Ch. 8: Conglomerate Firms and Internal Capital Markets 453
other unknowns. Because the estimator provides estimates ofE(q ̃j^2 ), E(ε^2 j)andE(ζj^2 ),
Whited (2001)also obtains estimates of theR^2 of the first equation, that is the propor-
tion of the variation of capital investment explained by the true Tobin’sq,aswellasthe
R^2 of the second equation, the proportion of the variation ofp’s (the proxy for Tobin’s
q) variation explained by the trueq.
Whited (2001)reestimates equations(4) and (5)correcting for the possible error mea-
surement error in the estimates of Tobin’sq. She finds that the corrected estimate ofβ
in equation(4)when estimated over conglomerate segments is insignificantly differ-
ent from the estimate ofβfor single-segment firms. Thus, she finds that the previous
findings of inefficient investment by conglomerates segments may be due to measure-
ment error. She also finds that the corrected estimates ofφandδin equation(5)are
insignificantly different from zero, suggesting that the previous finding that the firm’s
internal capital market is at least partially inefficient might also have been caused by
measurement error.
While the formal tests inWhited (2001)are specific to the model she investigates,
they raise a serious concern about the use of segment Tobin’sqs derived from COM-
PUSTAT data in all studies of intra-firm investment efficiency.
Maksimovic and Phillips (2007)argue that previous studies of investment using Com-
pustat data are subject to another form of measurement error: They exclude a major
type of investment expenditure by conglomerates. MP show that single-segment and
conglomerate firms differ both in the level of total investment and thetypeof invest-
ment. The overall level of capital expenditures on existing plants by conglomerates and
single-segment firms in U.S. manufacturing industries is similar. However, conglomer-
ates and single-segment firms differ markedly in their rates of purchases of new plants,
even when controlling for segment size. Thus, the COMPUSTAT based studies which
use segment capital expenditures as a proxy for investment and do not include acqui-
sitions exclude a major category of investment by conglomerates. Using LRD data, for
each single-segment firm and conglomerate segment MP predictFDjthe probability
that the segment will be run a financial deficit if it invests at the level predicted by its
productivity and industry conditions not taking account whether it is a conglomerate
segment or not. They then run the regressions of the following form:^26
acqj(orij)=zjγ+FDjβ+δcong×FDj+φcongj+φTFPj+ζj,where acqjis a measure of segmentj’s acquisition activity,ijis a measure of segment
j’s capital expenditures, TFPjis the segment’s industry standardized productivity and
congj is a dummy that takes a value of 1 if the segment belongs to a conglomerate
and 0 otherwise.Maksimovic and Phillips (2007)finds thatβ<0, so that a predicted
financing deficit leads to a reduction of acquisition and capital expenditure. However,
(^26) The regressions inMaksimovic and Phillips (2007)allow for differences across types of industries, but
these differences suppressed in this exposition.