432 V. Maksimovic and G. Phillips
data is called the Annual Survey of Manufactures, reporting is not voluntary for large
plants and is not voluntary once a smaller firm is selected to participate in a rotating
panel. All data has to be reported to the Censuc Bureau by law and fines can be levied
for misreporting.
Annual Survey of Manufactures offers several advantages over Compustat: First, it
is comprehensive and covers both public and private firms in manufacturing industries.
Second, coverage is at the plant level, and output is assigned by plants at the four-digit
SIC code level. Thus, firms that produce under multiple SIC codes are not assigned
to just one industry. Third, plant-level coverage means that plants can be tracked even
when they change owners.
Villalonga (2004a, 2004b)uses the Business Information Tracking Series (BITS)
database, also from the Bureau of the Census. BITS provides data between 1989 and
1996 for all U.S. business establishments, private and public, in all some 50 million
establishment-year observations.^14 For each establishment, the BITS database contains
data on the number of employees, the payroll and on the identity and revenue of the firm
that owns it. Each establishment is assigned to a 4-digit SIC code.
Because the BITS database covers all sectors of the economy and is not limited to
the manufacturing sector like the LRD, it is more comprehensive. However, since the
available data for each establishment is limited, BITS cannot be used to determine an
establishment’s productivity.
Villalonga (2004a, 2004b)links the BITS dataset with COMPUSTAT, enabling her to
determine the composition of a Compustat firm without relying on SFAS 14 disclosures.
She then recomputes the conglomerate discounts of the COMPUSTAT firms that she
has linked, using as comparables those COMPUSTAT firms that BITS data identifies as
being single-segment firms.
The results are startling. Villalonga finds that diversified firms trade at a significant
premiumover single-segment firms, as so classified using BITS. When COMPUSTAT
segment data is used to classify firms, Villalonga obtains the standard conglomerate
discount obtained in the earlier literature.
Villalonga explores several possible explanations for this discrepancy. A fundamen-
tal difference between BITS and COMPUSTAT is that former treats vertical integration
as a form of corporate diversification, whereas the latter does not. However, when Vil-
lalonga reconstitutes BITS segments to group together vertically integrated businesses
and recomputes the discount she still obtains a conglomerate premium.
These results highlight the fact that COMPUSTAT segments are related by con-
struction, at least in the eyes of the firms. Thus, measures of diversification based on
COMPUSTAT data may implicitly be measures of unrelated diversification. It is thus
possible that diversification, measured by COMPUSTAT is a measure of inefficient di-
versification (hence the discount). Villalonga also raises the possibility that Compustat
segments are lumped together to avoid disclosing to competitors which segments are
most lucrative.
(^14) An establishment is a location where a firm conducts business, such as a plant, a store or a warehouse.