internal and external capital markets - insead · internal and external capital markets urs c....
TRANSCRIPT
Internal and External Capital Markets
Urs C. Peyer *
Department of Finance INSEAD
April 25, 2002
Abstract – This study tests the proposition that firms that make efficient use of their internal capital markets can lower the cost of transacting in the external capital market. Using a large panel data set of diversified firms from 1980–1998, I show that diversified firms with an efficient internal capital allocation display a higher propensity to use external capital relative to comparable single segment firms. This result is robust to including other controls, such as measures of information asymmetry, capital needs, relative valuation and firm size. Further, a higher use of external capital by diversified firms relative to single segment firms is associated with a higher excess value, but only for efficient internal capital market users. I also demonstrate the robustness of these findings by employing a sample of firms that experience an increase in expected investment outlays. My findings support predictions from theoretical models, such as Stein (1997), and are consistent with the interpretation that diversified firms with an efficient internal capital market benefit from lower-cost access to external capital by alleviating information asymmetry problems between managers and investors. For helpful discussions and comments, I would like to thank Anil Shivdasani, Jennifer Conrad, Claudio Loderer, Henri Servaes, Steve Slezak and Marc Zenner, Mike Cliff, Eitan Goldman, Maria Nondorf, David Ravenscraft, Jeffrey Wurgler and seminar participants at Arizona State University, Boston College, Darden, Emory, Illinois, INSEAD, London Business School, University of Maryland, University of Miami, University of North Carolina, University of Pittsburgh, University of Toronto and Virginia Tech, the 2001 Young Scholar Conference at the College of William and Mary, the 2001 WFA meetings and the 14th Australasian Finance and Banking Conference. An earlier version of this paper received the 2001 Trefftz Award. *Urs Peyer, Department of Finance, INSEAD, Boulevard de Constance, 77305 Fontainebleau, France. Tel +33 (0)1 6072 4178; Fax +33 (0)1 6072 4045; email: [email protected]
1
Internal and External Capital Markets
This paper examines the interaction between internal and external capital markets. For the
purpose of this paper, I define an internal capital market as the mechanism by which headquarters
allocates capital to the various divisions of the firm. If headquarters allocates investment to the
divisions with the highest marginal return, then the firm uses its internal capital market
efficiently. The primary question in this study is whether and how a firm’s internal allocation is
related to its transactions with the external capital market.
Answering this question can help us to better understand how firms finance their
investments. Specifically, are there differences between single segment and diversified firms in
the sources of financing? What characteristics of diversified firms lead to more or less use of
external capital? The answers are important in the light of theories that try to explain the benefits
and costs of diversification.1 A potential benefit of diversification is to establish an internal
capital market (ICM). Creating an ICM can have at least two advantages. First, internal resource
allocation can be more efficient than allocation performed by the external capital market. This
issue is the focus of recent theoretical and empirical research investigating whether diversified
firms use their ICMs to efficiently reallocate capital. Theoretical models and arguments
predicting an efficiency gain from internal capital allocation are found in Weston (1970),
Williamson (1970, 1986), Gertner et al. (1994) and Stein (1997). Alternative models based on
agency conflicts emphasizing the drawbacks of internal allocation are developed by Scharfstein
and Stein (2000) and Rajan et al. (2000). Empirical tests of these models by Lamont (1997), Shin
and Stulz (1998), Scharfstein (1998) and Rajan et al. (2000), among others, suggest that capital is
reallocated internally, but that, on average, the reallocation is inefficient. On the other hand,
Maksimovic and Phillips (2001) and Khanna and Tice (2001) conclude that internal capital
markets are working efficiently by reallocating capital away from low productivity to high
productivity factories or stores.
A second potential advantage of an ICM is its effect on transactions with external capital
markets (ECM). Stein (1997) theoretically analyzes the interaction between the efficiency of
internal capital allocation, the size of the ICM (number of divisions and their correlation in
investment opportunities), the use of external capital and firm value. With information
asymmetries and agency problems between managers and outside investors, firms can be
financially constrained. Potentially, information asymmetry problems can be reduced through an
ICM. Take an external investor who bases her decision about how much to lend to a firm on her
2
estimate of the firm’s value-maximizing investment needs. According to the law of large
numbers, the precision of the estimate of the optimal amount of capital increases with the number
of projects in the firm if the projects’ capital needs are imperfectly correlated. The same logic
applies to single segment firms. However, the ICM allows HQ to reallocate capital to the highest
marginal return divisions. Investment in single segment firms cannot be reallocated across
divisions, which exacerbates the under- and overinvestment problem. Thus, lending to a
headquarters that oversees a portfolio of projects with imperfectly correlated capital needs, i.e., a
diversified firm, is different from lending to a portfolio of single segment firms in that
information asymmetry problems are less important. Therefore, diversified firms that allocate
capital efficiently in the ICM and firms with larger ICMs, i.e., firms with more divisions and
lower correlation of divisional investment opportunities, should be able to use more external
capital. On the other hand, HQ of a more diversified firm might loose the ability to efficiently
reallocate capital because HQ itself becomes less informed about all the possible investment
opportunities. Thus the impact of size of the ICM on a firm’s ability to raise external capital
should differ by its monitoring technology, i.e., the efficiency of internal capital allocation.
Diversified firms that are able to alleviate some of the information asymmetry and agency
problems in transacting with the external capital market should have a higher value because they
underinvest less and thus can raise external capital at a lower cost than single segment firms.
However, inefficient ICM users should not be able to access more external capital because they
do not have or do not use superior inside information about their projects. Thus, an external
investor should not be willing to invest more capital in such firms, since headquarters does not
allocate the capital in a value-maximizing fashion.
The above arguments suggest interesting cross-sectional relationships between the efficiency
of the internal capital allocation, the size of the ICM, the use of external capital and firm value.
Analyzing these interactions will further our understanding of the potential costs and benefits of
an internal capital market and hence diversification.
I test the above arguments in two ways. First, I employ a panel of diversified Compustat
firms from 1980–1998. In this sample, diversified firms use, on average, less external capital than
comparable single segment firms. However, the analysis indicates a significantly positive
correlation between the efficiency of internal capital allocation, as proxied by the relative value
added by allocation (RVA) of Rajan et al. (2000), and a firm’s use of external capital. Moreover,
diversified firms with a larger ICM use more external capital only if their internal capital
allocation is efficient. Firms with a large ICM allocating capital inefficiently use significantly less
external capital. In addition, the analysis also suggests that firms that allocate capital more
3
efficiently can reduce the impact of information asymmetry problems (e.g., Myers and Majluf,
1984) when raising capital externally. Further, I find a significant relationship between a firm’s
ICM characteristics, its use of external capital and excess value measures (Berger and Ofek,
1995; Lang and Stulz, 1994). Consistent with Rajan et al. (2000), it emerges that firms with a
more efficient internal capital allocation display a higher excess value, and firms with a higher
diversity in their investment opportunities (a proxy for the size of an ICM and potential agency
conflicts) display a lower excess value. A new finding is that firms that use more external capital
have a lower excess value. The important exceptions are firms with an efficient internal capital
allocation and firms with both, an efficient internal capital allocation and a large ICM – these
firms are valued significantly higher if they use relatively more external capital. On average, such
firms even have a positive excess value. The inferences from the first part of the paper, using
panel data on all diversified firms, strongly support the predictions from Stein’s (1997) model that
firms with an efficient internal capital allocation and firms with larger, but still efficient ICMs can
raise more external capital and that doing so increases firm value.
I try to alleviate concerns of endogeneity and simultaneity using different econometric
techniques. However, causality is difficult to establish in this framework. Therefore, I employ a
second approach. Many theoretical models investigating the use of external capital rely on the
assumption that a new, positive-NPV project arrives unexpectedly and that the entrepreneurs’
wealth and/or the firm’s internal funds are insufficient to cover the investment (e.g., Myers
(1977), Myers and Majluf (1984), Li and Li (1996), Stein (1997)). In order to mimic more closely
the setting in which these models are specified, I select a sample of diversified firms that operate
in industries that receive a positive shock to investment opportunities, proxied by industry median
q. To ensure that the change in q does not merely reflect a surprise in current cash flow, I require
that the industry’s median cash flow remain constant. This setting provides a natural experiment
to investigate whether diversified firms that receive an unexpected valuable project use more or
less external financing than comparable single segment firms.2 Consistent with the findings of the
panel data study, diversified firms use more external capital if their internal capital allocation is
more efficient. Also, diversity has a positive effect only if the internal allocation is efficient;
otherwise diversity negatively affects a firm’s use of external capital. As expected from
arguments such as Myers and Majluf (1984), firms with more information asymmetries use less
external capital. This relation is alleviated only if firms internally allocate capital efficiently. In
this setting, firms that are able to raise more external capital should find a profitable investment
opportunity. Nevertheless, the use of external capital is only significantly positively related to
excess value for firms with an efficient internal capital allocation and those with both, a higher
4
diversity and a more efficient internal capital allocation. In addition, changes in capital
expenditures at the firm and segment level suggest that the external capital raised is used to
increase investment.
Taken together, the evidence from a sample of firms that experience an exogenous shock to
investment opportunities suggests that ICM characteristics are important determinants of a
diversified firm’s ability to capture new growth opportunities by allowing the firm to use more
external capital. These findings highlight an additional, related advantage for firms with an
efficient ICM, namely easier access to external capital.
Prior empirical research on the interaction between internal and external capital markets
includes Comment and Jarrell (1995), Billett and Mauer (2002), Hadlock, Ryngaert and Thomas
(2001), and Fee and Thomas (1999). The study most similar to mine is Comment and Jarrell
(1995), who test Williamson’s (1970, 1975, 1986) argument that firms with ICMs transact less in
the external capital market. They find that, on average, diversified firms raise less external capital
but return more to their outside investors, and they conclude that there is no clear evidence that
diversification leads to less reliance on external capital markets. However, basing the conclusion
purely on average comparisons between diversified and single segment firms is problematic in the
light of findings by Rajan et al. (2000) and Scharfstein (1998) who show that firms are on average
allocating capital ineffic iently and by Berger and Ofek (1995) who find a valuation discount for
the average diversified firm relative to single segment firms. I extend Comment and Jarrell’s
analysis in several ways. First, I investigate the effects of ICM characteristics such as size and
efficiency on a firm’s use of external capital. Second, I compare diversified firms to their single
segment peers. Third, I control for other factors that may influence the use of external capital.
Fourth, I relate the use of external capital to firm value.
Berger and Ofek (1995) find that diversified firms use more debt, but conclude that the
difference is economically insignificant. Hadlock et al. (2001) find a less negative announcement
return to equity offerings for diversified firms than for single segment firms. Fee and Thomas
(1999) show that diversified firms have lower measures of information asymmetry. They link
those measures directly to excess value and find a negative relationship. My findings suggest that
efficient ICM users can reduce the cost of information asymmetry. Therefore, besides the direct
effect on pricing, there should also be an indirect effect through the firm’s ability to raise more
external capital.
Billett and Mauer (2002) find that diversified firms can increase firm value if capital is
transferred to segments with above-industry-average return on assets that would be financially
constrained if the divisions were single segment firms. However, their study does not analyze
5
whether the transfers were made due to relaxed credit constraints at the firm level or whether free
cash flow from other divisions was reallocated.
The remainder of the paper is organized as follows. The next section briefly reviews
underlying theories for my tests. Section 2 describes the sample, the tests, and the results for the
panel data set. In Section 3, I describe tests and show results for the industry shock sample.
Conclusions follow.
1 Underlying Theories
In a world with perfect markets, it does not matter whether investment is funded by internal
or by external capital markets. However, the source of financing can matter in the presence of
informational asymmetry and agency problems. Stein (1997) considers a model where managers
have better information about their projects’ success than external investors and use this
information efficiently to allocate capital to the divisions with the highest marginal return.
Furthermore, he assumes that managers derive private benefits that increase with the resources
under their control and that their tendency to overinvest is costly, such that the external capital
market may impose credit rationing. Under these assumptions, Stein shows that diversified firms
can sometimes raise more external capital than single segment firms and that doing so increases
firm value. A numerical example can help to illustrate the reasoning.
Assume that there are two projects. These two projects can be owned individually by two
single segment firms or a diversified firm can own both projects. Managers know which of the
projects (if any) are going to be successful. External investors, however, only have an ex ante
expectation about the probabilities of each project’s success. Assume that the expected
probability of a good outcome is p = 0.5 and that of a bad outcome is (1 – p) = 0.5. Investment in
the project can be either 1 or 2 units. In the bad state, investing 1 unit in a project results in a
verifiable gross return (y1) of 1.1, and investing 2 units results in y2 = 1.9. Therefore, the optimal
level of investment in the bad state is 1 unit. In the good state, I assume that the project’s return is
scaled up by a factor, θ = 1.4, such that θ (y2 – y1) > 1, which implies that the optimal investment
per project is 2 units in the good state. Without a revelation scheme, investment cannot be made
state-contingent and external investors have no means of telling which projects are good or bad.
Thus, the question an external investor faces is whether to invest 1 or 2 units per project when
projects are organized as single segment firms. Investing 1 unit in a single project firm provides
an expected NPV of pθ y1 + (1 – p)y1 – 1 = 0.32. Investing 2 units in a single project firm results
in an expected NPV of pθ y2 + (1 – p)y2 – 2 = 0.28. Therefore, external investors optimally invest
6
only 1 unit in single segment firms. Information asymmetries and agency problems thus result in
external capital constraints.
If a diversified firm owns both projects, the external investor determines the optimal
investment by computing the expected NPV per project for 2, 3 or 4 units of total capital raised.
Under the assumption that outcomes are independent across projects, the value of having an ICM
can be easily computed. If the external investor invests 2 units, the expected NPV is the
probability weighted average of the projects’ returns. The assumption that headquarters allocates
funds to the project with the highest marginal return, i.e., ICM efficiency, is now important in
determining the expected NPV, which is 2(1 – p)2y1 + 2p2 θ y1 + 2p(1 – p)θ y2 –2 = 0.65. Per
project, the expected NPV is 0.325, which is larger than the 0.32 that could be expected from a
single segment firm realizing a project.3 If the external investor invests 3 units, the expected NPV
is (1 – p)2(y1 + y2) + p2 θ (y1 + y2) + 2p(1 – p)(y1 + θ y2) – 3 = 0.68. Per project, the expected NPV
has increased to 0.34. The additional increase in value is due to the fact that, on average, more
positive NPV projects can be realized and only in one instance (both projects in the bad state) is
there more overinvestment compared to the previous scenario. As long as the expected benefit
from realizing more positive NPV projects is bigger than the cost of overinvesting, a diversified
firm can relax credit constraints relative to single segment firms. If the diversified firm was
allowed to raise 4 units, however, the expected NPV would drop to 0.56, which results in an
expected NPV per project of only 0.28. In this case, no reallocation occurs because each project is
always investing 2 units and a diversified firm is not more valuable than two separate single -
project firms.
Thus far, the example has shown that a diversified firm with an efficient ICM can use more
external capital and increase firm value. The increase in firm value has two sources. First, the
ability to transfer funds to the highest marginal return project (winner picking) is valuable. The
expected NPV per project increases from 0.32 for a single segment firm to 0.325 for a diversified
firm with an efficient ICM purely by combining two projects. Second, combining two projects
under the supervision of one headquarters can result in lower costs of information asymmetries.
In the example, the expected NPV per project of 0.325 in a diversified firm with 2 units of
investment increases to an expected NPV per project of 0.34 if the diversified firm receives 3
units of investment. This increase in value reflects a reduction in the cost of information
asymmetry.4 Note that this benefit only exists if the firm is using its ICM efficiently. Only then
can the external investor benefit from headquarters’ superior information by delegating the
investment allocation decision to management. To see this, assume the CEO has a pet project in
which she always invests 2 units, regardless of the project’s outcome. If the firm could raise 3
7
units of investment, then the expected NPV of the firm would be 0.60. Per project that is an
expected NPV of 0.30, which is lower than for a single project firm getting only 1 unit of
investment per project. Hence, efficient internal allocation is an important characteristic of an
ICM with respect to transactions in the external capital markets.
Allowing headquarters to increase the number of projects under its control makes it even
easier for outside investors to invest in the diversified firm. Assume an extreme case in which a
diversified firm owns 100 projects and each project’s outcome is independent. According to the
law of large numbers, an external investor would now expect roughly 50 projects in the good state
and 50 in the bad state. She would be willing to invest almost at the first-best level of, on average,
150 units. Therefore, a firm with a larger ICM should be able to use more external capital.
However, this prediction is based on the assumption that HQ can monitor many divisions without
a decrease in the quality of allocational efficiency. It also leads to the counterfactual prediction
that one huge firm could maximize value by making information asymmetry issues unimportant.
Extending the model, Stein (1997) shows that if monitoring becomes harder the more divisions a
firm accumulates, then allocational efficiency decreases with the size of the ICM. This suggests
that the size of an ICM might be non-linearly related to a firm’s ability to use more external
capital. Combining the two characteristics of an ICM, size and allocational efficiency, the model
predicts that firms with an efficient internal capital allocation and a large ICM use the most
external capital. Firms with a large ICM but inefficient allocation should use the least. Similarly,
the use of external capital should have a positive effect on firm value if the firm has an efficient
internal capital allocation or both an efficient internal capital allocation and a large ICM. A
negative relation between the use of external capital and size of the ICM is predicted if the
internal allocation is inefficient.
While this example is clearly a simplified version of investment allocation, it serves to
highlight that firms with efficient and larger ICMs (more divisions, and divisions with less-
correlated outcomes) should be able to relax some of the credit constraints otherwise imposed on
single segment firms and reduce the cost of information asymmetries. It also shows that the per-
project value of the diversified firm that raises more external capital should be higher than that of
both a diversified firm that does not raise more external capital and a single project firm.
Other papers, such as Stulz (1990), Froot et al. (1993) and Li and Li (1996) argue that if
diversification reduces cash flow volatility, the likelihood of over- and underinvestment is
reduced, and cash flows are more certain to cover the existing debt. Therefore, newly raised
external funds are less likely to be used to pay existing debt. This implies that a diversified firm
8
that has to finance a new, positive-NPV project with external capital before the existing debt is
due is more likely to receive external financing than is a single segment firm for a similar project.
Fluck and Lynch (1999) show that firms acquire marginally profitable single segment firms
that, because of agency problems, cannot find external financing as stand-alone firms. Within a
diversified firm, however, the conglomerate can raise funds sufficient to finance the marginally
profitable segment. Thus, diversified firms should be able to raise more external financing than
comparable single segment firms, and this should be value enhancing, even though diversified
firms might trade at a discount relative to their industry median peers.
Matsusaka and Nanda (2001) model a firm’s need to raise external capital for different levels
of internal resources. They assume a fixed deadweight cost of external capital, independent of
whether a diversified or single segment firm raises capital. In their model, an ICM is valuable
because it allows the diversified firm to avoid external financing in more instances than single
segment firms. However, there are cases where internal capital is insufficient and diversified
firms raise more external capital than comparable single segment firms, and doing so is valuable.
Matsusaka and Nanda conclude that efficient ICM firms do not necessarily access external capital
markets less often. Their analysis, however, holds properties of the internal capital market
constant, and does not address interactions between ICM and ECM for different organizational
forms. Important for this study is their finding that the level of internal capital available is a
significant determinant of external capital use.
In summary, the tests focus on the following predictions: (i) Firms with an efficient internal
capital allocation should be able to use more external capital. (ii) While more diversified firms
will face more difficulty in allocating capital efficiently, those that are efficient should have a
positive correlation between the size of the ICM and the use of external capital. (iii) Information
asymmetry problems will drive a wedge between the cost of internal and external capital, as in
Myers and Majluf (1984). These costs can be reduced if internal capital allocation is performed
efficiently. Therefore, measures of information asymmetry should be negatively correlated with a
firm’s use of external capital, although, less so for firms with an efficient internal capital
allocation, i.e., firms with better monitoring.
With respect to firm value, the model predicts: (iv) a positive correlation between efficiency
of internal capital allocation and firm value5 (v) a positive relation between the use of external
capital and value for efficient allocators, and for large diversified firms with an efficient internal
capital allocation.
9
2 Panel Data Sample
2.1 Sample Selection
I use all firms listed on Compustat’s industry segment files (including research files) for
1980–1998. Firms with incomplete segment information on sales, assets or capital expenditures
are dropped, as are firms with segments in the one-digit SIC codes of 0, 6 or 9.6 Firms with sales
less than $10 million are also excluded.7 Following Berger and Ofek (1995), I require the sum of
the segment sales to be within 1% of the net sales for the firm and the sum of the segment assets
to be within 25% of the firm assets. I apply a multiple to the remaining segment assets, such that
the sum of the recomputed segment assets adds up to total assets. I further restrict the sample to
firms with complete information on market value of equity and cash flow statement items.
Diversified firms are also dropped from the sample if imputed values for the segments are
missing. Imputed values are computed at the 3-digit SIC code level using only single segment
firms (at least five) with available data to compute the industry median.8 Additionally, firms are
excluded if their one year lagged value(s) of the variables described below is (are) missing. In
effect, this limits my sample to firms that survive any two-year period and have complete data
available in both years. Furthermore, to reduce endogeneity concerns, I use lagged values as a
proxy for ICM efficiency; therefore, I require that the lagged number of segments be at least two.9
Finally, I require that the firms have daily stock returns available on CRSP for at least 30 days in
the previous year in order to compute return volatilities.10 Imposing all of the data requirements
results in a sample of 8,538 diversified firm-years spread over the period 1981–1998. Over the
same time period, there are 34,065 single segment firms that pass the same screening process.
The number of diversified firms is fairly evenly distributed over time (not shown). There are
4,983 firm-years with 2 segments, 2,341 with 3 segments, 903 with 4 segments and 312 with 5 or
more segments.
2.2 Determinants of the Use of External Capital
According to Stein (1997), the key drivers of a firm’s use of external capital are the
efficiency of the internal capital allocation, the size of the ICM, and the degree of information
asymmetry. In addition, use of external capital can be affected by a firm’s need for capital and its
relative valuation.11 I estimate the following cross-sectional regressions for firm i and year t:
10
. value)(relative need) (capital
)efficiency ICM asymmetry on (informati asymmetry)n informatio(
)efficiency ICM size (ICM )efficiency ICM size (ICM
)efficiency (ICMsize) (ICM size) (ICM
capital externalnet Excess
1 ,9 ,8
1 ,71 ,6
21,51 ,4
1 ,32
1,21 ,1
,
−
−−
−−
−−−
++
×++
+×+×+
+++
=
titi
titi
titi
tititi
ti
γγ
γγ
γγ
γγγα
(1)
The subsequent sections describe the proxies used for the above variables and their
univariate statistics. Detailed definitions for all variables are given in Appendix 1.
2.2.1 Dependent Variable: Use of External Capital
I compute a measure of net external capital raised by diversified firms in excess of that
raised by single segment firms as follows. First, I compute net external capital as the proceeds
from the sale of debt, common and preferred stock minus the amount of debt retired and common
and preferred stock repurchased. To make the use of external capital comparable between
diversified and single segment firms, I compute an imputed value of net external capital based on
the median of single segment firms in the same 3-digit SIC code as the divisions of the diversified
firm. I match the median ratio of net external capital to sales to the individual segments of the
diversified firms according to year and industry. Then I multiply each segment’s imputed ratio by
segment sales and add up all of the imputed net external capital values to form a firm-level
imputed net external capital amount. I call the final measure Excess Net External Capital (EEC)
and compute it as follows:
assets of book value Laggedcapital externalnet Imputed capital externalNet
−
=capital external Excess net
A positive EEC implies that a diversified firm raises more net external capital than do
comparable single segment firms in its industries.
Table 1 shows that the median and mean EEC for diversified firms are significantly below
the median and mean EEC for single segment firms. The median for diversified firms is
–0.00549 and is significantly different from zero, which is the median of single segment firms, by
construction. The mean for diversified firms is 0.02966 and is significantly different from zero at
the 1% level, but this average is still significantly below the single segment mean, which is
0.04442. Consistent with Williamson (1975) these numbers suggest that diversified firms use
external capital markets less extensively than single segment firms. However, the use of external
capital should depend on a firm’s ICM characteristics. Thus a multivariate analysis is needed.
Further, the univariate test statistics should be interpreted with caution because the panel data
observations are not independent.
11
2.2.2 Measures of the Size of the ICM
The measure of size of the ICM has three basic aspects. The first is the number of different
operations or divisions; the second is the correlation of investment opportunities between these
divisions; the third is the size differences between the divisions. A measure of ICM size that
encompasses all three aspects is diversity, used by Rajan et al. (2000). Diversity is defined as the
standard deviation of the segment asset-weighted imputed q divided by the equally weighted
average imputed segment q. There is a significantly positive median for diversity of 0.286 (Table
1). A higher value of the variable can indicate more divisions, less dependence in the segments’
investment opportunities, and/or segments of more equal size. Thus, according to Stein (1997), a
higher value of diversity should be positively related to EEC if and only if a firm is internally
allocating capital efficiently.
Alternative measures that capture individual aspects of ICM size are the number of business
segments a firm reports or the inverse of the Herfindahl Index. For more detailed definitions see
Appendix 1.
2.2.3 Measures of ICM Efficiency
Stein (1997) shows that it is critical that headquarters be good at distinguishing between
good and bad projects and that the internal allocation be efficient. An ICM is considered efficient
if investment is allocated to the projects/segments with the highest marginal return.
I use two proxies to measure the allocational efficiency of the ICM. The first measure is the
relative value added by allocation (RVA) introduced by Rajan et al. (2000). I compute RVA as
follows:
( ) 11
−−−−= ∑∑
==
n
jss
ss
j
jjss
ss
j
jn
jj
j
j
j
j
j
BA
Capex
BA
Capexw
BA
Capex
BA
Capexqq
BA
BARVA ,
where BA is book value of assets of the firm, BAj is the book value of assets of segment j,
Capex is the firm’s capital expenditures, ss
j
ss
jBACapex is the asset-weighted average ratio of
single segment firms in the same industry as the segment of the diversified firm, wj is the ratio of
segment j assets to firm assets, qj is the asset-weighted Tobin’s q of single segment firms
operating in the same three-digit SIC industry as segment j, and q is the segment sales-weighted
qjs of the firm. BA, qj and wj are beginning-of-the-period values.
The expression ss
ss
j
j
j
j
BA
Capex
BA
Capex− is a proxy for transfers made between segments of a
diversified firm. It compares the segment’s investment ratio to the asset-weighted average
12
investment ratio of single segment firms in the same industry. The latter serves as a proxy for
what a segment’s investment ratio would have been were it a stand-alone entity.
∑=
−
n
jssj
ssj
j
jj BA
Capex
BA
Capexw
1 is a proxy for the overall funds available to a diversified firm relative
to its single segment peers. This term is subtracted from the industry-adjusted investment ratio to
correct for potential differences in availability of total capital that should not count as transfers.
( )qq j − identifies segments within a firm that have better-than-firm-average investment
opportunities. Thus, a firm with an efficient ICM should have a positive RVA because it transfers
capital to segments with better-than-firm-average investment opportunities and invests more than
single segment peers do in those segments.12
A second measure of the efficiency of internal allocation, used by Peyer and Shivdasani
(2001), is q-sensitivity of investment. q-sensitivity is defined as follows:
SalesFirmSalesFirm
CapexFirm
SalesCapex
qqSalesn
j jjj∑
=
−
×−×1
)( ,
where qj is the imputed Tobin’s q of segment j and q is the segment sales-weighted qjs of
the firm. Capex is the capital expenditures of the segment, and Firm Capex is the capital
expenditures of the firm. This measure is positive if a segment with a q above the firm’s average
q has an above-firm-average investment ratio (capital expenditures/sales) and a segment with
below-average q has a below-firm-average investment ratio. Therefore, q-sensitivity indicates
whether headquarters has invested relatively more in the high-q segments of the firm and
relatively less in the low-q segments based on the firm’s available resources.
A third measure is based on Maksimovic and Phillips (2001) and Schoar’s (2001) analysis of
the effect of differences in segment productivity. These papers suggest that a firm is efficiently
allocating capital if more investment is allocated to divisions with above average productivity. As
a proxy for segment productivity, Maksimovic and Phillips (2001) show that segment cash flow
can be used. I construct a measure called cash flow-sensitivity as in Peyer and Shivdasani (2001),
where the expression ( )qq j − in q-sensitivity is replaced with ( )cfcf j − . cfj is the cash flow to
sales ratio of segment j and cf is the average cash flow to sales ratio of the firm. This measure
also serves as a robustness check because it does not rely on imputed values but rather on
individual segment level information. 13
The measures of ICM efficiency use capital expenditures to proxy for segment investment.
Because the amount of capital expenditures is, in part, determined by a firm’s use of external
13
capital, the proxies for ICM efficiency are potentially simultaneously determined with my proxy
for a diversified firm’s use of external capital. To alleviate this problem, I use lagged values of
the measures of ICM efficiency as instruments.
Table 1 shows univariate statistics RVA, q-sensitivity and cf-sensitivity. RVA (cf-sensitivity)
has a median of –0.000005 (0) and a mean of –0.0004 (–0.00008) that is significantly negative at
the 10% level.14 Q-sensitivity has a median of 0 and a significantly positive mean of 0.001. For
single segment firms, these measures are always zero by definition.
Also interesting are the univaria te statistics in Panel B of Table 1. Mean and median EEC are
reported for firms stratified by RVA and diversity. Consistent with Stein’s (1997) predictions,
diversified firms with positive RVA and a measure of diversity larger than the sample median
diversity display the highest EEC. Firms with negative RVA and a measure of diversity larger
than the median display the lowest EEC indicating that ICM size can have very different effects
on a firm’s use of external capital depending on the efficiency of internal capital allocation.
2.2.4 Measures of Information Asymmetry
I use several measures for the degree of information asymmetry. First, I use the lagged ratio
of intangibles to total assets, expecting it to be negatively related to EEC. The advantage of this
ratio is that it is not affected by prices set in the external capital market, i.e., using this proxy, it
should be possible to identify the degree of information asymmetry that exists between managers
and outside investors.15
A second set of proxies is based on prices. Following Dierkens (1991) and Fee and Thomas
(1999), I compute residual variance and total variance of the daily stock returns over a calendar
year prior to the fiscal year-end. I use a market model to extract daily residual returns and
compute the variance over all of the available daily residual returns. The CRSP value-weighted
index, including dividends, is used as a proxy for the market return. As shown in Table 1, the
median daily residual variance (total variance) for diversified firms is 0.00057 (0.00063) and the
median for single segment firms is 0.00094 (0.00101).
As a third set of proxies, I use IBES analysts’ forecasts about a firm’s earnings per share. I
construct a standardized measure of analysts’ forecast dispersion using the standard deviation of
the one-year-ahead forecast of earnings per share standardized by the absolute value of the
average forecast. A higher value of this measure is expected to indicate greater information
asymmetry because it reflects a wider range of forecasts about the future earnings of a company.
The standardized analysts’ forecast dispersion could be computed for only 4,021 firm-years. For
3,370 firm-years, IBES information is missing. Another 1,147 observations are lost because only
one analyst’s forecast is available, and no standard deviation can be computed.
14
To test the argument made in section 1, that a diversified firm with an efficient internal
capital allocation can alleviate the effect of information asymmetry on the firm’s use of external
capital, I compute the following interaction variable. I create a dummy variable (RVADUMt–1)
that is equal to one if RVA at the beginning of the year (t-1) is greater than or equal to zero and
interact it with a proxy for information asymmetry.
2.2.5 Measures of Capital Need
Stein’s (1997) theory is based on the assumption that an entrepreneur has to raise external
capital for his projects because the financing needs exceed his personal wealth. Internally
generated cash flow from previous years is exogenous to the model. Matsusaka and Nanda (2001)
show that higher levels of internal capital reduce the need for costly external capital. As a proxy
for internal capital available to the firm, I compute excess internal cash flow. Internal cash flow is
defined as net cash flow from operations minus dividends. Excess internal cash flow is computed
in a similar way as EEC.16 I expect a firm with more excess internal cash flow to cover more of
its capital needs with internal capital. Hence a negative relation between excess internal cash flow
and EEC is expected, holding everything else constant. Table 1 shows that the median of excess
internal cash flow is significantly positive for diversified firms. Further, both median and mean
excess internal cash flow are higher for diversified firms than for single segment firms.
The need for capital is also determined by the available growth opportunities. As a proxy for
growth opportunities, I use the firm’s Tobin’s q at the beginning of the period. I expect firms with
a higher q to be in greater need of capital, holding everything else constant. One complication in
using q is that it might also be a proxy for information asymmetry. If so, one would expect a
negative relation between q and EEC.
2.2.6 Measures of Relative Valuation
Myers & Majluf (1984) show that firms that are overvalued are more likely to issue risky
new securities. Findings by Lucas and McDonald (1990), Asquith and Mullins (1986), Mikkelson
and Partch (1986), and Jung et al. (1996) confirm that firms are more likely to issue new
securities when their relative valuation is high. I use the stock return over the prior fiscal year and
lagged excess value as proxies for relative valuation. I follow Berger and Ofek (1995), and define
excess value as follows:
Excess value =
)(
logVIV , and ( )[ ]MSi
ni i SalesVMSalesVI /)( 1 ×= ∑ = ,
where V is the sum of market value of equity and book value of assets less the book value of
equity and deferred taxes, I(V) is the imputed firm value, Salesi is segment i’s sales, Mi(V/Sales)MS
15
is the sales multiplier (calculated as the median of the single segment firms in the same 3-digit
SIC code industry), and n is the number of segments per firm. An alternative way to compute
excess value is developed by Lang and Stulz (1994). They compute excess value as the difference
between Tobin’s q of the diversified firm and the segment asset weighted average of imputed
segment qs. Their imputed q is the average of the single segment firms’ qs. I compute the log of
the ratio of the firm’s Tobin’s q to the sum of segment sales-weighted imputed qs. The imputed
qs are median qs of single segment firms in the same 3-digit SIC code industry. A positive
relation to EEC is expected if higher stock returns and excess values indicate higher relative
valuations.
2.3 Results
Table 2 reports the regression results of equation (1) with EEC as the dependent variable.
Since the predictions relate to cross-sectional differences, Table 2 reports time-series averages of
coefficients of cross-sectional regressions run year-by-year. The reported t–statistics are based on
the time-series variation in the coefficients. This procedure is similar to that of Fama and
MacBeth (1973) and is also used in Rajan et al. (2000).17
Models 1–4 show results using different measures of the size of the ICM . In model 1 the
coefficient on diversity is significantly negative (–0.003). However, the interaction term with
allocational efficiency, RVADUM, is significantly positive (0.005). This supports the predictions,
and is consistent with the univariate statistics in Panel B of Table 1. The results still hold even
after introducing diversity squared as shown in model 2. Only firms with an efficient internal
capital allocation display a significantly positive relation with EEC. Similar inferences can be
drawn from the other two proxies of ICM size, the number of segments and the inverse of the
Herfindahl Index. For example, model 3 shows that the coefficient on the number of segments is
significantly negative (–0.026) but the interaction with RVADUM is significantly positive (0.040).
Increasing the number of segments from one to two, i.e. diversifying, decreases EEC by 0.026,
unless the firm has a positive measure of RVA, in which case, EEC increases by an overall 0.014
(0.040–0.026). Given the average difference in EEC between single segment and diversified firms
of 0.015, it seems that a change in the number of segments has an economically significant
impact on a firm’s use of external capital.
The coefficients on the proxies for ICM efficiency are always positive and significant at the
5% level. In model 1, the coefficient on RVA is 0.879. The point estimate suggests that a one-
standard deviation increase (0.0195) in RVA increases EEC by 0.017. Again, this corresponds to
about the average difference in EEC between single segment and diversified firms. Model 5 uses
16
q-sensitivity instead, and finds a significant positive coefficient. Firms with a more efficient
internal capital allocation use more external capital, consistent with Stein (1997). This conclusion
is also supported by model 6 where the coefficient on cf-sensitivity is significantly positive
(0.019). Thus, firms that allocate more investment to divisions with above firm average
productivity use more external capital.18
The measures of information asymmetry are expected to be negatively related to EEC. Of
specific interest for the tests in this paper is whether diversified firms with a more efficient ICM
display a less negative sensitivity to information asymmetry. Model 1 reports that the coefficient
on the lagged ratio of intangibles to total assets (–0.176), and the coefficient on residual variance
(–6.544) are significantly negative supporting Myers and Majluf (1984). Moreover, the
coefficient on the interaction variable between the lagged ratio of intangibles to total assets and
RVADUMt–1 is significantly positive (0.159). None of the implications change if total variance is
used instead of residual variance, as shown in model 7. Model 8 shows that the standardized
analysts’ forecast dispersion is significantly negatively related to EEC, with a coefficient of
–0.009.19 The coefficient on the interaction variable between the standardized analysts’ forecast
dispersion and RVADUMt–1 is significantly positive (0.003). The finding that the interaction
variables display a positive correlation with EEC is consistent with the notion that firms with an
efficient internal capital allocation can overcome some of the information asymmetry problems in
transacting with the external capital markets.
The proxies for need for capital are excess internal capital and beginning-of-the-year
Tobin’s q. Excess internal capital is significantly negatively related to EEC in every model. The
coefficient of –0.725 in model 1 suggests that a firm that has one dollar more internal capital than
a comparable single segment firm will use about 72.5 cents less external capital than its single
segment peers. Note that this coefficient is also significantly different from one, thus further
supporting the notion that market frictions make internal and external capital imperfect
substitutes. Tobin’s q is significantly positively related to EEC.
The coefficients on the measures of relative valuation are significantly positive in all the
models (in three cases, excess value is significant only at the 10% level). In model 1, the lagged
annual stock return has a coefficient of 0.037, and is significant at the 1% level. The coefficient
on lagged excess value is 0.009, and significant at the 5% level. Model 9 shows that the
coefficient on lagged excess value computed according to Lang and Stulz (1994) is also
significantly positive (0.037). These findings are consistent with the interpretation that firms are
more likely to issue new securities when their relative valuation is high.
17
Overall, the regression results show that firms with more divisions and with more
independent divisions use more external capital relative to their single segment peers only if the
firm is allocating capital efficiently in its ICM. Also, such firms can alleviate the impact of
information asymmetry problems when accessing external capital markets and use more external
capital. These findings suggest that the use of external capital significantly depends on the
characteristics of the ICM.
2.4 Robustness
In this section I examine the robustness of the findings presented in Table 2 by investigating
issues of supply and demand of external capital, including single segment firms in the analysis,
employing a different econometric method and by using different definitions of the dependent
variable. The exact definitions of the alternative dependent variables are given in Appendix 1.
Table 1 reports their univariate statistics.
2.4.1 Supply and Demand of External Capital
In essence, equation (1) is a reduced form of a supply equation and a demand equation for
external capital. Thus, a priori, it is not clear whether the coefficients on ICM size and efficiency
reflect supply side effects, as Stein’s (1997) model would imply. I perform the following test to
address the question of whether changes in the ICM characteristics affect the supply of external
capital holding demand constant.
I select a sample of diversified firms where the demand for external capital is held constant
but ICM characteristics are allowed to vary. Thus, holding demand constant, the coefficients
should reflect the effects of changes in the supply of external capital. This sample consists of
firms where the change in Tobin’s q, as a proxy for future investment opportunities, and the
change in internal cash flow between two consecutive years is within plus or minus 5%. 330 firms
pass this screen. Table 3 reports OLS regression results testing the following equation:
ies)opportunit investment( flow)cash internal(
)efficiency ICM asymmetry n informatio( asymmetry)n informatio()efficiency ICM size ICM(
)efficiency ICM( size) ICM(
76
5
43
21
∆+∆
+∆×∆+∆+∆×∆
+∆+∆+=∆
ββ
βββ
ββαEEC
(1')
In Table 3, all three models display significantly positive coefficients on the change in
diversity (0.035 in model 1) and the change in RVA (0.439 in model 1). Further, the coefficient
on the change in residual variance is marginally significantly negative (–1.466 in model 1)
indicating that firms with an increase in information asymmetry experience a reduction in the
supply of external capital unless they improve their allocational efficiency (coefficient of 1.122
18
on the interaction variable between the change in residual variance and a dummy variable equal to
one if RVA has increased). Models 2 and 3 additionally control for the changes in the demand for
external capital. With the exception of the coefficient on annual return (0.069), no demand side
coefficient is significantly different from zero. Furthermore, the increase in R-squared from
adding all the controls for the demand of external capital is only a marginal 0.018 from 0.182
(model 1) to 0.200 (model 2). In summary, the results reported here are at least not contradicting
the notion that the market’s supply of external capital is dependent on ICM characteristics.
2.4.2 Including Single Segment Firms
Even though I have defined an ICM as the mechanism by which HQ can allocate capital to
the different divisions, one could also argue that a single segment firm’s management allocates
capital in much the same way but just to different projects within the firm. Thus, the question
arises whether single segment firms are really different from diversified firms in their ability to
reduce the impact of information asymmetry on the use of external capital based on their
allocational efficiency.
The main measure employed thus far to proxy for the efficiency of internal capital
allocation, RVA, is zero by definition for all single segment firms. To allow for cross-sectional
variation among focused firms I employ a measure called the absolute value added by allocation
(AVA is defined in Appendix 1) used by Rajan et al. (2000).
Table 4 shows the results. As indicated by the coefficient on the multi-segment dummy of
–0.011, diversified firms, on average, use less external capital than comparable single segment
firms. This is consistent with the findings in the univariate analysis.
AVA is significantly positively related to EEC with a coefficient of 0.559. However, the
interaction variable between AVA and the multi-segment dummy is not significantly different
from zero.20 More interestingly, for single segment firms with a positive AVA there is no
significant reduction in the impact of information asymmetry on the use of external capital. This
is in stark contrast to the positive and significant coefficient on the interaction variable between
AVADUM, the multi-segment dummy and the measure of information asymmetry.
The results suggest that there are differences in the effect of internal capital allocation
between single segment and diversified firms.21 Only in diversified firms do we observe a
significant reduction in the impact of information asymmetry. The finding that AVA is positively
related to EEC is probably less surprising because firms that invest more than the median single
segment firm in the industry probably also use more external capital to finance their investment
than the median firm. Such an almost mechanical relation is not present in the RVA measure, but
it is precisely the reason why RVA is zero for all single segment firms. Thus, it is all the more
19
surprising that single segment firms cannot relax information asymmetry problems in the same
way that diversified firms can.22 Having shown that adding single segment firms has no effect on
the inferences drawn solely based upon the cross-sectional analysis of diversified firms, I
concentrate on diversified firms only.
2.4.3 Econometric Methodology
As an alternative to reporting time-series averaged coefficients from year-by-year cross-
sections, model 1 of Table 5 shows results using firm fixed-effects regressions. The coefficients
and their significance levels are very similar to those reported in Table 2. None of the above
conclusions are affected. However, the coefficients on Tobin’s q and excess value are now
insignificant, and the coefficient on residual variance decreases (significant at the 10% level).
2.4.4 Excess Net External Capital with Dividends and Interest
EEC, as defined thus far, does not consider dividends and interest payments as a decrease in
external capital. In model 2 of Table 5, the regressions are re-estimated using EEC including
interest and dividends as the dependent variable. None of the coefficients are significantly
different from the base case in model 2 of Table 2. Note that in this model, the definition of
excess internal capital is altered to include interest and dividend payments.23
2.4.5 Excess Net External Capital with Asset Sales
EEC does not consider proceeds from asset sales as an increase in external capital. There are
several reasons for this. First, determinants of asset sales are likely to be quite different from
determinants of new debt and equity issues. For example, Shleifer and Vishny (1992) show that
selling assets in a depressed industry can lead to relatively low sales prices because asset markets
become very illiquid. Schlingemann et al. (2001) find evidence that asset market liquidity is an
important determinant of which division is sold. In addition, Gertner et al. (1994) demonstrate
that diversified firms, especially those with efficient internal capital allocation, can redeploy
poorly performing assets internally and therefore reduce their transactions in the asset market.
Thus, a firm that allocates capital efficiently is expected to raise more external capital in the
financial markets but raise less capital by transacting in the asset markets. I use the sum of the
Compustat items ‘sale of property, plant and equipment’ and ‘sale of investment’ as a proxy for
asset sales, add it to net external capital raised, re-compute EEC and show the regression results
in model 3 of Table 5. The main difference from the regression using the base definition of EEC
is that the coefficient on RVA and its significance decrease. Thus, the asset market appears to be
used differently by firms with an efficient internal capital allocation. However, a complete
evaluation of these differences is beyond the scope of this paper.
20
2.4.6 Equity versus Debt Transactions
The measure of the use of external capital does not differentiate between transactions in the
debt and equity markets. According to Myers and Majluf (1984), however, one would expect that
information asymmetry problems have a more significant influence on raising equity than debt.
Thus, if allocational efficiency can reduce the cost of information asymmetry, the coefficient on
the interaction variable between information asymmetry and RVADUM should be especially
significant in a regression with new equity issued as the dependent variable. However, as long as
debt is also risky, information asymmetry will also affect a firm’s use of debt.
Excess increase in equity and excess increase in debt are used as dependent variables in
models 4 and 5 of Table 5 to study the effect of the size of the ICM and the allocational
efficiency. RVA is an important determinant of a firm’s use of equity as well as debt. Both
sources of external financing are also significantly negatively affected by information asymmetry
problems, and information asymmetry costs are reduced for firms with an efficient internal capital
allocation raising equity and debt. However, the coefficients are larger and more significant in the
equity regression than in the debt regression, indicating that the importance of allocational
efficiency is higher in the equity market than in the debt market.
Taken together, the robustness tests support the results shown in Table 2. Firms with an
efficient internal capital allocation and larger ICMs are able to alleviate some of the credit
constraints faced by single segment firms in transacting with the external capital markets.
2.5 Excess Value and Excess Net External Capital
In the previous section, I have found that ICM characteristics, such as the size and efficiency
of capital allocation are significant determinants of firm’s use of external capital. The question
now is whether the use of external capital has any impact on firm valuation – and, if so, what is
the direction? As demonstrated in section 1, Stein (1997) predicts a positive correlation between a
firm’s use of external capital and value for diversified firms with an efficient internal capital
allocation and firms with a large and efficient ICM.
I use excess value as a proxy for firm value relative to comparable single segment firms.24
The median (mean) excess values for diversified firms are reported in Table 1 as –14.87%
(–16.34%) using Berger and Ofek’s sales multiplier method and –15.54% (–13.37%) using Lang
and Stulz’s method. These means and medians are all significantly negative at the 1% level.25
Table 1, Panel C shows univariate statistics for excess value for firms stratified by RVA, EEC
and diversity. Interestingly, firms with positive RVA, high diversity and high EEC are firms that
display a positive median excess value of 3%. Firms with negative RVA, high diversity and high
21
EEC have a negative median excess value of 33%. Given the criticism by Graham et al. (2002),
Campa and Kedia (2000) and others about the validity of inferring value destruction due to
diversification from average excess value measures, the following tests exploit cross-sectional
differences rather than absolute levels.
The tests are based on the following model using firm i and year t:
1 ,81 ,7
1 ,6
1 ,51 ,4
1,31 ,21 ,1
,
value)excess(size) ICM of measure EEC(
size) ICM of measure EEC(
) size ICM of measure(size) ICM of measure(
) EEC(RVA)( EEC)(
valueExcess
−−
−
−−
−−−
+××+
×+
×++
×++++
=
titi
ti
titi
titititi
ti
RVADUM
RVADUM
RVADUM
γγ
γ
γγ
γγγβα
(2)
where iα is the firm-fixed effects, tβ is the year-fixed effects, EEC is the excess net external
capital and RVA is the relative value added by allocation. Beginning-of-the-period values of the
independent variables are used as instruments for the contemporaneous values to alleviate
endogeneity and simultaneity issues. Section 3 below presents results using an exogenous event
to further investigate issues of causality that clearly arise from equations (1) and (2).
Lamont and Polk (2001) find that the change in excess value is negatively related to the
lagged level of excess value indicating mean reversion in excess value. To control for this, I
include lagged excess value as an additional independent variable.
Table 6 reports firm fixed-effects regressions of equation (2). I use firm fixed-effects to
alleviate concerns that the status of diversification is endogenous and to control for other
unobservable cross-sectional heterogeneity (e.g., Campa and Kedia, 2000; Graham et al., 2002;
Fluck and Lynch, 1999).
In model 1 of Table 6, the coefficient on EEC is 0.068 but not significant.26 RVA is
positively related to excess value, consistent with the findings in Rajan et al. (2000). The
coefficient on RVA is 0.278, and is significant at the 5% level. Furthermore, the interaction
variable between EEC and RVADUM has a significantly positive coefficient, indicating that firms
with an efficient internal capital allocation that use more external capital are those with higher
valuation. The coefficients on diversity and the interaction between EEC and diversity are both
negative, albeit only the former significantly. This finding is consistent with Rajan et al. (2000)
and Lamont and Polk (2002) who find that higher diversity is associated with lower value. More
interestingly, the interaction between EEC, RVADUM and diversity is marginally significantly
positive with a coefficient of 0.069. Taken together, the data provide support for the argument
that firms that allocate capital efficiently, have a larger ICM and use more external capital are
22
also valued higher, i.e., interactions between internal and external capital markets do seem to
affect valuation. However, the insignificant coefficient on EEC is inconsistent with Stein (1997)
and seems to suggest that inefficiently allocating firms that use more external capital are not
significantly valued less.
In model 2, the number of segments rather than diversity is used as a measure of ICM size.
The results remain basically unchanged. One exception is that the coefficient on the number of
segments is not significantly negatively related to excess value, whereas diversity in model 1 is.
However, this finding is consistent with Lang and Stulz (1994), who show that excess value does
not decrease significantly beyond two segments.
There are two major concerns with this test of equation (2). First, Nickell (1981) shows that
using fixed-effect regressions in conjunction with dynamic panel data, e.g., panel data with a
lagged dependent variable, provides biased and inconsistent estimates. Second, Graham et al.
(2002) find that excess value, on average, decreases from before to after a merger. This decrease
is, to a large extent, caused by the target, which already has a negative excess value before the
merger. To control for this second problem, I re-estimate the models excluding firms that change
their number of segments.27 However, the inferences from regressions excluding firm-years
where the number of segments changed are qualitatively similar to the results discussed below,
and are omitted for brevity.
To address the first problem, and following Arellano and Bond (1991), I use GMM to
estimate equation (2) in first differences. Using first differences also eliminates the impact of a
firm fixed-effect. I employ lagged differences as instruments because the first differences of the
independent variables are still correlated with the residuals. In order for the lagged differences to
be valid proxies, the second order autocorrelation in residuals needs to be insignificant. Table 6
reports tests of the first and second order autocorrelations. The tests cannot reject the null
hypothesis of no second order autocorrelation in the residual.
When employing GMM, there is not much guidance on the optimal number of lags that
should be used as instruments. However, the Sargan test of over-identifying restrictions, reported
in Table 6, serves as a test of whether the set of instruments as a whole is uncorrelated with the
error term. The results reported use one lag only to keep the sample size as large as possible. For
those regressions, the p-value of the Sargan test is never below 0.1, confirming the validity of the
specification.28
Model 3 of Table 6 reports the coefficients using the Arellano and Bond technique.
Compared to the fixed-effect estimates, the number of observations has decreased from 8,538 to
6,368 because lagged differences are required as instruments. Note that the coefficient on EEC
23
changes its sign, and the magnitude and significance levels of all the coefficients are different.
The coefficient on EEC is now –0.102 and significant at the 1% level, indicating that firms that
raise more external capital without being efficient internal capital market users display lower
excess values. The impact of RVA is also estimated to be much larger with a coefficient that
increases from 0.278 to 1.089. The statistical significance in both cases is at the 5% level. More
importantly, the coefficient on the interaction variable between EEC, RVADUM and diversity has
increased from 0.069 to 0.116, and is significant at the 5% level. These results are consistent with
the predictions of Stein (1997), and emphasize that a firm that uses more external capital is valued
significantly higher if it has a large ICM and allocates capital efficiently in its ICM.
The coefficient on the lagged dependent variable is 0.530 and significant, consistent with the
finding of Lamont and Polk (2001). Relative to the fixed-effect model, the increase in the
coefficient is also consistent with the bias documented by Arellano and Bond (1991).
Model 4 shows Arellano and Bond regression results replicating model 2, which employs the
number of segments instead of diversity as a measure of ICM size. The coefficients and their
significance are similar to those in model 3. Finally, model 5 uses the Lang and Stulz (1994)
excess value measure. This change does not materially affect any of the inferences.
To summarize, the data suggest that firms with a la rger ICM and a more efficient internal
capital allocation raise more external capital, and doing so is associated with higher value,
consistent with the interpretation that interactions between internal and external capital markets
are important and reflected in firm valuation.
3 Industry Shock Sample
In this section I attempt to address the issue of causality of the results shown thus far by
studying firm behavior in the following situation. Theoretical models by Stein (1997), Li and Li
(1996) and others are based on the assumption that a new, positive-NPV project needs financing,
but the entrepreneurs’ wealth and/or the firm’s internal resources are insufficient to cover the
initial investment. In this section, I use a smaller sample that more closely mimics the setting in
which the models are specified. In this framework, firms are likely to underinvest if they cannot
access external capital markets to finance new investment projects. This implies that firms that
raise more external capital should increase investment, and a higher use of external capital should
lead to higher firm valuation. The question is whether ICM characteristics are an important
determinant of a firm’s ability to raise additional external capital in order to realize the growth
opportunity.
24
3.1 Sample Selection
The aim of the sample selection procedure is to choose firms that receive an exogenous,
positive shock to their investment opportunities. Such firms/divisions should display an increase
in Tobin’s q, holding everything else constant.29 However, Tobin’s q is not observable at the
segment level. Therefore, a sample of firms with operations in industries that have experienced a
significant increase in Tobin’s q is selected. Since industry q could increase as a result of
unexpected changes in industry cash flow, industry cash flow is required to remain constant.30
In order to select industries, only data on single segment firms are used. Industries are
defined at the 3-digit SIC code level. To make changes in industry q comparable across
industries, the change in the standardized industry median q between two consecutive years is
computed. The standardized industry median q is defined as follows:
Standardized industry median q =
−
q
σ
qqt ,
where qt is the industry median q at time t, q is the time-series average of industry median
qs, and qσ is the standard deviation of the time-series of industry median qs. As a control for the
industry cash flow, a measure of standardized industry median cash flow is defined as:
Standardized industry median cash flow =
−
cf
σ
cfcft ,
where cft is the industry median cash-flow-to-assets ratio at time t, cf is the time-series
average of the industry median cash-flow-to-assets ratio and, cfσ is the standard deviation of the
time-series of industry median cash-flow-to-assets ratios.
An industry is determined as having experienced a positive q shock if the change in the
standardized industry median q exceeds 1.25, and the change in the standardized industry median
cash flow is between –0.25 and +0.25.31 The rationale for using industry level qs rather than firm-
level qs is that changes in firm q could reflect the market’s view of idiosyncratic changes, such as
manageria l mistakes, which do not generally alter the set of investment opportunities. Industry-
level changes should better reflect changes in industry investment opportunities and industry cash
flow, thus allowing for a better control for expected capital needs. Using this procedure, 59 three-
digit SIC code industries with a positive q shock during 1980–1998 are obtained. Appendix 2 lists
all the sample-industries by event year and the change in their standardized industry q and cash
flow. One concern with this sample selection procedure is whether new firms entering the sample
could be responsible for the large increase in industry q. Appendix 2 shows that this is unlikely
25
because the number of single segment firms used to compute the annual standardized values is
very stable.
The final sample consists of diversified firms that have a segment in at least one of the
industries that experience a positive q shock. The selected diversified firms are also required to
have at least one segment in the industry with a positive q shock one year prior to the shock. In
addition, the sample selection criteria in section 2 are also observed. This results in a sample of
390 diversified firms with 497 segments in one of the selected industries. Appendix 2 shows the
number of segments per industry-year.
First, I investigate whether characteristics of a firm’s ICM help to explain the use of external
capital, given the exogenous shock to investment opportunities and controlling for differences in
the availability of internal cash flow. Next, the relation between the use of external capital and the
change in excess firm value is examined.
3.2 Determinants of the Use of External Capital
The research design in this section focuses on changes in the use of external capital by
diversified firms relative to comparable single segment firms. Changes are measured as the
difference between the values at t–1, the year before the industry shock, and t, the end of the year
in which the industry shock occurred.
3.2.1 Factors, Proxies and Predicted Effect
An increase in a segment’s investment opportunities should lead headquarters to allocate
more resources to that segment (Stein, 1997; Shin and Stulz, 1998). Firms that have an efficient
internal capital allocation and a larger ICM should be able to raise more external capital in order
to capture the new opportunities. Information asymmetry is again expected to have less of a
negative impact on a firm’s ability to use external capital if the firm is efficiently allocating
capital.
For the main part of the analysis I use independent variables as of t–1, the year-end before
the industry shock. Results using simultaneous changes of the independent variables between t–1
to t are also shown. However, the latter imposes a look-ahead bias because investors do not have
a measure of RVA at t at the time they have to make the decision of whether or not to supply
capital, and how much to supply. On the other hand, the exogenous change in at least one of the
segment’s investment opportunities requires a reallocation of the resources given that single
segment firms also change their investment strategy (not shown). If single segment firms are
credit constrained and diversified firms reallocate efficiently, RVA should increase due to the
26
exogenous shock to industry q, and thus establish causality.32 Given the trade-off, both results are
shown.
In the following analysis, I re-estimate a version of equation (1) with the change in EEC as
the dependent variable. I add a control for the relative size of the segment that operates in the
shocked industry, defined as the ratio of the segment assets in the positive q shock industry to
total assets (‘hit-size’). A positive coefficient on hit-size would indicate that firms that have a
larger fraction of their assets in one of the positive q shock industries (i.e., a diversified firm that
is more like a single segment firm) display a greater increase in their use of external capital than
diversified firms with only few assets in a positive q shock industry.
3.2.2 Results
Table 7, panel A shows univariate statistics. At t–1, the year before the industry shock, the
median EEC is positive for firms with an efficient internal capital allocation, i.e., those with
RVADUMt-1 equal to one, and negative for firms with an inefficient internal capital allocation.
This difference is significant at the 5% level. Similarly, in year t, efficient ICM firms have a
significantly higher median EEC than inefficient ICM firms. More importantly, those that
efficiently allocate capital in their ICM (measured by RVADUMt-1=1) increase EEC significantly
between t and t–1. The change in EEC for the other firms (RVADUMt-1=0) is insignificantly
different from zero. Similar results obtain if means are used and if the classification of efficient
versus inefficient internal capital allocation is based on the change in RVA rather than RVA at t–
1. The univariate analysis suggests an important effect of allocational efficiency on a firm’s use
of external capital. The following paragraph describes tests of whether these effects are still there
even after controlling for other determinants of a firm’s use of external capital.
Table 8 reports OLS regressions using the change in EEC as the dependent variable. Model 1
uses lagged values of the independent variables; model 2 uses the contemporaneous changes. In
model 1, the coefficient on diversity is significantly negative. The square term is not significantly
different from zero. The coefficient on RVA and the interaction between diversity and RVADUM
are significantly positive. Firms that allocate capital more efficiently raise more external capital,
and a larger ICM seems to help in doing so. Also, firms with more information asymmetry do not
increase their use of external capital, except if the internal capital allocation is efficient, as
indicated by the positive coefficient on the interaction variable between the intangible to total
assets and RVADUM in model 1 and standardized analysts’ forecast dispersion and RVADUM in
model 3. This suggests that diversified firms with more information asymmetry problems can
benefit from an efficient internal capital allocation by reducing the impact of information
problems on raising external capital.
27
The coefficient on internal cash flow at t–1 is marginally significantly positive. A positive
coefficient supports the notion that cash flow is used as collateral to raise new external capital. It
is also consistent with an interpretation that these firms are more profitable and, given the shock
to investment opportunities, should be given more capital to invest relative to less profitable or
less productive firms (Maksimovic and Phillips, 2002).
Measures of relative valuation are insignificant in this sample. The reduced importance of
lagged annual stock returns can be interpreted as supporting the sample selection. Here, the
reason for accessing external capital does not seem to be overvaluation but rather increased
investment opportunities. Further supporting this notion are the following statistics. Panel B of
Table 7 shows increased investment at the firm level (measured as the ratio of capital
expenditures to sales ratio minus the imputed ratio), more so in firms with an efficient internal
capital allocation and those that increase their use of external capital. Even more pronounced is
this pattern in the industry shock segments. Firms, classified as efficient internal capital allocators
increase the segment capital expenditures to sales ratio by 0.03 compared to inefficiently
allocating firms, with an increase of 0.01. The difference is highly significant. Moreover, the
change in segment investment (net of imputed capital expenditure to sales ratio) for firms with a
positive RVA at t–1 is significantly positive (0.014), while for firms with a negative RVA it is
significantly negative (–0.015). The univariate statistics also show median capital expenditure to
sales ratios (and adjusted ratios by imputed capital expenditures to sales ratios) for firms stratified
by the change in EEC. Firms with an increase in EEC increase investment; significantly more so
than firms that decrease EEC. These statistics are consistent with the notion that investment
opportunities increase, followed by an increase in investment, especially by firms with easier
access to external capital markets.
Model 2 of Table 8 shows the results using contemporaneous changes in the independent
variables. The change in diversity is negatively related to the change in EEC, unless the firm
increases its allocational efficiency measured by a dummy variable equal to one if the change in
RVA is positive, i.e., DUM( ∆ RVA ≥ 0). The coefficient on information asymmetry is negative,
but the interaction variable with DUM( ∆ RVA ≥ 0) is positive. This finding is in line with model 1
and suggests that firms, which improve their allocational efficiency due to the exogenous change
in investment opportunities, are less affected by changes in information asymmetry when they
transact with the external capital markets.
28
3.3 Changes in Excess Value and Changes in Excess Net External Capital
The dependent variable in this specification is the change in excess value between t–1 and t
computed, following Berger and Ofek (1995).33 The main variables of interest are EEC and its
interaction variables with ICM size and allocational efficiency.
Table 7, panel A shows the univariate statistics for excess value. Excess values are, on
average, negative for the sample as a whole in both years t and t–1. However, firms with an
efficient internal capital allocation display a significantly higher excess value in t and t–1 as well
as a significant increase between t and t–1. For firms classified as inefficient internal capital
allocators, excess value decreases. These results are generally consistent with Rajan et al. (2000)
and Cocco and Mahrt-Smith (2001) but a multivariate analysis has to show the inter-relation with
accessing external capital to finance growth on firm value.
Table 9 displays the OLS regression results.34 Model 1 shows that the change in EEC is
positively but insignificantly related to the change in excess value. However, the coefficients on
the interaction variables of EEC with RVADUM and diversity (0.828) and separately with
RVADUM only (0.223) are both significantly positive. On the other hand, neither the coefficient
on diversity (–0.013) nor on the interaction between the change in EEC and diversity (–0.200) is
significantly negative. The data are consistent with the interpretation that raising external capital
in situations where new growth opportunities require new investment is less harmful to
shareholders even for large, diversified firms with a relatively inefficient internal capital
allocation. However, relatively speaking, shareholders of large, diversified firms with an efficient
internal capital allocation benefit significantly more from an increase in external capital. This
argument is supported by the significance of the interaction variable between all three variables,
EEC, diversity and RVADUM.
In model 2 EEC at t–1 is used as an instrument for the change in EEC. The inferences are
similar to those from model 1 and support the notion that raising external capital is an important
determinate of firm value in this sample. Finally, model 3 shows results if the contemporaneous
change in RVA is used rather than the lagged value of RVA. Here, the contemporaneous change
in EEC and RVA are not significantly related to the change in excess value, and the interaction
variables are generally smaller and less significant than the coefficients in model 1.
As an additional test I also employ a two-stage procedure. The first stage regression to
predict the change in EEC is based on model 1 of Table 8. In the second stage regression, this
predicted change in EEC was used as an instrument. Even though the statistical significance of
the coefficients and the R-squared of the second stage regression decrease, none of the inferences
are affected (not tabulated).
29
In summary, diversified firms that have a larger ICM, allocate capital relatively more
efficiently and use more external capital can increase firm value. This finding is from a sample of
diversified firms that are likely to be in need of external capital in order to alleviate
underinvestment problems. The contribution of the industry shock sample analysis is twofold.
First, it shows that easier access to external capital markets can be achieved by diversified firms
with certain ICM characteristics and that a higher use of external capital is providing the firm
with the opportunity to capture new growth opportunities. Second, the sample selection procedure
should alleviate concerns about causality in the relations documented.
4 Conclusions
This study examines the interaction between internal and external capital markets. I find that
ICM characteristics are important determinants of a firm’s use of external capital. While firms
with a larger ICM, on average, use less external capital, firms with a larger ICM and a more
efficient internal capital allocation use significantly more external capital. In addition, the
analysis suggests that a more efficient internal capital allocation can help a firm to reduce the
impact of information asymmetry problems when raising external capital. More importantly,
firms using more external capital are valued lower, unless they have an efficient internal capital
allocation and a large ICM. Consequently, there is an additional benefit of diversification for
firms with an efficient internal capital allocation, namely, lower cost for external capital.
These findings are robust to different regression techniques and proxies for use of external
capital. The conclusions drawn are also unaffected if the sample is restricted to diversified firms
that have a division in an industry with a positive q shock, thus mimicking more closely a setting
in which a firm needs external financing to realize new, positive-NPV projects. This suggests a
partial answer to Zingales’s (2000) question about the factors that determine a firm’s ability to
capture new growth opportunities. This study shows the importance of a firm’s ICM
characteristics in financing, and thus capturing, new growth opportunities.
Taking a broader view, this study adds to research about decisions made in hierarchies
(ICMs) and markets (ECMs), surveyed in Stein (2001). Coase (1937) argues that firms exist
because transactions are less costly if made internally than externally, and Rajan et al. (2000)
conclude that there are important differences between hierarchies and markets. The results here
demonstrate that there are significant feedback effects from hierarchies to markets. Firms that
choose to transact in a hierarchy also change their ability to transact in markets.
30
Appendix 1 Description of Variables
MEASURES OF USE OF EXTERNAL CAPITAL Excess Net External Capital
(EEC) Excess net external capital = (net external capital – imputed net external capital) / lagged book value of assets [or standardized by lagged market value = market value of common equity plus book value of assets minus book value of common equity minus deferred taxes]. Net external capital = net common and preferred stock issued (#108-#115) plus net long-term debt issued (#111–#114) plus changes in short-term debt (#301). Numbers with #, refer to Compustat items. Imputed net external capital is computed as the segment sales (or asset) weighted sum of the median net external capital to sales (assets) ratio of single segment firms in the same 3-digit SIC code industry as the segment of the diversified firm. Sales-weighting is the standard.
Excess Net External Capital Including Dividends and Interest
Net external capital is defined as net common and preferred stock issued (#108–#115) plus net long-term debt issued (#111–#114) plus changes in short-term debt (#301) minus cash dividends (#127) minus interest paid (#15). Excess net external capital is computed as above.
Excess Net External Capital Including Asset Sales
Net external capital is defined as net common and preferred stock issued (#108–#115) plus net long-term debt issued (#111–#114) plus changes in short-term debt (#301) plus sale of PP&E (#107) plus sale of investments (#109) plus change of short-term investments (#309) plus sales of investing activities (#310) plus increases in other financing activities (#312). Excess net external capital is computed as above.
Excess Increase in Equity Increase in equity is defined as common and preferred stock issued (#108). Computing excess increase in equity follows the same steps as EEC.
Excess Increase in Debt Increase in debt is defined as long-term debt issued (#111). Computing excess increase in debt follows the same steps as EEC.
MEASURES OF ICM SIZE 1/Herfindahl Index
Inverse of the Herfindahl Index. ∑ ∑= =
=
n n
jjj SalesSales
1j
2
1 Index Herfindahl ,
where n is the number of segments and j refers to the segment. Number of Segments The number of segments a firm reports. Firms reporting more than five segments
are assigned the value five. Diversity
( )n
q
n
wqqw
n
jjn
j
j∑
∑ =
= −
− 1
1
2j
1, where wj is segment j’s share of total assets, qj is
imputed q, n is the number of segments and wq is the average asset weighted qj. wj and qj are beginning-of-the-period values.
MEASURES OF ICM EFFICIENCY Relative Value Added by
Allocation (RVA)
( ) BABA
Capex
BA
Capexw
BA
Capex
BA
Capexqq
n
jss
ss
j
jjss
ss
j
j
jj
j
j
j
j
−−−− ∑∑
== 1
n
1j BA , where
Capex is capital expenditures, qj is the asset-weighted average Tobin’s q of single-segment firms that operate in the same 3-digit SIC industry of segment j, n is the number of segments, BA is firm assets, BAj is segment assets and Capexj
ss/BAjss is the asset-weighted average Capex/asset ratio for the single
segment firms in the corresponding industry of segment j. wj is the ratio of segment assets to firm assets. BA, wj and qj are beginning-of-the-period values.
Absolute Value Added by Allocation (AVA)
( )
−−∑
=ss
ss
j
jn
jj
j
j
j
BA
Capex
BA
Capexq
BA
BA
1
1 ,
where Capex is capital expenditures, qj is the asset-weighted average Tobin’s q of single-segment firms that operate in the same 3-digit SIC industry of segment j, n is the number of segments, BA is firm assets, BAj is segment assets and Capexj
ss/BAjss is the asset-weighted average Capex/asset ratio for the single
segment firms in the corresponding industry of segment j. wj is the ratio of segment assets to firm assets. BA, wj and qj are beginning-of-the-period values.
31
Q-sensitivity ∑
=
−
×−×
n
j jj
j
SalesFirm
CapexFirm
SalesCapex
qqSalesFirm
Sales
1
)( , where Capex is
capital expenditures, qj is beginning-of-the-period median Tobin’s q of single segment firms that operate in the same 3-digit industry as segment j, q is the segment-asset-weighted average of the segment qs for the firm and n is the number of segments.
Cash Flow-sensitivity Cf-sensitivity is q-sensitivity except that qj is replaced with segment j’s cash flow to sales ratio and q is replaced with the median cash flow to sales ratio of single segment firms operating in the same 3-digit industry as segment j. Cash flow is defined as operating income plus depreciation.
MEASURES OF CAPITAL NEED Tobin’s q Tobin’s q is the market-to-book ratio, where market value is computed as
the market value of common equity plus book value of assets minus book value of common equity minus deferred taxes.
Excess Internal Cash Flow For firms with Compustat cash flow statements (#318=7), Internal cash flow is net cash flow from operations (#308) minus cash dividends (#127). For firms reporting a working capital statement, a cash statement by source and use of funds or a cash statement by activity (#318=1,2,3), internal cash flow is total funds from operations (#110) minus working capital change (#236) minus cash dividends (#127). Excess internal cash flow = (internal cash flow – imputed internal cash flow) / lagged book value of assets. Imputed internal cash flow is computed as the segment sales (assets) weighted sum of the median internal cash flow to sales (assets) ratio of single segment firms in the same 3-digit SIC code industry as the segment of the diversified firm.
Excess Internal Cash Flow Including Dividend and Interest
For firms with Compustat cash flow statements (#318=7), internal cash flow is defined as net cash flow from operations (#308) plus interest (#15). For firms reporting a working capital statement, a cash statement by source and use of funds or a cash statement by activity (i.e., #318=1,2,3), Internal cash flow is total funds from operations (#110) minus working capital change (#236) plus interest (#15). Excess internal cash flow is computed as above.
MEASURES OF INFORMATION ASYMMETRY Residual Variance Residual variance is computed over a calendar year by using daily returns
and a market model with the value-weighted CRSP index, including dividends as the market return. Variance is not annualized.
Total Variance Total variance is computed over a calendar year using the daily stock returns, including distributions. Variance is not annualized.
Intangible Assets / Total Assets Intangible assets (#33) divided by total assets (#6). Standardized Analysts’ Forecast
Dispersion The numerator is computed as the standard deviation of analysts’ forecasts of the firm’s one-year ahead fiscal year-end earnings per share (stdev). The denominator is the absolute value of the mean of the forecasts (meanest). Variables are from IBES.
MEASURES OF RELATIVE VALUATION Excess Value
(Berger and Ofek, 1995) Excess value =
)(log VIV , where ( )[ ]MSi
n
i i SalesVMSalesVI /)( 1 ×= ∑ = ,
where V is the sum of market value of equity and book value of assets less the book value of equity and deferred taxes, I(V) is the imputed firm value, Salesi is the segment i’s sales, Mi(V/Sales)MS is the sales multiplier (calculated as the median of the single segment firms in the same 3-digit SIC code industry), and n is the number of segments per firm.
Excess Value (Lang and Stulz, 1994)
Log of the ratio of the firm’s actual Tobin’s q at the end of the year to the sum of segment sales-weighted imputed qs.
Annual Stock Return Total return over a calendar year.
32
Appendix 2 Description of the Industry Shock Sample
Industries are defined at the 3-digit SIC code level using single segment firms only. Industries are selected based on the change in the standardized industry median q and the change in the standardized industry median cash flow, which are defined as follows:
Standardized industry median q =
−
q
σ
qq t ,
where qt is the industry median q at time t, q is the time-series average of industry median qs and qσ is
the standard deviation of the time-series of industry median qs.
Standardized industry median cash flow =
−
cf
σ
cfcf t ,
where cft is the industry median cash flow to assets ratio at time t, cf is the time-series average of the
industry median cash flow to assets ratio and cfσ is the standard deviation of the time-series of industry
median cash flow to assets ratios. An industry is selected to have experienced a positive q shock if the change in the standardized industry median q exceeds + 1.25, and the change in the standardized industry median cash flow is between –0.25 and +0.25. Diversified firms are selected into the sample if they have at least one segment in one of the selected industries in the year of the shock and have at least one segment in that respective industry prior to the shock. The year of the shock is indicated by t. The final sample consists of 390 diversified firms. Year of Shock
3-digit SIC
Change in standardized industry q, t–1 to t
Mean standardized industry q, t
Change in standardized industry cash flow, t–1 to t
Mean standardized industry cash flow, t
Number of single segment firms, t
Number of single segment firms, t–1
Number of segments in diversified firms
1982 221 1.36 0.61 0.07 -0.91 7 7 10 1982 333 1.26 0.49 -0.11 -1.25 4 4 1 1983 131 2.99 0.75 -0.14 -0.49 68 54 55 1983 138 1.85 0.83 -0.16 -0.42 20 21 42 1985 104 1.38 0.56 0.06 0.27 7 7 5 1985 225 1.65 0.80 0.10 -0.79 13 10 0 1985 245 1.51 -0.16 0.00 -0.28 14 15 4 1985 262 2.02 1.07 0.18 -0.01 7 6 3 1985 273 2.25 1.67 -0.16 -0.49 4 6 14 1985 539 2.31 1.16 -0.03 -0.88 4 6 1 1985 581 1.63 0.49 -0.05 -1.18 65 55 23 1986 262 1.39 2.46 0.17 0.16 7 7 3 1986 349 1.27 1.11 -0.21 0.01 5 8 13 1986 571 1.41 0.71 0.10 -0.05 12 10 1 1987 331 2.94 2.43 0.24 0.67 14 14 10 1987 333 2.56 1.25 -0.07 0.30 6 7 3 1988 232 1.47 0.47 -0.16 -0.02 12 11 2 1988 267 1.90 1.22 0.16 -0.12 9 10 10 1988 349 3.98 2.82 -0.17 0.54 6 3 9 1988 369 1.27 0.35 -0.20 -0.39 13 12 8 1988 473 1.91 -0.11 -0.18 -0.66 5 6 0 1988 509 2.23 1.14 0.21 -0.03 9 8 5 1988 512 1.27 0.48 0.03 -0.11 8 9 8 1988 531 2.13 1.16 0.14 -0.98 15 16 0 1988 599 1.86 1.13 0.17 -0.18 6 8 5
33
Year of Shock
3-digit SIC
Change in standardized industry q, t–1 to t
Mean standardized industry q, t
Change in standardized industry cash flow, t–1 to t
Mean standardized industry cash flow, t
Number of single segment firms, t
Number of single segment firms, t–1
Number of segments in diversified firms
1988 794 1.32 0.46 -0.19 0.27 7 7 2 1990 353 1.62 0.56 0.20 0.14 12 12 28 1990 373 1.81 -0.22 -0.05 -0.24 5 6 2 1991 205 3.34 0.78 0.06 0.26 5 5 2 1991 273 2.54 2.07 0.13 0.23 6 6 5 1991 282 1.57 -0.19 0.00 -0.47 6 6 10 1991 333 1.55 0.28 -0.07 -0.62 5 6 5 1991 346 1.33 0.27 -0.19 -0.63 6 6 2 1991 354 3.16 2.28 -0.13 -0.87 9 12 16 1991 363 3.41 1.77 -0.21 1.04 8 6 3 1991 371 2.36 0.87 0.19 -0.51 32 29 23 1991 736 1.91 0.78 0.12 -0.55 14 18 9 1991 738 2.12 1.60 -0.08 0.40 19 19 9 1991 808 1.34 1.04 0.00 0.00 11 6 3 1993 104 2.58 2.47 -0.01 0.25 32 33 2 1993 205 1.98 1.16 0.23 0.99 6 5 1 1993 282 1.99 1.55 -0.23 -1.05 6 6 13 1993 331 1.91 1.78 0.09 0.30 28 30 11 1993 871 1.27 0.64 -0.20 -0.50 19 19 12 1994 346 2.13 1.79 -0.03 -0.20 7 6 5 1994 473 1.28 0.21 -0.22 -0.71 7 6 1 1995 394 1.73 0.15 0.14 -0.96 28 23 8 1995 484 1.96 0.85 -0.24 -0.27 15 12 5 1995 489 2.32 0.99 0.01 0.03 7 6 4 1995 492 2.23 -0.16 -0.25 -0.47 34 35 40 1995 505 1.38 0.60 -0.20 0.02 6 5 5 1995 874 2.67 2.39 -0.25 -0.20 7 8 8 1996 230 1.60 0.50 0.14 0.12 6 6 1 1996 245 1.67 1.00 0.03 -0.30 10 8 1 1996 501 1.82 -0.11 0.00 -1.01 12 9 4 1996 591 1.45 0.84 -0.02 -0.80 12 14 3 1997 805 1.80 1.05 0.02 -0.87 17 21 3 1998 399 2.53 1.42 0.19 -0.02 8 10 9 1998 781 1.59 0.73 0.02 -0.55 10 16 7
Mean 1.95 0.97 -0.02 -0.26 12.92 12.58 8.42 Sum 762 742 497
34
References
Ackerman, R. W., 1968, Organization and the investment process: A comparative study, Unpublished doctoral dissertation, Harvard Business School.
Arellano, M., and S. Bond, 1991, Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations, Review of Economic Studies 58, 277-297.
Asquith, P., and D. W. Mullins, 1986, Equity issues and offering dilution, Journal of Financial Economics 15, 61-100.
Berger, P. G., and E. Ofek, 1995, Diversification’s effect on firm value, Journal of Financial Economics 37, 39-65.
Billett, M. T., and D. C. Mauer, 2002, Cross subsidies, external financing constraints, and the contribution of the internal capital market to firm value, University of Iowa Working Paper.
Blanchard, O. J., F. Lopez-de-Silanes, and A. Shleifer, 1994, What do firms do with cash windfalls? Journal of Financial Economics 36, 337-360.
Campa, J. M., and S. Kedia, 2000, Explaining the diversification discount, Harvard Business School Working Paper.
Chevalier, J. A., 2000, Why do firms undertake diversifying mergers? An examination of the investment policies of merging firms, University of Chicago Working Paper.
Coase, R. H., 1937, The nature of the firm, Economica 4, 386-405.
Comment, R., and G. A. Jarrell, 1995, Corporate focus and stock returns, Journal of Financial Economics 37, 67-87.
Cocco, J. and J. Mahrt-Smith, Return sensitivity to industry shocks: Evidence on the (in-)efficient use of internal capital markets, London Business School Working Paper.
Denis, D. J., D. K. Denis, and A. Sarin, 1994, The information content of dividend changes: Cash flow signaling, overinvestment, and dividend clienteles, Journal of Financial and Quantitative Analysis 29, 567-588.
Dierkens, N., 1991, Information asymmetry and equity issues, Journal of Financial and Quantitative Analysis 26, 181-199.
Fama, E. F., and J. D. MacBeth, 1973, Risk, return and equilibrium – empirical tests, Journal of Political Economy 81, 607-636.
Fee, E. C., and S. Thomas, 1999, Corporate diversification and asymmetric information: Evidence from stock market trading characteristics, University of Pittsburgh Working Paper.
Fluck, Z., and A. W. Lynch, 1999, Why do firms merge and then divest? A theory of financial synergy, Journal of Business 72, 319-346.
Froot, A. K., D. S. Scharfstein, and J. C. Stein, 1993, Risk management: Coordinating corporate investment and financing policies, Journal of Finance 48, 1629-1658.
35
Gertner, R. H., D. S. Scharfstein, and J. C. Stein, 1994, Internal versus external capital markets, Quarterly Journal of Economics 109, 1211-1230.
Graham, J., M. Lemmon, and J. Wolf, 2002, Does corporate diversification destroy value?, Journal of Finance 57, 695-720.
Grossman, S., and O. Hart, 1986, The costs and benefits of ownership: A theory of vertical and lateral integration, Journal of Political Economy 94, 691-719.
Hadlock, C., M. Ryngaert, and S. Thomas, 2001, Corporate structure and equity offerings: Are there benefits to diversification? Journal of Business 74, 613-636.
Harford, J., and G. D. Haushalter, 2000, Cash-flow shocks, investment, and financial constraint: Evidence from a natural experiment, University of Oregon Working Paper.
Jung, K., Y. Kim, and R. M. Stulz, 1996, Timing, investment opportunities, managerial discretion, and the security issue decision, Journal of Financial Economics 42, 159-186.
Khanna, N., and S. Tice, 2001, The bright side of internal capital markets, Journal of Finance 56, 1489-1528.
Lamont, O., 1997, Cash flow and investment: Evidence from internal capital markets, Journal of Finance 52, 83-110.
Lamont, O., and C. Polk, 2002, Does diversification destroy value? Evidence from industry shocks, Journal of Financial Economics 63, 51-77.
Lamont, O., and C. Polk, 2001, The diversification discount: Cash flow vs. returns, Journal of Finance 56, 1693-1722.
Lang, L. H. P., and R. M. Stulz, 1994, Tobin’s q, corporate diversification, and firm performance, Journal of Political Economy 102, 1248-1280.
Li, D. D., and S. Li, 1996, A theory of corporate scope and financial structure, Journal of Finance 51, 691-709.
Lucas, D. J., and R. L. McDonald, 1990, Equity issues and stock price dynamics, Journal of Finance 45, 1019-1034.
Maksimovic, V., and G. Phillips, 2002, Do conglomerate firms allocate resources inefficiently across industries?, Journal of Finance 57, 721-767.
Matsusaka, J. G., and V. Nanda, 2001, Internal capital markets and corporate focusing, forthcoming, Journal of Financial Intermediation.
Mikkelson, W. H., and M. M. Partch, 1986, Valuation effects of security offerings and the issuance process, Journal of Financial Economics 15, 31-61.
Miller, M. H., and K. Rock, 1985, Dividend policy under asymmetric information, Journal of Finance 40, 1031-1051.
Myers, S., 1977, Determinants of corporate borrowing, Journal of Financial Economics 5, 147-175.
Myers, S., and N. Majluf, 1984, Corporate financing and investment decisions when firms have information investors do not have, Journal of Financial Economics 13, 187-221.
36
Nickell, S., 1981, Biases in dynamic models with fixed effects. Econometrica 49, 1417-1426.
Peyer, U. C., and A. Shivdasani, 2001, Leverage and internal capital markets: Evidence from leveraged recapitalizations, Journal of Financial Economics 59, 477-515.
Rajan, R., H. Servaes, and L. Zingales, 2000, The cost of diversity: The diversification discount and inefficient investment, Journal of Finance 55, 35-80.
Scharfstein, D. S., 1998, The dark side of internal capital markets II: Evidence from diversified conglomerates, NBER Working Paper.
Scharfstein, D. S., and J. C. Stein, 2000, The dark side of internal capital markets: Divisional rent-seeking and inefficient investment, Journal of Finance 55, 2537-2564.
Schlingemann, F. P., R. M. Stulz, and R. A. Walkling, 2001, Divestitures and the liquidity of the market for corporate assets, forthcoming Journal of Financial Economics.
Schoar, A. S., 2001, Effects of corporate diversification on productivity, forthcoming Journal of Finance.
Shleifer, A., and R. W. Vishny, 1992, Liquidation values and debt capacity: A market equilibrium approach, Journal of Finance 47, 1343-1367.
Shin, H., and R. M. Stulz, 1998, Are internal capital markets efficient?, The Quarterly Journal of Economics 113, 531-552.
Shyam-Sunder, L., and S. Myers, 1999, Testing static tradeoff against pecking order models of capital structure, Journal of Financial Economics 51, 219-244.
Stein, J. C., 1997, Internal capital markets and the competition for corporate resources, Journal of Finance, 111-134.
Stein, J. C., 2001, Agency, information and corporate investment, forthcoming in Handbook of Economics and Finance. Eds: G. Constantinides, M. Harris and R. Stulz. Amsterdam: North Holland.
Stulz, R. M., 1990, Managerial discretion and optimal financing policies, Journal of Financial Economics 26, 3-28.
Villalonga, B., 2000, Does diversification cause the “diversification discount”?, University of California, Los Angeles Working Paper.
Weston, F. J., 1970, Mergers and acquisitions in business planning, Rivista Internazionale di Scienze Economiche e Commerciali, 309-320.
Williamson, O. E., 1970, Corporate control and business behavior. Prentice Hall, Englewoods Cliffs, N.J.
Williamson, O. E., 1975, Markets and hierarchies: Analysis and antitrust implications. Collier Macmillan, New York.
Williamson, O. E., 1986, Economic organization: Firms, markets and policy control. New University Press, New York.
Zingales, L., 2000, In search of new foundations, Journal of Finance 55, 1623-1653.
37
Table 1
Summary Statistics Univariate Statistics using data from 1980 to1998. Variables are defined in Appendix 1. Single segment firms are those with only one segment reported on the Compustat segment file. Diversified firms are those with more than one segment reported. t–1 indicates that the one-year lagged value of the variable is used. Medians and means of the panel data are reported. Significance levels are indicated by *, **, *** corresponding to a 10%, 5%, 1% significance level. Panel A reports univariate statistics by classifying firms as either single segment or diversified based upon the number of segments a firm reports. Panels B and C report univariate statistics for diversified firms only. Panel A: Single Segment and Diversified Firms Variable Median
Single Segment
Median Diversified
Mean Single Segment
Mean Diversified
Standard Deviation Diversified
Excess Net External Capital (EEC) 0 -0.00549*** 0.04442*** 0.02966*** 0.24803 Excess Net External Capital Including
Dividends and Interest 0 -0.00880*** 0.04198*** 0.02415*** 0.24712
Excess Net External Capital Including Asset Sales
0 -0.00645*** 0.04840*** 0.03316*** 0.24776
Excess Increase in Equity 0 -0.00510*** 0.03210*** 0.02091*** 0.20233 Excess Increase in Debt 0 0.0155*** 0.04551*** 0.0505*** 0.26495 Excess Internal Cash Flow 0 0.00394*** -0.00336*** -0.00389** 0.13096 Excess Internal Cash Flow Including Dividends and Interest
0 0.00314*** -0.00725*** -0.00386** 0.13287
1/Herfindahl Index Based on Sales 1 1.77810*** 1 1.89100*** 0.68418 Diversity 0 0.28617*** 0 0.31481*** 0.18982 Residual Variance 0.00094*** 0.00057*** 0.00186*** 0.00129*** 0.00377 Total Variance 0.00101*** 0.00063*** 0.00192*** 0.00135*** 0.00379 Intangible Assets / Total Assets 0 0.00906** 0.04561*** 0.06147*** 0.10985 Annual Stock Return 0.05101*** 0.08191*** 0.15464*** 0.14838*** 0.56826 Excess Value (Berger and Ofek, 1995, with sales multiplier)
0 -0.14866*** -0.00913 -0.16339*** 0.70348
Excess Value (Lang and Stulz, 1994) 0 -0.15542*** 0.03193*** -0.13374*** 0.59810 Relative Value Added by Allocation (RVA) 0 -0.000005 0 -0.0004* 0.01950 Absolute Value Added by Allocation (AVA) 0.000 -0.000489*** 0.00104*** -0.00110*** 0.00879 Q-sensitivity 0 0 0 0.00117** 0.02797 Cf-sensitivity 0 0 0 -0.00008* 0.00411 Number of Observations 34065 8538 34065 8538 8538
Panel B: Excess Net External Capital by Diversity and Allocational Efficiency for Diversified Firms Excess Net External Capital RVA ≥ 0 at t–1 RVA < 0 at t–1 Diversity ≥ Median Diversity Median 0.005 -0.016 (Mean) (0.054) (-0.013) Diversity < Median Diversity Median 0.001 -0.009 (Mean) (0.025) (-0.006) Panel C: Excess Value, Allocational Efficiency and Excess Net External Capital for Diversified Firms Excess Value (Berger and Ofek, 1995) RVA ≥ 0 at t–1 RVA < 0 at t–1 Diversity ≥ Median Diversity EECt–1 ≥ Median EEC Median 0.030 -0.330 (Mean) (0.045) (-0.359) EECt–1 < Median EEC Median -0.066 -0.169 (Mean) (-0.129) (-0.190) Diversity < Median Diversity EECt–1 ≥ Median EEC Median 0.022 -0.277 (Mean) (0.031) (-0.319) EECt–1 < Median EEC Median -0.106 -0.200 (Mean) (-0.121) (-0.200)
38
Table 2
Excess Net External Capital
Time-series averaged coefficients of cross-sectional OLS regressions on a year-by-year basis are reported (Fama-MacBeth, 1973). All the data are for the period 1980–1998. The dependent variable is excess net external capital. The t-statistics are based on the time-series standard deviation of the coefficients and are reported underneath the coefficients. R-squared are time-series averages. The Herfindahl Index is based on sales. The number of segments range from 1 to 5, where 5 includes firms with 5–10 segments. RVADUMt–1 = 1 if RVA ≥ 0 at t–1. The addition, t–1, means that the one-year lagged value of the variable is used. In Model 5, RVADUMt–1 = 1 if q-sensitivity ≥ 0 at t–1. In Model 6, RVADUMt–1 = 1 if cf-sensitivity ≥ 0 at t–1. For other variable definitions see Appendix 1. The significance levels are indicated by *, **, *** corresponding to 10%, 5%, 1% significance levels. Model 8 only includes firms with available information on the standard deviation of analysts’ forecasts reported in IBES. Excess Net External Capital
Model (1) (2) (3) (4) (5)
Measures of Size of ICM Diversity, t–1 -0.003**
(2.064) -0.009** (2.184)
-0.010** (2.119)
(Diversity, t–1)2 -0.002 (1.057)
-0.002 (1.054)
Number of Segments, t–1 -0.026** (2.154)
(Number of Segments, t–1)2 -0.000 (0.243)
1/Herfindahl Index , t–1 -0.015** (2.101)
(1/Herfindahl Index, t–1)2 -0.001 (0.483)
Measures of ICM Efficiency Relative Value Added by Allocation
(RVA), t–1 0.879** (2.339)
0.887** (2.375)
0.865** (2.279)
0.875** (2.353)
Q-sensitivity, t–1 0.318** (2.850)
(Size of ICM, t–1) × (RVADUMt–1) 0.005*** (3.609)
0.017** (2.781)
0.040** (2.467)
0.044** (2.568)
0.019*** (2.902)
(Size of ICM, t–1)2 × (RVADUMt–1) 0.001 (0.802)
0.004 (1.244)
0.008** (2.763)
0.002 (0.972)
Measures of Information Asymmetry Residual Variance -6.544***
(3.200) -6.782*** (3.293)
-6.269*** (3.033)
-5.963** (2.884)
-6.998*** (3.422)
Intangible Assets / Total Assets, t–1 -0.176*** (5.804)
-0.182*** (5.814)
-0.191*** (4.922)
-0.194*** (5.040)
-0.182*** (5.472)
(Intangible Assets / Total Assets, t–1) × (RVADUMt–1)
0.159** (2.269)
0.153** (2.133)
0.119** (2.216)
0.122** (2.338)
0.168** (2.338)
Measures of Capital Need Tobin’s q (Beginning of Year) 0.024***
(3.514) 0.024*** (3.420)
0.024*** (3.474)
0.024*** (3.381)
0.026*** (3.506)
Excess Internal Cash Flow -0.725*** (14.259)
-0.725*** (14.261)
-0.730*** (14.272)
-0.726*** (14.268)
-0.724*** (14.095)
Measures of Relative Valuation Excess Value (Berger and Ofek), t–1 0.009**
(2.154) 0.009** (2.174)
0.008* (1.947)
0.009** (2.435)
0.009** (2.165)
Annual Stock Return, t–1 0.037*** (4.405)
0.037*** (4.388)
0.037*** (4.384)
0.0375*** (4.364)
0.036*** (4.187)
Number of Observations 8538 8538 8538 8538 8538 Average R-squared 0.266 0.271 0.272 0.294 0.276
39
Table 2 (continued)
Excess Net External Capital Excess Net External Capital
Model (6) (7) (8) (9)
Measures of Size of ICM Diversity, t–1 -0.006**
(2.013) -0.009** (2.198)
-0.016* (1.832)
-0.014** (1.960)
(Diversity, t–1)2 -0.002 (0.965)
-0.002 (1.067)
-0.008 (0.945)
-0.007 (1.064)
Measures of ICM Efficiency Relative Value Added by Allocation
(RVA), t–1 0.888**
(2.381) 0.867** (2.259)
0.845** (2.156)
Cf-sensitivity, t–1 0.019** (1.998)
(Size of ICM, t–1) × (RVADUMt–1) 0.017** (2.460)
0.017** (2.747)
0.021*** (3.693)
0.012** (2.013)
(Size of ICM, t–1)2 × (RVADUMt–1) 0.001 (0.893)
0.001 (1.001)
0.002 (0.668)
0.001 (0.521)
Measures of Information Asymmetry Residual Variance -6.464***
(2.921) -6.999***
(3.286) Total Variance -6.299***
(3.160)
Intangible Assets / Total Assets, t–1 -0.146*** (4.115)
-0.182*** (5.817)
-0.191*** (5.748)
(Intangible Assets / Total Assets, t–1) × (RVADUMt–1)
0.163** (2.154)
0.152** (2.128)
0.150** (2.098)
Standardized Analysts’ Forecast Dispersion
-0.009** (2.745)
Standardized Analysts’ Forecast Dispersion × (RVADUMt–1)
0.003** (2.103)
Measures of Capital Need Tobin’s q (Beginning of Year) 0.027***
(3.268) 0.024*** (3.423)
0.016* (1.868)
0.008 (1.395)
Excess Internal Cash Flow -0.603*** (12.371)
-0.724*** (14.057)
-0.644*** (10.560)
-0.697*** (13.343)
Measures of Relative Valuation Excess Value (Berger and Ofek), t–1 0.011**
(2.362) 0.009** (2.184)
0.010* (1.876)
Excess Value (Lang and Stulz), t–1 0.037*** (4.197)
Annual Stock Return, t–1 0.038*** (4.065)
0.037*** (4.392)
0.039*** (3.107)
0.047*** (4.636)
Number of Observations 8538 8538 4021 8538 Average R-squared 0.300 0.271 0.272 0.255
40
Table 3
Robustness Tests
Changes in Excess Net External Capital With Constant Demand for Capital OLS regressions using the sample of firms where the change in Tobin’s q and the change in cash flow between two consecutive years are within plus/minus 5%. The data are from 1980–1998. Year dummies are not reported. The dependent variable is the change in excess net external capital (∆ EEC). All the changes are measured between t–1 and t, where t is the end of the year for which Tobin’s q and cash flow are constant. Excess value is computed according to Lang and Stulz’s (1994) method. DUM(∆ RVA ≥ 0) is a dummy variable equal to one if the change in the RVA is positive. Cash flow is operating income before depreciation. For other variable definitions, see Appendix 1. On the first line, the coefficients are reported with their significance level indicated by *, **, *** corresponding to 10%, 5% and 1% significance level based on White-adjusted standard errors. In brackets, the absolute values of the t-statistics are reported.
Dependent Variable
Change in Excess Net External Capital
Model (1) (2) (3)
Measures of Size of ICM Change in Diversity 0.035**
(2.210) 0.035** (2.211)
0.035** (2.212)
Measures of ICM Efficiency Change in Relative Value Added by
Allocation 0.439** (2.294)
0.421** (2.001)
0.594*** (2.531)
Change in Diversity × DUM(∆ RVA ≥ 0) 0.051** (2.412)
0.046** (2.339)
0.050** (2.408)
Measures of Information Asymmetry Change in Residual Variance -1.466*
(1.701) -1.410* (1.655)
-1.535* (1.812)
Change in Residual Variance × DUM(∆ RVA ≥ 0)
1.122** (2.419)
1.023** (1.981)
1.099** (2.293)
Intangible Assets / Total Assets, t–1 0.041 (0.512)
Measures of Capital Need Change in Excess Internal Cash Flow -0.014
(0.118) -0.015 (0.121)
Change in Cash Flow -0.000 (0.912)
-0.000 (1.000)
Change in Tobin’s q -0.013 (0.363)
-0.013 (0.366)
Measures of Relative Valuation Change in Excess Value (Lang and Stulz) -0.015
(0.933) -0.014 (0.892)
Annual Return 0.069** (2.164)
0.069** (2.170)
Number of Observations 330 330 330 Adjusted R-squared 0.182 0.200 0.198
41
Table 4
Excess Net External Capital and Single Segment Firms
Time-series averaged coefficients of cross-sectional OLS regressions on a year-by-year basis are reported. The t-statistics are based on the time-series standard deviation of the coefficients and are reported, in brackets, underneath the coefficients. R-squared are time-series averages. The Multi-segment dummy is one if the firm reports more than one segment, and zero otherwise. AVADUMt–1 = 1 if AVA ≥ 0 at t–1, where t–1 indicates that the one-year lagged value of the variable is used. Other variables are defined in Appendix 1. Model 2 only includes observations for which the standard deviation of analysts’ forecasts, as reported by IBES, is available. Dependent Variable Excess Net External Capital
Model (1) (2)
Multi-segment Dummy -0.011*** (3.211)
-0.009** (2.528)
Measures of Size of ICM Diversity, t–1 -0.012**
(2.243) -0.013*** (4.451)
(Diversity, t–1)2 -0.003 (0.364)
-0.004 (0.995)
Measures of ICM Efficiency Absolute Value Added by Allocation (AVA) , t–1 0.559***
(7.642) 0.421*** (7.140)
Absolute Value Added by Allocation (AVA) , t–1 × (Multi-segment Dummy)
-0.041 (0.716)
-0.036 (0.877)
(Diversity, t–1) × (AVADUMt–1) 0.005** (2.161)
0.013** (2.110)
(Diversity, t–1)2 × (AVADUMt–1) 0.001 (0.722)
0.001 (0.471)
Measures of Information Asymmetry Residual Variance -2.577**
(2.334)
Intangible Assets / Total Assets , t–1 -0.309*** (4.129)
(Intangible Assets / Total Assets, t–1) × (AVADUMt–1) 0.090 (1.045)
(Intangible Assets / Total Assets, t–1) × (AVADUMt–1) × (Multi-segment Dummy)
0.211** (2.855)
Standardized Analysts’ Forecast Dispersion -0.013** (2.468)
Standardized Analysts’ Forecast Dispersion × (AVADUMt–1) 0.003 (0.865)
Standardized Analysts’ Forecast Dispersion × (AVADUMt–1) × (Multi-segment Dummy)
0.010** (1.958)
Measures of Capital Need Tobin’s q (Beginning of Year) 0.001
(0.106) 0.001 (0.113)
Excess Internal Cash Flow -0.619*** (9.407)
-0.600*** (6.726)
Measures of Relative Valuation Excess Value (Berger and Ofek), t–1 0.029***
(9.544) 0.036*** (8.103)
Annual Stock Return, t–1 0.045*** (8.360)
0.057*** (8.760)
Number of Observations 42603 20331 Average R-squared 0.221 0.205
42
Table 5
Robustness Tests
Model 1 is estimated using firm fixed-effects with year dummies (not reported). Models 2-5 report time-series averages of coefficients of cross-sectional regressions using different dependent variables. Excess increase in equity is the difference between common and preferred stock issued (Compustat item 108) by the diversified firm minus the imputed equity issued, standardized by lagged book value of assets. Excess increase in debt is the difference between long-term debt issued (Compustat item 111) by the diversified firm minus the imputed long-term debt issued, standardized by lagged book value of assets. RVADUMt–1 = 1 if RVA ≥ 0 at t–1, where t–1 indicates that the one year lagged value of the variable is used. For other variable definitions see Appendix 1. In models 2, 4 and 5, excess internal cash flow includes interest and dividends (see Appendix 1 for a more detailed definition). For model 1, the absolute values of the heteroskedasticity robust t-statistics are reported in brackets. For models 2-5, t-statistics are based upon time-series standard deviations of the coefficients (Fama-MacBeth, 1973). The R-squared reported are time-series averages except for model 1. The significance levels are indicated by *, **, *** corresponding to 10%, 5%, 1% levels, respectively. All the data are for the period 1980–1998. Dependent Variable Excess Net
External Capital Fixed-Effects
Excess Net External Capital Including Interest and Dividends
Excess Net External Capital with Asset Sales
Excess Increase in Equity
Excess Increase in Debt
Model (1) (2) (3) (4) (5)
Measures of Size of ICM Diversity, t–1 -0.008**
(2.164) -0.009** (2.155)
-0.007* (1.874)
-0.012** (2.268)
0.005 (0.766)
(Diversity, t–1)2 0.001 (0.227)
-0.000 (0.149)
-0.001 (1.094)
0.004 (1.061)
0.001 (1.089)
Measures of ICM Efficiency Relative Value Added by
Allocation (RVA) , t–1 0.308*** (7.711)
0777** (1.960)
0.674* (1.790)
0.455*** (5.225)
0.145** (2.413)
(Diversity, t–1) × (RVADUMt–1) 0.016** (2.742)
0.017*** (3.091)
0.012** (2.167)
0.010*** (2.983)
0.012** (2.111)
(Diversity, t–1)2 × (RVADUMt–1) 0.001 (0.096)
0.001 (0.437)
-0.000 (0.751)
0.001 (0.705)
0.001 (0.624)
Measures of Information Asymmetry Residual Variance -1.023*
(1.645) -8.307*** (3.886)
-5.022** (2.490)
-0.745 (0.739)
-0.196 (0.112)
Intangible Assets / Total Assets, t–1
-0.132** (2.689)
-0.156*** (4.887)
-0.108** (2.475)
(Intangible Assets / Total Assets, t–1) × (RVADUMt–1)
0.213** (6.867)
0.171** (2.231)
0.081* (1.794)
Standardized Analysts’ Forecast Dispersion
-0.071*** (5.910)
-0.034*** (3.611)
Standardized Analysts’ Forecast Dispersion × (RVADUMt–1)
0.065** (2.056)
0.010* (1.682)
Measures of Capital Need Tobin’s q (Beginning of Year) -0.000
(0.306) 0.025*** (3.392)
0.022*** (3.543)
0.026*** (3.964)
0.003 (0.904)
Excess Internal Cash Flow -0.649*** (29.400)
-0.687*** (15.780)
-0.783*** (17.358)
-0.326*** (4.857)
-0.406*** (4.517)
Measures of Relative Valuation Excess Value (Berger and Ofek),
t–1 0.004 (0.614)
0.004 (0.835)
0.014*** (3.070)
-0.002 (0.795)
-0.003 (0.546)
Annual Stock Return, t–1 0.027*** (6.175)
0.041*** (4.709)
0.032*** (3.779)
0.028*** (4.159)
0.033*** (4.197)
Number of Observations 8538 8538 8538 4021 4021 R-squared 0.172 0.257 0.222 0.261 0.121
43
Table 6
Excess Value and Excess Net External Capital
Models 1 and 2 use firm fixed-effects with year dummies (not reported). Models 3-5 use the Arellano-Bond (1991) technique. These regressions are estimated in first differences. In addition, the lagged (differenced) dependent variable is instrumented by the second lagged difference, thus reducing the sample size to 6,368. The dependent variable is excess value computed either according to Berger and Ofek’s (1995) sales multiplier method or according to Lang and Stulz’s (1994) method. RVADUM t–1 = 1 if RVA ≥ 0 at t–1, where t–1 indicates that the one-year lagged value of the variable is used. NA signifies not available. For other variable definitions, see Appendix 1. The absolute values of the heteroskedasticity robust t-statistics are reported in brackets. The significance levels are indicated by *, **, *** corresponding to 10%, 5%, 1% levels, respectively. All the data are for the period 1980–1998.
Dependent Variable Excess Value According to:
Berger and Ofek
Lang and Stulz
Model (1) (2) (3) (4) (5)
M ethod FE FE Arellano-Bond
Arellano-Bond
Arellano-Bond
Excess Net External Capital (EEC), t–1
0.068 (1.500)
0.071 (1.611)
-0.102*** (3.381)
-0.096*** (3.109)
-0.078** (2.460)
Relative Value Added by Allocation (RVA), t–1
0.278** (2.080)
0.276** (2.065)
1.089** (2.304)
1.097** (2.321)
0.398** (2.445)
EEC, t–1 × RVADUMt–1 0.145** (2.497)
0.141** (2.410)
0.152** (2.274)
0.158** (2.660)
0.105** (2.572)
Diversity, t–1
-0.032** (2.406)
-0.041* (1.799)
-0.036* (1.731)
(Diversity, t–1) × RVADUMt–1 0.022* (1.826)
0.026 (1.219)
0.023 (1.297)
EEC, t–1 × (Diversity, t–1) -0.033 (1.108)
-0.085** (1.987)
-0.055** (2.092)
EEC, t–1 × RVADUMt–1 × (Diversity, t–1)
0.069* (1.714)
0.116** (2.138)
0.188** (2.379)
Excess Value (Berger and Ofek), t–1 0.400*** (36.446)
0.416*** (37.195)
0.530*** (20.343)
0.532*** (20.396)
0.369*** (19.794)
Number of Segments – 1, t–1
-0.002 (0.241)
-0.018 (1.011)
(Number of Segments – 1, t–1) × RVADUMt–1
0.013 (0.922)
0.026 (1.693)
EEC, t–1 × (Number of Segments – 1, t–1)
-0.059** (2.432)
-0.101** (2.123)
EEC, t–1 × RVADUMt–1 × (Number of Segments – 1, t–1)
0.098** (2.002)
0.113** (1.984)
Number of Observations 8538 8538 6368 6368 6368
R-squared 0.583 0.580 NA NA NA
Sargan test: prob > chi2 NA NA 0.19 0.20 0.13
H0: no autocorrelation in first order (p-value)
NA NA 0.00 0.00 0.00
H0: no autocorrelation in second order (p-value)
NA NA 0.22 0.24 0.22
44
Table 7
Summary Statistics for the Industry Shock Sample Univariate statistics of diversified firms, which have at least one segment in the industry classified as experiencing a positive q shock. An industry is determined as having experienced a positive q shock if the change in the standardized q is greater or equal to 1.25 and the change in the standardized industry cash flow is between – 0.25 and + 0.25. For a more detailed description of the sample selection procedure, see Appendix 2. For definitions of the variables, see Appendix 1. Panel A reports medians on the first line, and means on the second line in brackets. Panel B reports medians only. Here Capex is the ratio of capital expenditures to sales. Imputed Capex is the median capital expenditures to sales ratio of single segment firms operating in the same industry as the segment. At the firm level imputed Capex is the segment sales-weighted sum of segment imp uted Capex. The significance levels for means and medians are indicated by *, **, *** corresponding to 10%, 5% and 1%, respectively. The p-values are based on mean comparison t-tests and Wilcoxon rank sign tests for medians. All the data are for the period 1980–1998. There are 390 firms in this sample with 497 segments in one of the shocked industries. Panel A: Univariate Statistics of Use of External Capital and Excess Value Variables Median
(Mean) RVA = 0
t–1 RVA < 0
t–1 p-value
difference ∆ RVA
= 0 ∆ RVA
< 0 p-value
difference -0.002 0.003 -0.009 0.05 0.000 -0.004 0.49 Excess Net External Capital,
t–1 (0.012) (0.022) (-0.003) (0.08) (0.013) (0.010) (0.91) 0.001 0.015 -0.011 0.00 0.017 -0.019 0.00 Excess Net External Capital,
t (0.029) (0.054) (-0.009) (0.00) (0.076) (-0.021) (0.00) 0.003 0.010 -0.002 0.01 0.016 -0.014 0.03 Change in Net External
Capital (0.017) (0.032) (-0.006) (0.03) (0.063) (-0.031) (0.01)
-0.087 -0.019 -0.168 0.01 -0.072 -0.090 0.63 Excess Value (Berger and Ofek), t–1 (-0.095) (-0.031) (-0.191) (0.02) (-0.076) (-0.103) (0.76)
-0.100 0.030 -0.248 0.00 -0.009 -0.168 0.01 Excess Value (Berger and Ofek), t (-0.109) (0.033) (-0.322) (0.00) (0.003) (-0.209) (0.02)
-0.010 0.058 -0.077 0.00 0.055 -0.063 0.01 Change in Excess Value (Berger and Ofek) (-0.014) (0.064) (-0.131) (0.00) (0.079) (-0.105) (0.01)
Panel B: Univariate Statistics of Firm and Segment Investment before and after the Industry Shock Variables Median RVA = 0
t–1 RVA < 0
t–1 p-value
difference ∆ EEC = 0
∆ EEC < 0
p-value difference
Firm Capex-Imputed Capex, t–1
-0.003 0.004 -0.009 0.00 0.001 -0.005 0.19
Firm Capex-Imputed Capex, t
-0.004 0.008 -0.015 0.00 0.009 -0.016 0.00
Change in (Firm Capex-Imputed Capex)
-0.001 0.004 -0.006 0.02 0.008 -0.011 0.01
Segment Capex-Imputed
Capex, t–1 -0.001 0.005 -0.006 0.08 0.001 -0.004 0.23
Segment Capex-Imputed Capex, t
-0.002 0.019 -0.021 0.00 0.018 -0.021 0.01
Change in (Segment Capex-Imputed Capex)
-0.001 0.014 -0.015 0.00 0.012 -0.016 0.00
Segment Capex, t–1
0.062 0.066 0.055 0.00 0.063 0.060 0.50
Segment Capex, t
0.088 0.097 0.069 0.00 0.099 0.066 0.00
Change in Segment Capex
0.018 0.030 0.010 0.00 0.033 0.005 0.00
45
Table 8
Changes in Excess Net External Capital
OLS regressions using the industry shock sample for the period 1980–1998. Year dummies are not reported. The dependent variable is the change in excess net external capital (∆ EEC). All the changes are measured between t–1 and t, where t is the end of the year in which the industry shock occurred. Excess Value is computed according to Berger and Ofek’s (1995) sales multiplier method. RVADUM is a dummy variable equal to one if RVA is not negative, and zero otherwise. DUM(∆ RVA ≥ 0) is a dummy equal to one if the change in RVA between t–1 and t is not negative. Hit-size is the ratio of segment(s) assets (segments that belong to the shocked industry) to total assets of the firm. For other variable definitions see Appendix 1. On the first line the coefficients are reported with their significance level indicated by *, **, *** corresponding to 10%, 5% and 1% significance based on White-adjusted standard errors. The absolute values of the t-statistics are reported in brackets underneath.
Dependent Variable
Change in Excess Net External Capital
Model (1) (2) (3)
Measures of Size of ICM Diversity, t–1
-0.016** (2.308)
-0.018** (2.176)
Diversity2, t–1
0.000 (0.180)
-0.003 (1.595)
Change in Diversity
-0.020** (2.051)
Measures of ICM Efficiency Relative Value Added (RVA), t–1
2.640** (2.144)
2.543*** (2.933)
Change in RVA
1.830* (1.696)
(Diversity, t–1) × RVADUMt–1
0.026** (2.239)
0.023** (2.482)
(Diversity, t–1)2 × RVADUMt–1
-0.003 (1.263)
0.002 (0.409)
Change in Diversity × DUM( ∆ RVA ≥ 0)
0.056** (2.533)
Measures of Information Asymmetry Intangible Assets / Total Assets, t–1
-0.207*** (3.565)
-0.321** (2.600)
Intangible Assets / Total Assets, t–1 × RVADUMt–1
0.137** (2.761)
Intangible Assets / Total Assets, t–1 × DUM( ∆ RVA ≥ 0)
0.149** (2.234)
Standardized Analysts’ Forecast Dispersion, t–1
-0.014** (2.263)
Standardized Analysts’ Forecast Dispersion, t–1 × RVADUMt–1
0.009* (1.788)
Measures of Capital Need Excess Internal Cash Flow, t–1
0.264* (1.926)
0.274* (1.901)
Change in Excess Internal Cash Flow
-0.325** (2.002)
Measures of Relative Valuation Annual Return, t–1
0.012 (0.523)
0.004 (0.208)
0.016 (0.694)
Measure of Size Hit-size
0.047 (1.177)
0.011 (0.229)
-0.012 (0.306)
Number of Observations 390 390 229 Adjusted R-squared 0.382 0.185 0.231
46
Table 9
Changes in Excess Value and Changes in Excess Net External Funds
OLS regressions using the industry shock sample for the period 1980–1998. Year dummies are not reported. The dependent variable is the change in excess value between t–1 and t. Excess Value is computed according to Berger and Ofek’s (1995) sales multiplier method. The variable ∆ EEC is the change in excess net external capital. RVADUM is a dummy variable equal to one if RVA is not negative and zero otherwise. DUM(∆ RVA ≥ 0) is a dummy equal to one if the change in RVA between t–1 and t is not negative, and zero otherwise. Hit-size is the ratio of segment(s) assets (segments that belong to the shocked industry) to total assets of the firm. For other variable definitions see Appendix 1. On the first line the coefficients are reported with their significance level indicated by *, **, *** corresponding to 10%, 5% and 1% significance based on White-adjusted standard errors. The absolute values of the t-statistics are reported in brackets underneath.
Dependent Variable
Change in Excess Value
Model (1) (2) (3) Change in Excess Net External Capital
( ∆ EEC) 0.126 (0.818)
0.148 (1.036)
Relative Value Added by Allocation (RVA), t–1
1.702** (2.553)
1.803** (2.725)
Change in Relative Value Added by Allocation ( ∆ RVA)
0.098 (0.361)
Diversity, t–1 -0.013 (1.562)
-0.012 (1.231)
-0.016* (1.718)
(Diversity, t–1) × RVADUM t–1 0.042*** (3.214)
0.041** (2.213)
(Diversity, t–1) × DUM(∆ RVA ≥ 0) 0.038** (2.749)
( ∆ EEC) × RVADUM t–1 0.223** (2.016)
( ∆ EEC) × (Diversity, t–1) -0.200 (0.620)
-0.420 (1.643)
( ∆ EEC) × (Diversity, t–1)× RVADUM t–1 0.828** (2.139)
( ∆ EEC) × DUM(∆ RVA ≥ 0) 0.147 (0.535)
( ∆ EEC) × (Diversity, t–1)× DUM(∆ RVA ≥ 0)
0.600** (2.336)
Excess Net External Capital (EEC), t–1 0.189 (1.323)
(EEC, t–1) × RVADUM t–1 0.255** (2.250)
(EEC, t–1) × (Diversity, t–1) -0.230 (0.899)
(EEC, t–1) × (Diversity, t–1)× RVADUM t–1 0.630** (2.261)
Excess Internal Cash Flow, t–1 0.002 (0.011)
0.035 (0.133)
0.185 (0.657)
Excess Value (Berger and Ofek), t–1 -0.358*** (3.882)
-0.364*** (3.868)
-0.362*** (3.928)
Hit-size 0.119 (0.926)
0.073 (0.579)
0.082 (0.684)
Number of Observations 390 390 390 Adjusted R-squared 0.479 0.430 0.371
47
Endnotes
1 For an overview, see Campa and Kedia (2000). 2 Cocco and Mahrt-Smith (2001) also study how diversified firms react to industry shocks and find that the
option to re-allocate capital in the ICM is most often abused by discounted conglomerates in the event of high industry returns.
3 For simplicity, I do not include an effort dilution cost which might exist because headquarters can appropriate some of the divisional managers’ private benefits (see Grossman and Hart (1986) and Stein (1997)). Including such an additional cost of diversification can reduce the benefits documented below.
4 The numerical values in this example are chosen to show the main determinants of the interactions between internal and external capital markets. Allowing for correlation in projects’ outcome (i.e., capital needs), differences in projects’ returns between the good and bad outcome and different ex ante outcome probabilities will affect the probability that the diversified firm can raise more external capital relative to single segment firms, unless the diversified firm can own more than two projects. Note that if the ex ante expectation of the good state occurring is high enough that overinvestment is cheaper than underinvestment, external investors will not impose credit constraints on single project firms. Under this assumption, a diversified firm might receive less external capital by increasing the number of projects under one roof.
5 This prediction is not unique to Stein (1997). Rajan et al. (2000) also predict this relation and empirically find support for it.
6 This is consistent with Lamont and Polk’s (2002) sample selection process. SIC 9 contains mostly nonoperating divisions. SIC 0 contains agriculture operations with an average of only about 40 single segment firms per year, which is insufficient to compute imputed values. SIC 6 contains financial firms, where the market-to-book ratio is difficult to interpret and many cash flow statement variables are not available.
7 I convert all dollar values to their 1990 level by applying a GDP deflator. The $10 million size limit mainly eliminates single segment firms.
8 The analysis has also been done at the 2-digit SIC level without changing any of the conclusions. 9 Results are also shown if only diversified firms with the same number of segments in two consecutive
years are used. 10 Only 19 firms have less than 127 trading days; 42 have less than 200 but more than 30. The mean and
median of the residual variance are not significantly different between the firms with less than 128 trading days and the firms with more than 127 trading days. None of the results changes significantly if the limit is set either at 127 or at 200 trading days.
11 Note that agency problems are not treated as a separate determining factor, although, they might have an indirect impact through other factors, such as ICM efficiency (e.g., Rajan et al., 2000; Scharfstein and Stein, 2000).
12 Chevalier (2000) shows that in her sample the ranking of imputed q and firm q only correspond in about 60% of the cases using the 3-digit SIC level to impute q. This is a potential problem in using RVA. Note, however, that the value added due to reallocation is higher when the differences in divisional investment opportunities are higher. Thus, in situations where the divisional qs are close, the ranking might be incorrect, but then investment should also not differ substantially.
13 Scharfstein (1998) uses segment cash flow for this reason. 14 Rajan et al. (2000) find a significantly negative mean RVA of –0.0012. Using only data up to 1993, as
Rajan et al. (2000), the mean RVA in my sample is a significantly negative –0.001. 15 Proxies based on prices can be lower either because the firm is well enough diversified such that the cost
of information asymmetry is reduced or because information asymmetry actually is lower. 16 For a detailed description, see Appendix 1. Note that none of the variables used to compute EEC is also
used to compute excess internal capital. Section 2.4 reports regression results using alternative definitions.
17 A robustness test with respect to the estimation technique, using firm fixed-effects, is shown in Table 5. 18 As noted before, the findings are robust to limiting the sample to the 7,035 firms that report the same
number of segments in two consecutive years. 19 Note that residual variance is not used in model 8 because residual variance itself might be determined, in
part, by analysts’ forecasts. However, qualitatively similar results obtain if residual variance is included.
48
20 AVA for single segment firms is defined using the industry-median q. Using the firms’ own q there is a
significant difference in the coefficient on AVA between single segment firms and diversified firms. The coefficient is higher for single segment firms.
21 Ackerman (1968) compares resource allocation in integrated and conglomerate companies. 22 Replacing diversity by the number of different 4-digit SIC codes reported by Compustat as a measure of
the size of the ICM, does not lead to a significant coefficient for single segment firms. Results for diversified firms are similar to the ones reported in model 3 of Table 2 (omitted for brevity).
23 The main reason for excluding dividends is that they are a strong commitment, and changes are very costly (e.g., Shyam-Sunder and Myers, 1999). This potential complication will become important in testing the relation between EEC and firm value since firms with higher dividends would display a lower EEC. Miller and Rock (1985) demonstrate that an increase in dividends is viewed as a positive signal regarding firm value. Denis et al. (1994) empirically confirm such a relationship. Thus, firms with higher dividends are expected to be valued higher. If dividends are used to signal rather than being viewed as an external capital market transaction, they potentially induce a negative correlation between EEC and firm value that is unrelated to the tests of interest in this study.
24 Graham et al. (2001), Campa and Kedia (2000) and Villalonga (2000) show that an average negative excess value does not imply that diversification per se destroys value. Rather, firms might endogenously choose to diversify and/or acquire targets that are significantly discounted even as stand alone firms. I address these issues in three ways, described in more detail below. First, as suggested by Campa and Kedia firm-fixed effect regressions are used. Second, lagged excess value, as in Lamont and Polk (2002), is added as a control and a consistent panel data technique (Arellano-Bond, 1991) is employed to estimate the regression in first differences and using lagged differences as instruments. Third, the industry shock sample with an exogenous event, described in section 3 below, is used to study changes around the event, holding other things constant.
25 Following Berger and Ofek (1995) in excluding observations where excess values are less than –1.386 or more than 1.386, neither the univariate nor the regression results change significantly. Results using the Berger and Ofek asset multiplier excess value are qualitatively similar and are not reported here.
26 In untabulated regressions, similar results obtain if sales growth, log of assets and EBIT/sales are included (e.g., Berger and Ofek, 1995).
27 As Graham et al. (2001) show, increases in the number of segments are caused by merger & acquisition activities in about two-thirds of the cases in their sample.
28 The Sargan test rejects the null with p < 0.05 if more than two lags are used. However, as shown in Arellano and Bond (1991), the Sargan test rejects the null too often in situations with lagged dependent variables and with a larger number of instruments.
29 That is, if the market sees a positive chance that the project will be realized or sold. 30 Lamont (1997), Blanchard et al. (1994) and Harford and Haushalter (2000) employ event study
methodologies to investigate the effect of shocks to cash flow on firms’ use of funds. Blanchard et al. (1994) and Harford and Haushalter (2000) analyze how firms transact with the ECM after the shock. However, they do not explore whether differences exist between single segment and diversified firms.
31 The cut-off values of 1.25, –0.25 and 0.25 are arbitrary. If a normal distribution of q is assumed, my standardization procedure computes a standard normal variable, where the value 1.25 corresponds to the 10th percentile using a one-tailed test. However, since the change between two standardized values is used, the probability is path dependent. To test whether the procedure inadvertently picks up industries that recover from a very low realization of the standardized q, the standardized industry median q and cash flow in the year of the shock, t, are shown in Appendix 2. The average standardized industry median q across all industries in the year of the shock is 0.97, and no industry has a standardized q below –0.22.
32 The predictions of Maksimovic and Phillips’ (2002) neoclassical model are that divisions with relatively lower productivity should decrease their size, given a positive demand shock, and higher productivity divisions should increase their size. Using cf-sensitivity rather than RVA, the conclusions drawn from such a measure of allocational efficiency are not different from those presented (not shown).
33 Qualitatively similar results are obtained if the Lang and Stulz’s (1994) method is used (not shown). 34 Since the industry shock sample includes no firm in two consecutive years, tests in this section are less
likely to be influenced by possible estimation bias introduced by uncontrolled time-series correlation nor is there a problem of using a lagged dependent variable.