Ch. 8: Conglomerate Firms and Internal Capital Markets 431
material by exceeding 10% of the firms’ consolidated totals.^9 The CIS database contains
information for such segments on net sales, earnings before interest and taxes (EBIT),
depreciation, capital expenditures, and assets, as well as the total number of reported
segments for the firm. This data is available for all active Compustat firms except utility
subsidiaries and is easy for most researchers to access.
There are, however, several well-known problems with CIS data. Firms self-report
segment data and changes in the number of reported segments may reflect changes in
reporting practice.Hyland (1997)finds that up to a quarter of reported changes in the
number of segments stem from changes in reporting policy, not changes in the level of
diversification.^10 The reporting requirement also only applies to segments that meet a
10% materiality condition. Thus, segments reported by large firms may be span several
industries.^11 Moreover, there is no presumption that a self-reported segment approxi-
mates a single industry. According to SFAS 14, a segment is distinguished by the fact
that its constituents “are engaged in providing a product service or a group of related
products and services... to unaffiliated customers”. Thus, segments may be vertically
integrated. The 4-digit SIC in which they are classified by CIS are assigned by COMPU-
STAT, not by the firms themselves. This last problem is quite severe: using Census data
Villalonga (2004a, 2004b)shows that in over 80% of cases the SIC code assigned by
COMPUSTAT is not the code of the segment’s largest industry. Taken together, these
problems raise the possibility that a substantial number of segments are misclassified
into 4-digit SIC codes and that a substantial number of firms that report only one seg-
ment in fact operate in related or vertically integrated industries.^12
Several researchers have used alternative data sources from the US Bureau of Census
which do not rely on data which is aggregated up to segment level by firms.Maksimovic
and Phillips (1998, 2001, 2002, 2007)andSchoar (2002)use the Longitudinal Research
Database (LRD), maintained by the Center for Economic Studies at the Bureau of the
Census.^13 The LRD database contains detailed plant-level data on the value of ship-
ments produced by each plant, investments broken down by equipment and buildings,
and the number of employees. The LRD tracks approximately 50,000 manufacturing
plants every year in the Annual Survey of Manufactures (ASM) from 1974 to 2003. The
ASM covers all plants with more than 250 employees. Smaller plants are randomly se-
lected every fifth year to complete a rotating five-year panel. Note that while the annual
(^9) Revised disclosure requirements, SFAS 131, superseded SFAS 14 in 1997. Most of the studies that use
Compustat data discussed in this review rely on pre-1997 data. Under SFAS 131 firms do not have to report
line of business data unless they are organized that way for performance evaluation (Berger and Hahn, 2003).
(^10) See alsoDenis, Denis and Sarin (1997), Pacter (1993)andHayes and Lundholm (1996).
(^11) Villalonga (2004a, 2004b)notes that the maximum number of 4-digit segments belonging to a single firm
for her sample of firms drawn from the BITS database of the U.S. Bureau of Census is 133.
(^12) Note that the definition of relatedness according to SFAS 14 does not correspond to the SIC classification.
Thus, divisions from different 2-digit SIC codes may be related according to SFAS 14.
(^13) For a more detailed description of the Longitudinal Research Database (LRD) seeMcGuckin and Pascoe
(1988).