Ch. 8: Conglomerate Firms and Internal Capital Markets 435
econometric techniques in an attempt to control for the endogeneity of the diversifica-
tion decision—firm fixed effects, simultaneous-equation estimation using instrumental
variables and Heckman’s two-step procedure. Their data is from COMPUSTAT and
their sample and the measurement of excess value follow the earlier literature. Seg-
ments of multiple-segment firms are valued using median sales and asset multipliers of
single-segment firms in that industry. The imputed value of a segment is obtained by
multiplying segment sales (asset) with the median sales (asset) multiplier of all single-
segment firm-years in that SIC. The imputed value of the firm is the sum of the segment
values.
Campa and Kedia find a strong negative relation between a firm’s choice to be diver-
sified and its value. Firms that are diversified have a lower value than firms that do not.
However, once the endogeneity between the decision to be diversified and firm value is
taken into account, the diversification discount always drops, and sometimes turns into
apremium.
The statistical modeling of the endogeneity of conglomerate status, in turn, raises
questions about the nature of the decision to become conglomerate. In their statistical
specification, Campa and Kedia implicitly assume that the decision to remain diversified
is itself endogenous in each period. This is appropriate if the decision to diversify is
easily reversible. However, if the decision is costly to reverse, then it is natural to focus
attention on the endogeneity of the decision to diversify (as opposed to the endogeneity
of the decision to maintain conglomerate status), or more generally on changes in the
level of diversification.
Villalonga (2004a, 2004b)focuses on the decision to become diversified. Using a
Compustat for the years 1978–1997 she identifies 167 firm years in which single-
segment firms diversified. Her control sample consists of 40,757 single-segment firm
years. She adopts a two-stage procedure. In the first stage, she uses a probit model to
obtain the probability that a firm becomes diversified, which she terms the propensity to
diversify. For the probits Villalonga tries several specifications, including one that uses
the same explanatory variables asCampa and Kedia (2002).
In the second stage Villalonga controls for the estimated propensity to diversify in
determining whether becoming diversification destroys value. She uses two types of
matching estimators (the methods proposed byDehejia and Wahba, 1999, andAbadie
and Imbens, 2002) andHeckman’s (1979)correction for selection bias. As inCampa
and Kedia’s (2002)tests, Heckman’s method directly corrects for biases due to unob-
served characteristics of firms that choose to diversify. The matching estimators use the
estimate of the propensity to merge as one of the characteristics for finding matching
non-diversifying single-segment firms that are comparable to the diversifying single-
segment firms. Consistent withCampa and Kedia (2002), Villalonga finds that the
decision to diversify did not affect the value of the 167 firms that she identifies as having
diversified during her sample period.
Lamont and Polk (2002)adopt a different approach and a difference definition of
the extent of diversification in their study of the relation between diversification and
value. They argue that a key characteristic of an industry is the ratio of investment to