust
enforcement policy variation discussed above and allows different price
and quantity effects during census years other than 1967, 1972, and
1977, the period of the strictest stance toward vertical mergers by U.S.
antitrust authorities. If enforcement or potential enforcement limits
foreclosure possibilities, we should be more likely to see foreclosuretype
price and quantity responses during the lax enforcement period.
Table 11 shows the results of this exercise. Each set of estimates was
obtained using the benchmark specification and sample of the previous
two subsections, except here the price and quantity effects are allowed
to vary depending on whether or not one of the conditions above is
met. The “interacted” markets in each specification are chosen so that
foreclosure effects should be more likely in those markets than in the
remainder of the sample.32
When integration coefficients are allowed to differ in markets with
30
For these two specifications we consider cement plants’ service areas to be BEA economic
areas (described above) rather than the much larger USGS cement markets. Cement
monopolies are very rare in the USGS markets because of the way they are constructed,
and furthermore the model’s assumed transport cost–driven advantage of local cement
monopolists is likely to be much stronger when considered within the smaller market
definition. We also realize that the upstream market structure measures used here are
themselves endogenous outcomes that might have been influenced by the vertical structure
itself rather than simply capturing pre-existing conditions. However, the variation in foreclosure
potential across markets in our two other specifications below is essentially
exogenous.
31
While imported cement may not always be consumed in the market in which it is
brought to port, the fraction moved to other markets is likely to be small. Land transport
of cement is expensive relative to its value, and “ports” in the Customs data include interior
cities on navigable waterways, such as St. Louis, Minneapolis, and Cleveland. One possible
factor that would counteract the ability of cement consumers to substitute imports is the
inclusion in imports of clinker (cement in an intermediate, postkiln, pellet-like stage) that
is purchased and then ground into cement and sold by domestic cement firms. Unfortunately,
the data do not reveal importers’ identities.
32
Here we use the specification without market fixed effects to highlight differences
across markets with differing foreclosure potential. All specifications control for the HHI
in the ready-mixed concrete market, demand density (price regressions) or demand (quantity
regressions), and year effects. Only specifications using the market share measure of
the extent of vertical integration are shown for space reasons, though similar results are
obtained using the number of integrated firms.
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
TABLE11
MarketExAnteForeclosurePotential,VerticalIntegration,andConcretePricesandQuantities
VerticallyInte-
gratedCement
Monopolist
CementHHIAbove
Mean
LowImport
Presence
LightAntitrustEn-
forcementPeriod
Quantity-
Weighted
Average
Price
(1)
Total
Quantity
(2)
Quantity-
Weighted
Average
Price
(3)
Total
Quantity
(4)
Quantity-
Weighted
Average
Price
(5)
Total
Quantity
(6)
Quantity-
Weighted
Average
Price
(7)
Total
Quantity
(8)
2
R.087.783.087.784.096.783.088.782
VerticalintegrationintensityϪ.132*
(.035)
.769*
(.161)
Ϫ.125*
(.029)
.758*
(.134)
Ϫ.085
(.058)
.551*
(.251)
Ϫ.107*
(.032)
.884*
(.205)
(Verticalintegrationintensity)#
indicator
.022
(.050)
Ϫ.158
(.207)
.008
(.126)
Ϫ.583
(.351)
Ϫ.057
(.063)
.201
(.266)
Ϫ.030
(.043)
Ϫ.322
(.206)
Impliedeffectininteracted
markets
Ϫ.110*
(.041)
.611*
(.172)
Ϫ.117
(.124)
.175
(.329)
Ϫ.143*
(.029)
.752*
(.145)
Ϫ.137*
(.037)
.562*
(.131)
Note.—Thistableshowsmarket-levelpriceandquantityregressionsinwhichthecoefficientontheextentofintegrationmeasureisallowedtovaryfordifferingtypesofmarkets.Allspecifications
areestimatedfromasampleof1,870CEAmarket-years.AllregressionscontrolfortheHHIanddensityofdemandinthemarketaswellasyeareffects(coefficientsnotreported).Seethetext
fordetails.Standarderrorsareclusteredbymarket.
*Significantatthe5percentlevel.
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
282 journal of political economy
vertically integrated monopolist cement firms and few ready-mixed firms
(cols. 1 and 2), the implied price effects are not significantly different
in these markets than in the rest of the sample. The negative priceintegration
and positive quantity-integration relationships remain even
in those markets that most closely match the polar case of the model
above. (Both of their magnitudes are about 20 percent smaller than in
the rest of the sample, but the differences are insignificant.) When this
comparison is broadened to all markets with above-average cement
HHIs, again there are no statistically significant differences in the priceand
quantity-integration correlations. However, a combination of the
magnitude of the interaction terms and their imprecision leads to the
implied price and quantity effects of vertical integration being insignificant—and
in the case of quantities, considerably smaller in size—in
markets with high cement HHIs. In the third set of interactions, markets
in which importing cement is unlikely to be an option still see negative
(positive) and significant associations between average prices (total
quantities) and the share of vertically integrated firms. The magnitudes
of the coefficients are slightly larger in no-import markets, but again
the differences are insignificant. Finally, when we allow differential price
and quantity effects during periods of lax antitrust policy attitudes toward
vertical mergers, we do not see foreclosure’s implications surfacing
in these periods. The relationships between market-level outcomes and
the extent of integration remain the same in these years as during the
tougher enforcement regime. The implied price effect is slightly larger,
the quantity effect slightly smaller, but neither significantly so.
Therefore, in final-goods markets that theory indicates are a priori
more susceptible to foreclosure, we still see a negative correlation between
integration and prices and a positive correlation between integration
and quantities. In none of the specifications is the magnitude
of the negative price-integration relationship statistically different in
“high–foreclosure potential” markets from the remainder of the sample.
In three of four specifications, integration does appear to be associated
with smaller (though still positive) quantity gains in such markets, but
this difference is statistically weak.
D. Does Vertical Integration Force Exit of Unintegrated Producers?
Foreclosure models predict that firms can use leverage afforded by vertical
integration to force unintegrated rivals out of business in both the
upstream and downstream industries. We test this implication in this
subsection.
The first test directly compares exit probabilities across integrated
and unintegrated producers. We regress an indicator for plant death
by the next census (i.e., it does not exist in the following census) on
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
cementing relationships 283
TABLE 12
Integration and Exit Probabilities: Likelihood of Exit across Integrated and
Unintegrated Plants
Concrete (N p 24,042) Cement (N p 1,010)
(1) (2) (3) (4) (5) (6) (7) (8)
2
R .091 .097 .092 .098 .446 .472 .447 .473
Vertical integration
indicator
Ϫ.053*
(.014)
Ϫ.004
(.014)
Ϫ.050*
(.014)
Ϫ.003
(.014)
Ϫ.083*
(.036)
.005
(.036)
Ϫ.082*
(.036)
.005
(.036)
Multiunit
indicator
Ϫ.087*
(.007)
Ϫ.084*
(.007)
Ϫ.233*
(.054)
Ϫ.232*
(.054)
TFP (revenue-
based)
Ϫ.075*
(.013)
Ϫ.058*
(.013)
Ϫ.033
(.051)
Ϫ.023
(.050)
Note.—The table shows results from linear probability models in which the dependent variable is an indicator equal
to one if the plant exits by the next Census of Manufactures. Market-year fixed effects are included in all regressions.
* Significant at the 5 percent level.
an integration status indicator as well as a full set of market-year fixed
effects.33
Table 12 contains the results for both concrete and cement plants.
Integrated plants in both industries are less likely to exit than their
unintegrated competitors in the same market-year. The implied differences
are nontrivial: integrated concrete plants have a 5.3-percentagepoint
lower probability of exit over a five-year intercensus period and
cement plants an 8.3-percentage-point lower probability. As points of
comparison, the unconditional intercensus exit rates across all plants
are 30.3 percent in ready-mixed concrete and 16.4 percent in cement.
Unlike the price and quantity results, these exit patterns are consistent
with foreclosure.
We add the multiunit indicator and plant TFP controls as shown in
columns 2–4 and 6–8.34
Both are negatively correlated with exit in both
industries. Interestingly, here multiunit status has greater influence on
the vertical integration coefficient than TFP does. It is significant in
every specification, and controlling for it alone is enough to make vertical
integration’s correlation with exit fall to zero in both industries.
When we control for both factors simultaneously, the vertical integration
indicators for both concrete and cement plants remain statistically insignificant
and have small magnitudes.
So again we see that controlling for multiunit status and TFP explains
33
We do not count as exits plants that leave the cement or concrete industries but
continue to operate in a different industry. The exits with which we concern ourselves—
complete disappearances of plants from the Census of Manufactures universe—correspond
to the cessation of economic activity at the former plant’s location. We have also run
Probit exit models and found results similar to those reported here.
34
Note that in these regressions, we use revenue-based TFP rather than the physical
output–based measure used above, since we do not have physical output for cement plants.
Moreover, revenue-based TFP may be the better control variable here since it captures
both technology and demand fundamentals, both of which affect survival.
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
284 journal of political economy
TABLE 13
Integration and Exit Probabilities: Market-Level Exit
Rates of Non–Vertically Integrated Plants
Concrete Cement
Observations 1,725 126
2
R .265 .537
Market share of vertically
integrated firms
.033
(.061)
.084
(.163)
Note.—The dependent variable is constructed as the number of non–vertically integrated
exiting plants divided by the average number of non–vertically integratedplants
in the market over the last two censuses. Market fixed effects are included in both
specifications. Standard errors are clustered by market.
away any impact of vertical integration not connected to size or productivity.
Vertically integrated plants are more likely to survive not because
of any particular facet of being integrated, but instead because
they are in large firms and have higher productivity levels. This is inconsistent
with foreclosure, which implies that integration itself has a
survival benefit. A productivity-based story of vertical merger effects, on
the other hand, does not predict a differential impact of integration
except through its correlation with productivity and size.35
We next look at how market-level vertical integration affects exit of
unintegrated producers. We regress the exit rate of unintegrated producers
in a market on the extent of integration, controlling for market
and year fixed effects, the industry’s HHI in the local market, and demand
growth between the present and future censuses.36
The results
are in table 13.
The extent of integration is positively related to unintegrated producers’
exit rates in both industries, reflecting in part the differences
in likelihood of across-plant exits seen above. However, the link here
is statistically insignificant and economically small: in concrete (cement),
raising the market share of vertically integrated producers from
35
These results are consistent with earlier empirical evidence. Many studies have found
that more productive plants are less likely to exit. See, e.g., Baily, Hulten, and Campbell
(1992), Olley and Pakes (1996), Aw, Chen, and Roberts (2001), and several of the papers
reviewed in Bartelsman and Doms (2000). Further, a separate benefit of size for survival
was observed in Dunne, Roberts, and Samuelson (1989), perhaps because larger firms
have a greater ability to raise financial capital internally or because firm size acts as a sort
of “permanent TFP” proxy for the firm, capturing the long-run component of productivity
that is conflated with noise in the contemporaneous measure.
36
For the sake of symmetry with our entry rate analysis below, we construct the marketlevel
exit rate by dividing the number of exiting unintegrated plants by the average number
of unintegrated plants in the market in the current and future censuses rather than just
the count in the current period. For small percentage changes, both measures are closely
comparable. We also weight observations by market demand so as not to give undue
influence to markets with very small numbers of plants, where even a single plant exit
can create a large exit rate. We do not report coefficients on the controls for space reasons;
concrete HHI is negatively related to exit rates, but the results are otherwise unremarkable.
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
cementing relationships 285
zero to one—that is, across the extremes of the data—implies only a
3.3- (8.4-) percentage-point increase in the exit rate of unintegrated
producers. This is a fraction of the average plant-level exit rates above,
especially for concrete. Controlling for multiunit status and average
TFP in the market makes little difference in the estimate of the vertical
integration coefficient, and the controls themselves have insignificant
impacts on exit rates; we suppress these results for space reasons.
We see that, as foreclosure predicts, being vertically integrated is correlated
with a lower exit probability for a plant in either industry. However,
this relationship is driven by the fact that integrated plants are in
larger firms and have higher productivity levels. Once this is controlled
for, there is no longer per se a survival benefit of vertical integration.
Further, increases in the extent of integration within a market do not
lead to significantly higher exit rates among unintegrated producers in
that market.
E. Does Vertical Integration Reduce Entry?
Many foreclosure models, as well as the FTC report on cement-concrete
integration, imply that vertical mergers can lead to higher entry barriers.
This prediction has strong implications for long-run welfare, since giving
firms the ability to preserve their market power by restricting the entry
of new competitors is socially harmful. We explore the link between
entry and vertical integration here.
We compute two types of entry rates for each market. One is plantbased
and is calculated by dividing the number of new plants in the
market by the average number of plants over the current and previous
censuses.37
The second entry rate is firm-based. Here definition issues
are more complex. Unlike plants, which are geographically unique,
firms can operate in several markets. We must therefore take a stand
on how to treat entry of existing firms into new geographic markets.
For example, if a company that has operated in Texas for 15 years builds
a plant in Ohio, is that entry? We consider it as such because of the
highly geographically segmented nature of these industries’ product
markets. Note that this entry could occur not only by the firm building
a new plant but also by its purchase of an existing one. That is, a market
could experience firm entry but not plant entry. Firm entry rates for a
37
Again we use the average number of establishments in the denominator rather than
just the prior value. This keeps small markets that would otherwise have infinite or extremely
high computed entry rates from distorting the results. This transformation preserves
relative orderings. We also weight by market demand in our regressions to further
adjust for the small-market effect.
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
286 journal of political economy
market are the number of entry episodes divided by the average number
of firms operating in the market across the current and past censuses.38
The empirical specification is similar to the market-level exit tests
above: we regress entry rates in a market-year, which are based on entry
between the past and current censuses, on the market share of vertically
integrated establishments in the previous census. As in the exit regressions,
we control for market and year fixed effects as well as market
demand growth over the period to capture short-run entry impacts of
demand fluctuations. We estimate specifications using both overall entry
rates, which include entry by integrated and unintegrated producers,
and entry rates of unintegrated producers alone.
The results are presented in table 14, with panel A showing results
for overall entry rates and panel B for unintegrated producers. There
are no significant connections between entry rates and the market share
of vertically integrated firms. The economic magnitude of the vertical
integration coefficients is also small. As with the market-level exit rate
results, even a move from one extreme to the other—that is, a market
shifting from a situation in which integrated producers are completely
absent to one with 100 percent market share—implies small entry rate
changes compared to typical values in the data. For example, even the
largest implied effect, that on the plant-level entry rates of unintegrated
cement producers, moves entry rates by less than half of their mean
and a third of their standard deviation. Moreover, the coefficients are
not even negative for concrete. In unreported specifications for the
concrete industry in which we add multiunit firms’ market shares, these
shares are tied to lower entry rates (and are typically significant). However,
this also made the implied vertical integration effect more positive.
F. Discussion
The evidence suggests that foreclosure effects are not quantitatively
important when cement and ready-mixed producers vertically integrate.
More extensive integration in a market is not tied to higher prices and
38
There is an additional data difficulty regarding computation of firm-level entry rates.
While firm identifiers in the census remain constant if the plant-owning entity does not
change, there is an exception to this. Single-unit and multiunit firms are numbered according
to different systems, so when a plant in a single-unit firm becomes part of a
multiunit firm, its firm identifier changes from one format to the other. This presents no
measurement problems when a formerly single-unit plant is bought by a previously multiunit
firm; the plant simply assumes the identifier of the purchasing firm. However, if
the plant is purchased by a firm that was previously only a single-unit firm or if the plant’s
owner itself becomes multiunit by purchasing or opening another plant, the plant’s firm
identifier could change even though no new firm actually entered the market. We therefore
count as firm entry episodes only those instances in which a firm identifier changes
and the plant either was part of a multiunit firm in the previous census or was previously
a single-unit firm and its new firm identifier existed in the previous census.
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
cementing relationships 287
TABLE 14
Integration and Entry Rates
Concrete Cement
Plant-Level
Entry Rate
Firm-Level
Entry Rate
Plant-Level
Entry Rate
Firm-Level
Entry Rate
A. Overall Entry Rate (Number of Entrants Divided by
Average Plants/Firms in the Market over the Last Two
Censuses)
Observations 1,725 1,725 127 127
2
R .321 .354 .282 .423
Market share of integrated
firms in base year
.046
(.068)
.025
(.090)
Ϫ.024
(.127)
Ϫ.024
(.152)
B. Unintegrated Producer Entry Rate (Number of Unintegrated
Entrants Divided by Average Plants/Firms in
Market over the Last Two Censuses)
Observations 1,725 1,725 127 127
2
R .330 .366 .234 .332
Market share of integrated
firms in base year
.067
(.064)
.051
(.071)
Ϫ.073
(.106)
Ϫ.043
(.115)
Note.—This table shows the impact of integration on entry rates. Panel A shows the results of regressing overall
plant- and firm-level entry rates in a market (see the text for definitions) on the market share of integrated firms in
the base year. Panel B shows similar results for entry rates of unintegrated producers alone. Observations are weighted
by market demand. (Similar patterns were observed using lagged number of vertically integrated firms as the explanatory
variable, not reported here.) Standard errors are clustered by market.
lower quantities, but rather the opposite. These patterns are robust to
numerous empirical specifications and sources of identification. Even
when we focus on markets in which one would expect foreclosure opportunities
to be greater, the negative correlation between integration
and prices remains. Exit rates are lower for integrated plants in both
the upstream and downstream industries, as foreclosure would predict,
but we find that this can be explained by the fact that integrated plants
are more productive and are in larger firms, not because of a particular
effect of vertical integration itself. Finally, entry rates do not fall when
markets become more integrated.
It seems instead that the data are consistent with a story in which any
effects of vertical integration on outcomes arise because integrated firms
are larger and, especially, more productive. As these more efficient producers
expand their presence in a market, they pass on part of their
cost advantages to their customers in lower prices. This reduces average
prices in the market directly and also induces their higher-cost unintegrated
producers to lower their prices.39
Lower prices in turn increase
quantities sold. Those unintegrated producers that cannot remain profitable
when facing greater low-cost competition exit.
39
A negative correlation between price and productivity levels has also been documented
in previous empirical work (e.g., Roberts and Supina 1996; Eslava et al. 2004; Foster et
al. 2005; Syverson, forthcoming).
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
288 journal of political economy
TABLE 15
Differences between Integrated and Unintegrated Producers
Labor
Productivity
Real
Revenue
Total
Employment
Total
Hours
Nonproduction
Worker Ratio
Capital-
Labor
Ratio
A. Ready-Mixed Concrete
Observations 31,061 31,770 31,485 31,075 31,480 18,703
2
R .217 .213 .168 .166 .181 .331
Vertical inte-
gration
indicator
.292*
(.019)
.280*
(.034)
.015
(.033)
.047
(.034)
Ϫ.018*
(.007)
.099*
(.030)
B. Cement
Observations 1,476 1,484 1,480 1,476 1,480 902
2
R .552 .56 .584 .574 .492 .546
Vertical inte-
gration
indicator
.116*
(.046)
.820*
(.106)
.652*
(.097)
.648*
(.098)
Ϫ.034*
(.012)
.284*
(.098)
Note.—This table reports differences in key dependent variables (listed at the head of each column) across integrated
and unintegrated producers in the ready-mixed concrete and cement industries. The reported coefficients are those
for an indicator variable denoting that a plant is in a vertically integrated firm. Market-year fixed effects are included
in all specifications.
* Significant at the 5 percent level.
V. Why Do Integrated Producers Have Higher Productivity Levels?
The fact that the productivity advantage of vertically integrated producers,
rather than integration itself, can explain the patterns in the
data raises its own issue. What efficiency gains are harnessed through
integration, or is the connection merely a correlation? We look more
closely at how integrated producers differ from their unintegrated counterparts
to try to answer this question.
A. How Are Integrated Producers Different?
We first compare a number of plant-level technology measures across
integrated and unintegrated producers. As with the plant-level price,
TFP, and quantity comparisons above, we regress the variable of interest
on an indicator for integration status and a full set of market-year fixed
effects. This again focuses on comparisons among integrated and unintegrated
producers in the same market.
These comparisons for each industry’s plants are shown in table 15.
Differences among ready-mixed concrete plants are in panel A and
among cement producers in panel B.40
Notice first the labor productivity
differences. Integrated ready-mixed producers’ labor productivity levels
40
The sample size differences across dependent variables reflect variation in the availability
of the underlying production data necessary to compute the measures. See the
discussion in the Data Appendix.
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
cementing relationships 289
are 33 percent (29 log points) higher than those of their unintegrated
counterparts in the same market. This is considerably larger than the
TFP differences seen in table 5 and reflects that fact that, as can also
be seen in the table, integrated plants are more capital intensive. In the
cement industry, integrated producers also have higher labor productivity
levels, by just under 10 percent, than unintegrated cement plants
in their market.
Besides being more productive, integrated ready-mixed plants do a
larger volume of business than unintegrated plants in their markets;
their real revenue is on average about 30 percent higher. Despite their
higher output, integrated ready-mixed plants do not employ significantly
more labor (workers or hours), as manifested in the higher labor productivity
result. There are differences in labor composition, however: nonproduction
workers constitute a smaller share of the labor force in
integrated plants.41
This suggests that integrated producers may be able
to harness gains coming from the consolidation of administrative duties,
a point to which we shall return below. In a comparison of cement
plants, integrated producers are also considerably larger in terms of
total sales, have a smaller share of nonproduction workers, and have
higher capital-labor ratios than their unintegrated counterparts. Unlike
concrete plants, they also hire more labor.
Integrated plants are indeed different. However, these patterns are
less useful for determining whether the differences are caused by vertical
integration. As above, we look at additional comparisons that are less
likely to confound the causal impacts of vertical integration with selection
on pre-existing heterogeneity. Panel A of table 16 compares plants
that were unintegrated in the previous census but became integrated
to those that remained unintegrated (we control for market-year fixed
effects in this and all these comparisons). Those becoming integrated
in the interim had 10 percent faster labor productivity growth than their
market counterparts that remained unintegrated. As in the physical
quantity comparisons above, sales revenue at these newly integrated
ready-mixed plants shrink (at least in relative terms; the reported coefficients
give relative growth rates).42
Also as previously mentioned,
41
The U.S. Census Bureau defines production workers (all other employees are considered
nonproduction workers) as “workers (up through the line-supervisor level) engaged
in fabricating, processing, assembling, inspecting, receiving, packing, warehousing,
shipping (but not delivering), maintenance, repair, janitorial, guard services, product
development, auxiliary production for plant’s own use (e.g., power plant), record keeping,
and other closely associated services (including truck drivers delivering ready-mixed concrete).
Exclude proprietors and partners” (form MC-31000).
42
While in the absence of randomized integration at the plant level we cannot be sure
that these estimates reflect exclusively causal effects, our productivity-based explanation
for the results above requires only that vertical integration is correlated with productivity.
We also find it difficult to form a reasonable explanation for the observed patterns that
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
290 journal of political economy
TABLE 16
Becoming Integrated: Ready-Mixed Concrete Continuers and Entrants
Growth of
Labor
Productivity
Real
Revenue
Total
Employment
Total
Hours
Nonproduction
Worker Ratio
Capital-
Labor
Ratio
A. Changes among Continuers (Conditioning on Being Unintegrated in
Previous Census)
Observations 15,919 16,358 16,274 15,933 9,166 16,271
2
R .194 .274 .204 .217 .189 .180
Vertical inte-
gration
indicator
.105*
(.049)
Ϫ.399*
(.060)
Ϫ.396*
(.061)
Ϫ.439*
(.064)
Ϫ.030
(.022)
.018
(.076)
B. Integrated Entrants Compared to Unintegrated Entrants
Observations 7,681 8,005 7,871 7,687 7,870 5,405
2
R .330 .339 .325 .323 .34 .433
Vertical inte-
gration
indicator
.336*
(.047)
.125
(.079)
Ϫ.157*
(.073)
Ϫ.162*
(.078)
Ϫ.044*
(.014)
.263*
(.073)
C. Integrated Entrants Compared to Unintegrated Incumbents
Observations 18,038 18,310 18,220 18,045 18,217 12,760
2
R .240 .228 .193 .195 .222 .378
Vertical inte-
gration
indicator
.358*
(.038)
Ϫ.281*
(.063)
Ϫ.510*
(.060)
Ϫ.514*
(.064)
Ϫ.061*
(.012)
.259*
(.058)
Note.—This table reports differences in key dependent variables (listed at the head of each column) across integrated
and unintegrated producers. Panel A compares growth rates across integrated and unintegrated continuers (plants that
survive for two consecutive censuses). Panel B compares integrated and unintegrated entrants (plants appearing in
their first census). Panel C compares integrated entrants to unintegrated incumbents. Market-year fixed effects are
included in all specifications.
* Significant at the 5 percent level.
though, the decline in sales at acquired concrete plants is on average
more than compensated by new entrants. Panels B and C of table 16
compare integrated entrants to, respectively, unintegrated entrants and
unintegrated incumbents. Integrated entrants have higher sales than
unintegrated entrants, though they hire less labor. New ready-mixed
plants in integrated firms are also noticeably more productive than both
unintegrated entrants and unintegrated incumbents. As with the comparisons
among all producers, integrated entrants have lower nonproduction
worker ratios and are more capital intensive than both comparison
groups.
Similar comparisons for cement plants—not reported here—found
no significant differences among either continuing plants or entrants.
This suggests, interestingly, that the differences seen between integrated
and unintegrated cement producers are driven by selection into inteimplies
reverse causation, i.e., why integrating firms would explicitly target plants that are
expected to increase in productivity and shrink in the near future.
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
cementing relationships 291
gration based on pre-existing differences, whereas some of the differences
among ready-mixed concrete plants are more likely to be a result
of becoming integrated (or being newly built by an integrated firm).
Moreover, vertically integrated firms may have the largest impact on
concrete markets in terms of productivity through their newly built
plants. We return to these points below.
B. A Specific Mechanism
The findings above, along with a consideration of practical features of
the ready-mixed concrete industry, suggest an explanation for the observed
integration-productivity link: improved logistical coordination.
Operating in the ready-mixed industry requires delivering a perishable
product to time-sensitive buyers in multiple locations. Having several
plants in a local area and coordinating deliveries through a central office
could benefit a firm by consolidating overhead (one dispatcher might
handle deliveries from several plants that would each have separate
dispatchers in single-unit firms) and allowing more efficient use of available
resources through cross-plant substitution.43
This type of productivity gain is consistent with the above findings.
Integrated producers’ lower nonproduction worker ratios suggest that
they have lower overhead labor requirements. The pursuit of coordination
efficiencies explains why integrated producers might reshuffle
production locations in markets by reducing output at purchased plants
and replacing it with production from their newly built plants, perhaps
located nearer spatially shifting demand within the market. Such efficiencies
are also used to motivate some stated practices of integrated
producers. For instance, the large integrated producer Lafarge (a firm
that was just starting to develop a sizable presence in the United States
at the end of our sample period) describes its preferred placement of
ready-mixed operations in its 2004 20-F filing: “We aim to place our
ready mix concrete plants in clusters in each micro market in which we
operate in order to optimize our delivery flexibility, capacity and backup
capability. . . . We evaluate each micro market in which we operate
periodically and dismantle and move plants to locations where they can
43
In an earlier version of this paper, we explored in detail two other potential sources
of integrated producers’ productivity advantage: elimination of double marginalization
and improved inventory management and production smoothing. We did not find evidence
that either was the source of the observed productivity gap. See Hortac¸su and
Syverson (2006) for details.
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
292 journal of political economy
be used more profitably” (Lafarge 2005, 39). Plant clustering allows
firms to better take advantage of any coordination efficiencies.44
This raises the question, however, of why cement firms need to be the
coordinating body. A possible answer is that they need not be. The readymixed
industry has seen considerable consolidation over our sample—
in 1963, 3,999 firms owned 4,621 ready-mixed plants, but by 1997, only
2,898 firms owned 5,252 plants—and much of this consolidation came
through horizontal rather than vertical mergers. As seen in table 2
above, the fraction of ready-mixed plants in multiunit firms more than
doubled between 1963 and 1997, from 24.8 to 55.6 percent. The corresponding
change in the fraction of vertically integrated plants was
from 1.8 to 10.6 percent. Clearly, then, the majority of merged readymixed
concrete plants were folded into firms without cement operations.
Coordination and its possible efficiency gains, if they are tied
primarily to firm size, have therefore not been exclusive to vertically
integrated firms.45
We explore this notion further by repeating our earlier comparisons
across integrated and unintegrated plants, but this time adding to the
specification the total ready-mixed sales of the plant’s firm in the CEA.
If logistics coordination benefits are facilitated by having a greater extent
of local concrete operations, controlling for the size of the firm’s local
ready-mixed sales will let us see whether the relationships observed
above between vertical integration and concrete plants’ prices, productivity
levels, scale, and technologies are tied to this.
The results of this exercise, shown in table 17, are instructive, especially
in comparison to the earlier plant-level price, TFP, and quantity
contrasts as well as panel A of table 15. When we control for local
concrete sales, the magnitude of vertically integrated plants’ price and
44
We note that nothing about these efficiency gains requires spatial proximity between
an integrated firm’s cement and concrete operations, or even that the upstream division
supply the downstream division at all. Indeed, Lafarge owns about two dozen ready-mixed
plants in Colorado and Wyoming, even though its nearest cement distribution terminal
is 490 miles away in Oklahoma City. We contacted some of these plants, and they confirmed
that they receive all their cement inputs from sources outside of Lafarge. A similar situation
exists for Florida Rock Industries, an integrated producer that owns ready-mixed plants
in Florida, southern Georgia, Virginia, and Maryland. The firm’s cement facilities,however,
are all located in Florida, and the company’s financial filings state that the company’s
concrete operations purchased cement from 10 outside suppliers (Florida Rock Industries
2005). Phone calls to management verified complete external cement sourcing among
the company’s 30 or so ready-mixed plants in Virginia and Maryland. We also note that
these practices are hard to reconcile with foreclosure motives. Foreclosure necessarily
implies that integrated downstream producers are at the very least being partially supplied
by upstream producers in their firm.
45
That is not to say that a large firm in any industry could harness coordinating economies,
of course. But cement shares a final demand sector with ready-mixed, and the two
industries do have other key elements in common, such as the prominence of logistical
concerns.
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
cementing relationships 293
TABLE 17
Differences between Integrated and Unintegrated Ready-Mixed Concrete
Plants, Controlling for Firm’s Total Sales in Local Market
Output
Price TFP
Physical
Output
Labor
Productivity
Nonproduction
Worker Ratio
Capital-
Labor
Ratio
Observations 12,553 8,555 12,553 31,061 31,468 18,702
2
R .426 .329 .624 .292 .187 .35
Vertical integration
indicator
Ϫ.012
(.007)
Ϫ.015
(.014)
Ϫ.237*
(.042)
.097*
(.019)
Ϫ.039*
(.007)
Ϫ.037
(.031)
Firm’s local readymixed
sales
Ϫ.009*
(.002)
.055*
(.004)
.633*
(.008)
.161*
(.004)
.017*
(.001)
.113*
(.006)
Note.—This table reports differences in key dependent variables (listed at the head of each column) across integrated
and unintegrated producers in the ready-mixed concrete industry. The reported coefficients are those for a vertically
integrated plant dummy and the coefficient on the (log of the) total concrete sales of the plant’s firm in the local
market. Market-year fixed effects are included in all specifications.
* Significant at the 5 percent level.
productivity differences falls and their significance disappears. In other
words, while plants in vertically integrated firms charge lower prices and
are more productive on average than unintegrated plants, they do not
have significantly different prices or TFP levels than plants in unintegrated
firms having equally sized local concrete sales. Recall that most
of the plant- and market-level relationships between vertical integration
and outcomes above become insignificant once we control for multiunit
status and TFP. These results suggest that firm size and TFP are integrally
related, specifically, that the firms’ horizontal size downstream is linked
to how efficient its individual concrete plants are. Their productivity
advantage in turn allows them to charge lower prices and makes them
more likely to survive.46
Some differences do remain, however, between
integrated and unintegrated plants whose firms have similar downstream
sales: integrated plants are smaller (i.e., the same sales are spread across
more plants), they still have a labor productivity advantage, and they
use less overhead labor.
On the whole, this evidence, both anecdotal and from our data, appears
to suggest that vertical integration per se might not be the primary
source of integrated producers’ productivity advantage. Rather, integrated
producers—as with any producer with sufficient size and perhaps
the necessary complementary managerial talent, even one with a purely
horizontal firm organization—seem to be able to take advantage of the
46
The strictest interpretation of the logistics coordination story would imply that the
large-firm and TFP impacts would be perfectly collinear, since one completely explains
the other. We do not observe this in the data, probably because, among other reasons,
multiunit status is a noisy measure of a firm in the downstream market, and there are
almost surely other firm-specific determinants of productivity that are not related to firm
size.
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
294 journal of political economy
logistical coordination gains from having geographically clustered plants
whose logistics are centrally managed.
VI. Concluding Remarks
We have used unusually detailed data to investigate vertical integration
in the cement and ready-mixed concrete industries, with a particular
focus on the possible foreclosure effects of integration. These industries
hold several features that make them an excellent candidate for an
empirical case study on the issue. Their downstream markets are highly
geographically segmented (especially for ready-mixed, where the vast
majority of output is shipped less than 100 miles), providing us with
considerable variation to empirically identify effects of interest. Second,
cement and ready-mixed are relatively homogeneous in physical attributes
and have little brand differentiation. Therefore, the competitive
effects we find—on prices, for instance—more likely arise from changes
in market structure than from alterations in product mix. Third, we
have access to detailed plant-level production information for these
industries, affording more variation in vertical market structures
(thousands of producers operating in hundreds of local markets over
a 34-year time span) than was typically available to previous researchers
and allowing us to study interacting effects that previous studies by
necessity examined in isolation. We also explore for the first time elements
of theoretical models that have not yet (to our knowledge) been
studied empirically, including vertical integration’s long-run competitive
impacts on entry and exit. Finally, vertical integration among producers
in these two industries has been the focus of substantial policy attention
both in the United States and elsewhere.
We find little evidence that foreclosure takes place when firms integrate
in these industries. Instead, prices fall, quantities rise, and entry
rates do not change when markets become more integrated (measured
by either the market share or the number of vertically integrated firms
in the market).
These patterns are consistent with an alternative, efficiency-based
mechanism: namely, that greater vertical integration in a market corresponds
to the expansion of larger, more productive integrated firms
as they take market share from higher-priced, less efficient producers.
This alternative explanation is reflected in the fact that, once we control
for firm size and productivity impacts, there is very little role for vertical
integration itself in explaining plant- and market-level outcomes. We
went on to show evidence suggesting that linking integration, firm size,
and productivity are gains in logistical coordination efficiencies tied to
having multiple concrete plants in the same market. However, these
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
cementing relationships 295
efficiencies appear not to be tied to vertical integration itself, but rather
firm size in the downstream market more generally.
Our results, if correct, point to an interesting trade-off for the social
welfare implications of vertical integration. It may be that in the cement
and ready-mixed industries, vertical integration is just one way for large,
efficient firms to enter a market. In fact there has been extensive consolidation
in the ready-mixed industry over the past four decades, and
vertical integration has been one way in which industry plants have been
folded into larger firms. As we show above, the churning process underlying
such consolidations may have had significant impacts on productivity
in the industry. This presents again the familiar difficulty of
balancing efficiency and market power considerations in merger
analysis.
We conclude by emphasizing that one should be careful when trying
to generalize the results in this paper to other industries or markets.
Vertical integration decisions of firms are driven by many different considerations,
of which vertical foreclosure may or may not be important
in any particular case. Our study is an attempt to study the short- and
long-term effects of the decision to integrate in these particular industries.
The fact that we do not find evidence of foreclosure here does
not mean that market power gains made possible through vertical integration
are not quantitatively important elsewhere.
Data Appendix
We describe here details on the construction of our production variables. We
first note that while the Census of Manufactures microdata contain enormous
amounts of production information, they do not offer full coverage for every
variable discussed below. Very small plants (typically with fewer than five employees)—called
administrative record establishments—have imputed data for
most production variables. Administrative record plants amount to roughly onesixth
of cement establishments and one-third of ready-mixed plants, but because
of their small size they constitute a much smaller share of employment (0.6
percent in cement—these are almost surely grinding-only plants without kilns—
and 5.1 percent in ready-mixed) and output (0.8 and 4.3 percent in cement
and ready-mixed, respectively). Because of the imputations, we exclude administrative
record plants from analyses that compare production variables such as
productivity, output, or prices. However, we are of course able to use these plants
when computing entry and exit rates or integration status. Additionally, not
every variable was collected in each census. For example, equipment capital
stocks were not collected in 1963 and 1997, making it impossible to compute
TFP values and capital-labor ratios during these years. We have removed likely
imputes (they are not explicitly flagged) from our sample using methods described
in Roberts and Supina (1996) and Syverson (forthcoming).
Labor hours.—Production worker hours are reported directly in the Census of
Manufactures microdata. This value is then scaled up to total hours by multiplying
by the ratio of total employees to production workers. This assumes, in essence,
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
296 journal of political economy
that average nonproduction worker hours equal average production worker
hours within plants.
Capital stocks.—Equipment and building capital stocks are plants’ reported
book values of each capital type deflated by the book to real value ratio for the
corresponding three-digit industry. (These industry-level equipment and structures
stocks are taken from published BEA data.) Any reported machinery or
building rentals by the plant are inflated to stocks by dividing by a type-specific
rental rate.47
The total productive capital stock is the sum of the equipmentkit
and structures stocks.
Real materials and energy use.—Materials and energy inputs are plants’ reported
expenditures on each divided by their respective industry-level deflators from
the NBER Productivity Database.
Labor productivity.—We measure labor productivity as plant output per workerhour,
where output is the real value of shipments and hours are constructed as
above.
Total factor productivity (revenue- and physical quantity–based).—We measure productivity
using a standard TFP index. Plant TFP is computed as its logged output
minus a weighted sum of its logged labor, capital, materials, and energy inputs.
That is,
TFP p y Ϫ a l Ϫ a k Ϫ a m Ϫ a e ,it it lt it kt it mt it et it
where the weights aj are the input elasticities of input . While inputsj ෈ {l, k, m, e}
are plant-specific, we use industry-level input cost shares to measure the input
elasticities.48
These cost shares are computed using reported industry-level labor,
materials, and energy expenditures from the NBER Productivity Database (which
is itself constructed from the Census of Manufactures). Capital expenditures are
constructed as the reported industry equipment and building stocks multiplied
by their respective capital rental rates in cement and ready-mixed concrete’s
corresponding two-digit industry.
For both industries, we construct TFP measures using plants’ reported revenues
(deflated to 1987 dollars using price indexes from the NBER Productivity
Database) as an output measure. This is the standard measure used in the
literature and is the one we use in the investigation of plant exit patterns. (This
revenue-based measure is used both so that the results from both industries can
be compared and because as explained in Foster et al. [2005], revenue-based
productivity measures embody both technology- and demand-side profitability
47
Capital rental rates are taken from unpublished data constructed by the Bureau of
Labor Statistics for use in computing its multifactor productivity series. Formulas, related
methodology, and data sources are described in U.S. Bureau of Labor Statistics (1983)
and Harper, Berndt, and Wood (1989).
48
This index makes a few implicit assumptions that are worth discussing further. One
is that returns to scale are constant. If the scale elasticity were instead different from one,
each of the input elasticities aj should be multiplied by the scale elasticity. In earlier work,
Syverson (2004) finds that returns to scale in ready-mixed concrete are essentially constant.
Also, using industry-level input elasticities assumes that all plants in the industry use the
same technology. While a common assumption in the literature, the fact that we observe
different capital intensities at vertically integrated plants could raise questions about
whether this is appropriate. However, different input intensities can also result from producers
facing different factor prices. In this case, it would imply that integrated producers
face lower capital prices than unintegrated producers. Given that integrated plants are
owned by firms that are on average substantially larger than those owning unintegrated
plants, and as such may have access to easier credit, we find such a factor price difference
plausible.
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
cementing relationships 297
fundamentals that determine exit probabilities.) However, for ready-mixed
plants when data are available, we also construct a TFP measure based on the
plant’s physical output. This removes the influence of within-industry price variation
on the output measure. This is the TFP concept used in all other exercises
that use only ready-mixed concrete plants.
We must make one adjustment to the output data when computing physicaloutput
TFP. Since ready-mixed plants can produce multiple products (though
most do not, as discussed above), but inputs are reported on an establishmentwide
rather than product-specific basis, we must impute the share of inputs
allocated to ready-mixed production in multiproduct plants. We do so by dividing
reported ready-mixed output by its share of total establishment sales. This adjustment
in effect assumes that inputs are used proportionately to each product’s
revenue share. (For example, a plant producing 1,000 cubic yards of readymixed
concrete accounting for 80 percent of its revenues will have the same
TFP as a completely specialized plant producing 1,250 cubic yards with the same
measured inputs.)
Output and factor prices.—We use product-level revenue and physical production
and consumption data from the Census of Manufactures Product and Materials
Supplements to compute ready-mixed plants’ unit concrete output prices
and cement factor input prices. We then adjust these to a common 1987 basis
using the corresponding four-digit industry-level shipments deflators from the
NBER Productivity Database.
There are two important notes regarding these calculated unit prices. First,
the value of shipments (sales revenue) data are collected on a free-on-board
basis, that is, exclusive of any shipping costs. Prices should reflect not the delivered
cost of the ready-mixed but rather what one could buy it for at the plant
gate. Second, the unit prices are annual averages. This can be shown to be
equivalent to a quantity-weighted average of all transaction prices charged by
the plant during the year. We do not observe product-specific data for administrative
record plants, so they are dropped from the analysis as in the core
sample. We also remove a small number of gross outliers having prices greater
than five times or less than one-fifth the median in a given year and limit the
sample to those plants with ready-mixed sales accounting for over one-half of
yearly revenues. (This sample criterion is not very restrictive in practice; most
ready-mixed producers are specialists.) Finally, we attempt to exclude any non–
administrative record plants that have (mostly because of incomplete reporting)
physical quantities imputed by the Census Bureau. Unfortunately, these imputes
are not flagged. To distinguish and remove imputed product-level data from
the sample, we use the techniques described in detail in Roberts and Supina
(1996) and Foster et al. (2005).
Imputing construction-sector employment when County Business Patterns data are missing.—County
Business Patterns data occasionally have missing observations because
of data disclosure regulations. The construction sector’s ubiquity and
abundance of small firms allow full disclosure of total employment in nearly all
counties, however (employment data are withheld in roughly 1.5 percent of the
county-year observations). We impute employment when missing by multiplying
the number of establishments in each of nine employment ranges (which are
always reported) by the midpoint of their respective employment ranges and
summing the result. The impact of using imputes is likely to be even less than
their proportion indicates, since the typically small nondisclosure counties are
less likely to contain ready-mixed plants.
Construction of antitrust policy and spatial demand instruments.—In our instruThis
content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
298 journal of political economy
mental variable specifications, we use three sets of instruments. Two use variation
in antitrust enforcement across our markets and the other works off spatial
patterns of demand (i.e., construction activity) within concrete and cement
markets.
The first antitrust policy instrument is an indicator equal to one if the smallest
firm in a market has a share of at least 4 percent. This is based on the fact that
U.S. antitrust authorities’ horizontal merger guidelines have changed over time.
During a particularly strict period, any merger involving a firm with more than
4 percent market share was open to challenge. Thus in markets in which the
smallest firm was already above that threshold, any merger in the market would
be suspect. Since the guidelines applied across industries, these guidelines were
obviously exogenous to structure in any given ready-mixed concrete market and
should be exogenous shifters of horizontal market structure. We interact the
indicator with another equal to one during the 1967, 1972, and 1977 censuses
since this is the period in which the policy was in effect.
The second antitrust policy instrument comes directly from the FTC’s antitrust
actions in the cement and ready-mixed industries. In January 1967, in response
to industry complaints about the unpredictability of merger challenges, the FTC
announced a set of guidelines that promised challenges to any acquisitions by
cement producers of any “substantial” ready-mixed concrete company. “Substantial”
was defined as a ready-mixed firm that was one of the four largest in
its market or one that used over 50,000 barrels of cement per year (Wall Street
Journal 1967). This rule was dropped by 1977. Since the combined market share
of substantial firms plausibly varies across markets for reasons unrelated to equilibrium
outcomes (at the very least once exogenous demand is conditioned on),
some markets are likely to receive more FTC scrutiny than others simply because
of the specifics of the rule. As with the changes in the horizontal merger guidelines,
the fact that the rule was in effect only during the 1967–77 censuses also
creates intertemporal variation in enforcement likelihoods. We therefore construct
for each market-year in our sample the total market share of substantial
concrete firms. The first criterion is easily evaluated using plant-level revenue
aggregated to the firm level within the market. For the cement use criterion,
we employed the census materials supplement. These data allow us to observe
the physical quantity of cement purchases made by a subsample of ready-mixed
concrete plants. We regressed cement purchases on plant revenues for these
plants (which tended to be the largest in the industry) to obtain a total firm
revenue value such that firms with greater real sales than this would be expected
to buy more than 50,000 barrels of cement. This allowed us to determine which
firms met the FTC’s second definition of “substantial” even if we did not observe
physical cement purchases for the entire firm. Finally, we interact this share of
substantial firms with an indicator variable denoting the 1967, 1972, and 1977
censuses as above. This variation in likely FTC scrutiny is used to instrument
for the extent of integration.
The third instrument set reflects measures of the dispersion of demand within
our sample markets. They are constructed from the county-level construction
employment data aggregated to the market level above. We use three measures:
the standard deviation in both the (log) level and growth rates of demand across
the market’s counties, and the HHI of counties’ construction employment shares
in the market (here, employment is in levels rather than logs). The logic of
these instruments is as follows. As with the market-level aggregates, they should
be exogenous to ready-mixed market outcomes because of concrete’s small share
in construction costs. Further, demand (or demand growth) variation across
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
cementing relationships 299
areas in a market plausibly affects the returns to a firm owning multiple plants.
If demand is concentrated in a small space, a firm may be able to serve it
sufficiently with a single plant (albeit possibly a large one, depending on the
level of demand). But if demand is instead spread spatially, the industry’s high
transport costs may make it worthwhile to own several plants throughout the
market. Both the standard deviation of logged employment and the HHI of
employment levels measure the market’s spatial demand variation, for a given
demand level (which we control for directly in the regressions). The standard
deviation in demand growth rates measures the market’s demand volatility,which
may give firms an incentive to own multiple plants in different locations in a
market to diversify away idiosyncratic plant-level risk.
References
Allen, Bruce T. 1971. “Vertical Integration and Market Foreclosure: The Case
of Cement and Concrete.” J. Law and Econ. 14 (April): 251–74.
Asker, John. 2004. “Diagnosing Foreclosure Due to Exclusive Dealing.” Manuscript,
New York Univ., Stern School Bus.
Aw, Bee Yan, Xiaomin Chen, and Mark J. Roberts. 2001. “Firm-Level Evidence
on Productivity Differentials and Turnover in Taiwanese Manufacturing.” J.
Development Econ. 66 (October): 51–86.
Baily, Martin N., Charles Hulten, and David Campbell. 1992. “Productivity Dynamics
in Manufacturing Plants.” Brookings Papers Econ. Activity: Microeconomics,
pp. 187–249.
Bartelsman, Eric J., and Mark Doms. 2000. “Understanding Productivity: Lessons
from Longitudinal Microdata.” J. Econ. Literature 38 (September): 569–94.
Bernheim, B. Douglas, and Michael D. Whinston. 2004. “Anticompetitive Exclusion
and Foreclosure through Vertical Agreements.” Manuscript, Northwestern
Univ.
Bolton, Patrick, and Michael D. Whinston. 1993. “Incomplete Contracts, Vertical
Integration, and Supply Assurance.” Rev. Econ. Studies 60 (January): 121–48.
Bork, Robert H. 1978. The Antitrust Paradox: A Policy at War with Itself. New York:
Basic Books.
Chipty, Tasneem. 2001. “Vertical Integration, Market Foreclosure, and Consumer
Welfare in the Cable Television Industry.” A.E.R. 91 (June): 428–53.
Coase, Ronald H. 1937. “The Nature of the Firm.” Economica, n.s., 4 (November):
386–405.
Dunne, Timothy, Mark J. Roberts, and Larry Samuelson. 1989. “The Growth
and Failure of U.S. Manufacturing Plants.” Q.J.E. 104 (November): 671–98.
Eslava, Marcela, John Haltiwanger, Adriana Kugler, and Maurice Kugler. 2004.
“The Effects of Structural Reforms on Productivity and Profitability Enhancing
Reallocation: Evidence from Colombia.” J. Development Econ. 75 (special issue;
December): 333–71.
Florida Rock Industries. 2005. “10-K Report for 2004.” http://www.flarock.com/
flarock/index.asp?bwsp72819243&bwwp44.
Foster, Lucia, John Haltiwanger, and Chad Syverson. 2005. “Reallocation, Firm
Turnover, and Efficiency: Selection on Productivity or Profitability?” Working
Paper no. 11555 (August), NBER, Cambridge, MA.
FTC (U.S. Federal Trade Commission). 1966. Economic Report on Mergers and
Vertical Integration in the Cement Industry. Washington, DC: U.S. Government
Printing Office.
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
300 journal of political economy
———. 1985. “Cement Industry: Rescission of Enforcement Policy with Respect
to Vertical Mergers in the Cement Industry.” Fed. Register 50 (May): 21507–8.
Grimm, Curtis M., Clifford Winston, and Carol A. Evans. 1992. “Foreclosure of
Railroad Markets: A Test of Chicago Leverage Theory.” J. Law and Econ. 35
(October): 295–310.
Grossman, Sanford J., and Oliver D. Hart. 1986. “The Costs and Benefits of
Ownership: A Theory of Vertical and Lateral Integration.” J.P.E. 94 (August):
691–719.
Harper, Michael J., Ernst R. Berndt, and David O Wood. 1989. “Rates of Return
and Capital Aggregation Using Alternative Rental Prices.” In Technology and
Capital Formation, edited by Dale W. Jorgenson and Ralph Landau. Cambridge,
MA: MIT Press.
Hart, Oliver, and Jean Tirole. 1990. “Vertical Integration and Market Foreclosure.”
Brookings Papers Econ. Activity (special issue), pp. 205–76.
Hastings, Justine S. 2004. “Vertical Relationships and Competition in Retail Gasoline
Markets: Empirical Evidence from Contract Changes in Southern California.”
A.E.R. 94 (March): 317–28.
Hastings, Justine S., and Richard J. Gilbert. 2005. “Market Power, Vertical Integration
and the Wholesale Price of Gasoline.” J. Indus. Econ. 53 (December):
469–92.
Hortac¸su, Ali, and Chad Syverson. 2006. “Cementing Relationships: Vertical Integration,
Foreclosure, Productivity, and Prices.” Working Paper no. WP-06-
21, Center Econ. Studies, Washington, DC.
Johnson, Kenneth P. 1995. “Redefinition of the BEA Economic Areas.” Survey
Current Bus. 75 (February): 75–81.
Johnson, Ronald N., and Allen M. Parkman. 1987. “Spatial Competition and
Vertical Integration; Cement and Concrete Revisited: Comment.” A.E.R. 77
(September): 750–53.
Kreps, David M., and Jose´ A. Scheinkman. 1983. “Quantity Precommitment and
Bertrand Competition Yield Cournot Outcomes.” Bell J. Econ. 14 (Autumn):
326–37.
Lafarge. 2005. “Report on Form 20-F.” http://www.lafarge.com/lafarge/
PUBLICATION/20050407/04072005-publication_finance-20f_2004-uk.pdf.
Martin, Stephen, Hans-Theo Normann, and Christopher M. Snyder. 2001. “Vertical
Foreclosure in Experimental Markets.” Rand J. Econ. 32 (Autumn): 466–
96.
McBride, Mark E. 1983. “Spatial Competition and Vertical Integration: Cement
and Concrete Revisited.” A.E.R. 73 (December): 1011–22.
Nocke, Volker, and Lucy White. Forthcoming. “Do Vertical Mergers Facilitate
Upstream Collusion?” A.E.R.
O’Brien, Daniel P., and Greg Shaffer. 1992. “Vertical Control with Bilateral Contracts.”
Rand J. Econ. 23 (Autumn): 299–308.
Olley, G. Steven, and Ariel Pakes. 1996. “The Dynamics of Productivity in the
Telecommunications Equipment Industry.” Econometrica 64 (November):
1263–97.
Ordover, Janusz A., Garth Saloner, and Steven C. Salop. 1990. “Equilibrium
Vertical Foreclosure.” A.E.R. 80 (March): 127–42.
Posner, Richard A. 1976. Antitrust Law: An Economic Perspective. Chicago: Univ.
Chicago Press.
Rey, Patrick, and Jean Tirole. 2007. “A Primer on Foreclosure.” In Handbook of
Industrial Organization, vol. 3, edited by Mark Armstrong and Robert H. Porter.
Amsterdam: North-Holland, forthcoming.
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms
cementing relationships 301
Riordan, Michael H. 1998. “Anticompetitive Vertical Integration by a Dominant
Firm.” A.E.R. 88 (December): 1232–48.
Roberts, Mark J., and Dylan Supina. 1996. “Output Price, Markups, and Producer
Size.” European Econ. Rev. 40 (April): 909–21.
Rosengren, Eric S., and James W. Meehan Jr. 1994. “Empirical Evidence on
Vertical Foreclosure.” Econ. Inquiry 32 (April): 303–17.
Salinger, Michael A. 1988. “Vertical Mergers and Market Foreclosure.” Q.J.E. 103
(May): 345–56.
Shea, John. 1997. “Instrument Relevance in Multivariate Linear Models: A Simple
Measure.” Rev. Econ. and Statis. 79 (May): 348–52.
Snyder, Christopher M. 1995. “Empirical Studies of Vertical Foreclosure.” Indus.
Econ. Conf. Papers and Proc. (Univ. Melbourne and Bur. Indus. Econ.) 95/23:
98–127.
Syverson, Chad. 2004. “Market Structure and Productivity: A Concrete Example.”
J.P.E. 112 (December): 1181–1222.
———. Forthcoming. “Prices, Spatial Competition, and Heterogeneous Producers:
An Empirical Test.” J. Indus. Econ.
U.S. Bureau of Labor Statistics. 1983. Trends in Multifactor Productivity, 1948–81.
Bull. no. 2178. Washington, DC: Government Printing Office.
USGS (U.S. Geological Survey). 2004. Minerals Yearbook. Vol. 1. Metals and Minerals.
Washington, DC: U.S. Government Printing Office.
Vernon, John M., and Daniel A. Graham. 1971. “Profitability of Monopolization
by Vertical Integration.” J.P.E. 79 (July/August): 924–25.
Wall Street Journal. 1967. “FTC Announces Formal Merger Guidelines for Cement,
Food-Distribution Industries.” January 18.
Waterman, David, and Andrew A. Weiss. 1996. “The Effects of Vertical Integration
between Cable Television Systems and Pay Cable Networks.” J. Econometrics 72
(May): 357–95.
Williamson, Oliver E. 1971. “The Vertical Integration of Production: Market
Failure Considerations.” A.E.R. Papers and Proc. 61 (May): 112–23.
This content downloaded from
147.251.185.56 on Tue, 26 Apr 2022 13:23:31 UTC
All use subject to https://about.jstor.org/terms