Content uploaded by José-Miguel Gaspar
Author content
All content in this area was uploaded by José-Miguel Gaspar
Content may be subject to copyright.
The Role of Commonality Between CEO and Divisional Managers
in Internal Capital Markets
José-Miguel Gaspar
*
Massimo Massa
**
Abstract
We study the role played by the informal links, or “connections”, that relate the CEO and the divisional
managers. We argue that these links make it more likely that CEO and divisional managers share
similar beliefs, interpret information in a similar way, and belong to the same power group within the
firm. This creates trust that leads the CEO to channel more resources to the segments run by the
“similar” managers. Trust should have a positive impact on the value of the firm if it reduces the
misallocation of resources induced by the lobbying efforts of managers and the “socialist” allocation of
capital across divisions (Rajan et al, 2000). We study this phenomenon, by using data on a large sample
of multi-segment US corporations from 1996 to 2004. We show that the segments run by “connected”
managers receive more investment and exhibit lower sensitivity to their own cash flow short-falls (and
more sensitivity to the other segments’ cash-flow). At the firm-level, having more connected managers
presiding over high Q segments improves the transfer of resources by preventing wasteful socialistic
reallocation. This impact is stronger in the case the pressure for a socialistic allocation is higher — i.e.,
firms characterized by higher diversity in investment opportunities across divisions.
JEL Classification: G34, G30, L25
Keywords: trust; connections; internal capital markets; distribution of power.
*
Finance department, ESSEC Business School, Avenue Bernard Hirsch, 95021 Cergy-Pontoise, France.
Tel.: (33)-1-34433374. email: gaspar@essec.fr
. ** Finance department, INSEAD, Boulevard de Constance
77300 Fontainebleau, France. Tel.: (33)-1-60724481. email: massimo.massa@insead.edu
. We thank
Wolfgang Buehler, Gilles Chemla, François Derrien, Thierry Foucault, Markus Glaser, Ernst Maug,
Ludwig von Thadden, Geoffrey Tate, David Thesmar, and seminar participants at Mannheim
University, HEC Paris and University of Paris Dauphine for helpful comments. José-Miguel gracefully
acknowledges financial support from the ESSEC Private Equity Chair. Any remaining errors are our
own.
The Role of Commonality Between CEO and Divisional Managers
in Internal Capital Markets
Abstract
We study the role played by the informal links, or “connections”, that relate the CEO and the divisional
managers. We argue that these links make it more likely that CEO and divisional managers share
similar beliefs, interpret information in a similar way, and belong to the same power group within the
firm. This creates trust that leads the CEO to channel more resources to the segments run by the
“similar” managers. Trust should have a positive impact on the value of the firm if it reduces the
misallocation of resources induced by the lobbying efforts of managers and the “socialist” allocation of
capital across divisions (Rajan et al, 2000). We study this phenomenon, by using data on a large sample
of multi-segment US corporations from 1996 to 2004. We show that the segments run by “connected”
managers receive more investment and exhibit lower sensitivity to their own cash flow short-falls (and
more sensitivity to the other segments’ cash-flow). At the firm-level, having more connected managers
presiding over high Q segments improves the transfer of resources by preventing wasteful socialistic
reallocation. This impact is stronger in the case the pressure for a socialistic allocation is higher — i.e.,
firms characterized by higher diversity in investment opportunities across divisions.
JEL Classification: G34, G30, L25
Keywords: trust; connections; internal capital markets; distribution of power.
2
1. Introduction
Budgeting investment resources in multi-segment firms is a complex and lengthy negotiation
process between corporate headquarters and operating divisions. This process, in which the
company’s CEO plays the role of arbiter, determines the overall investment policy of the firm.
Investment is therefore at least partially dependent on the relative bargaining power of each
divisional manager and therefore ultimately on his relation with the CEO.
1
This paper studies the role played by the existence of commonalities between the CEO and
divisional managers on the resource allocation process
within
the firm. We focus on the
informal links, or “connections”, that relate people that belong to the same organization. We
consider the case when the CEO and the divisional managers have the same educational and
career background, or share characteristics due to affiliation with the same generation inside
the firm (they are in the same age group, they joined the firm at the same time, they got
promoted to their current position at the same time).
We argue that these features make it more likely that the CEO and the divisional
managers share similar beliefs and interpret the same piece of information in a similar way
(Boot and Thakor, 2004). Moreover, they also make it more likely that the CEO and the
divisional managers belong to the same power group within the firm. Managers with common
backgrounds, that met at the time of joining the firm or that belong to the same circle of
influence, are more likely to “stay in touch” and share information about their activities. They
are also more likely to belong to an “old boys” network and therefore share vested interests.
Both elements — the “thinking alike” factor and the direct personal interaction — concur to
create trust between the CEO and the divisional managers.
We argue that CEOs tend to channel more resources to the segments run by the managers
they perceive “closer” and trust. We also argue that this has an impact on firm value. Testing
for the value impact of trust is difficult, as we cannot directly observe whether the resource
transfers to the different segments induce investment above or below the optimal level.
However, we can observe the effect of trust on the misallocation of resources induced by the
lobbying efforts of managers.
Scharfstein and Stein (2000) argue that the CEO chooses a “socialist” allocation of capital
across divisions to prevent division managers from wasting too much effort in lobbying. This
leads to the suboptimal equilibrium of all divisions investing the same. Rajan et al. (2000)
argue that managers choose on purpose investments that destroy value to prevent ex-post
poaching by other divisions. To minimize the effects of this behavior, the CEO misallocates
investment resources in the direction of smaller or less profitable divisions. The amount of
1
Bower (1970) is a seminal reference that describes this process in detail.
3
misallocation is a function of the division’s diversity in resources and opportunities. If diversity
is high, transfers are made so as to improve the incentives to undertake efficient investments.
2
Therefore, resources flow toward the most inefficient divisions, leading to misallocation of
resources and less valuable firms.
Trust between CEO and divisional managers helps to reduce this misallocation. The more
the divisional managers at the helm of the productive segments trust the CEO, the less they
need to implement defensive investment schemes. In other words, if transfers are required by
the inability of the CEO to commit ex-ante to distribute the surplus, the existence of trust
between CEO and the managers presiding over the best divisions should reduce this wealth
destruction and increase value. At the same time, more trusted managers at the helm of the
good segments will probably have a higher ability to influence the CEOs’ decisions concerning
the allocation of resources preventing “socialistic” misallocations. These considerations suggest
that, in the presence of diversity between the different segments of the same conglomerate,
trust between the CEO and the divisional managers at the helm of segments with high
investment opportunities would limit the destruction of value due to the need of diverting
resources to the less productive segments.
We test these hypotheses focusing on multi-segment US corporations from 1996 to 2004.
We collect information on the cohort the CEOs and the divisional managers belong to in terms
of: age, time they joined the company, type of education, degree, school, and time they got
appointed to the position they hold. We then construct an index of Connection between CEO
and divisional managers on the basis of such information and then we define managers sharing
a higher degree of commonality with the CEO as “connected”.
We show that commonality with the CEO helps the divisional managers to get allocated
more resources. Our parameter estimates show that a one-standard deviation increase in the
Commonality index implies a 4.4% increase in the investment of the segment relative to the
average investment in the firm. This allocation is particularly significant in moments when the
segment is underperforming. We also show that the investment of segments run by connected
managers is less sensitive to their own cash flows but more sensitive to the cash-flows of the
other segments of the firm during bad times.
2
“Since incentives are distorted away from the optimal because of diversity of opportunities and resources, transfers
will be made in a direction that makes divisions less diverse — from divisions that are large and have good
opportunities to divisions that are small and have poor investment opportunities. Thus, the diversified firm may
misallocate some funds at the margin relative to the first-best to prevent greater average investment distortions
[defensive investment schemes]. The more diverse a firm’s divisions are, the greater the need to reallocate funds in
this way.” (Rajan et al., 2000)
4
At the firm level, and in the spirit of Rajan et al., (2000), we construct a measure that
captures how much the more attractive segments of the firm and the less attractive ones differ
in terms of the commonality of their managers with the CEO: the “connection gap”. The
connection gap is the difference in connections between segments with higher investment
opportunities and segments with lower investment opportunities. We show that transfers to
the segments with lower investment opportunities are less when the managers in segments with
good investment opportunities predominate (i.e., display a higher index of connection).
These findings are simultaneously reflected in firm value. The firm’s excess value — i.e., the
difference between the firm Q and the firm’s imputed Q (the weighted average of the single
segments Qs the firm operates in) — is positively related to the difference in connection between
the more productive segments of the firm and the less productive ones. An increase of one
standard deviation in the connection gap corresponds to a 2.3% absolute decrease in the
average conglomerate discount, equivalent to a 13% percent decrease. We then interact
connection gap with the degree of diversity of the investment opportunities in the
conglomerate. We show that most of the impact of the connection gap occurs via the channel
of diversity. The higher the diversity (and therefore the potential for misallocation of resources
due to “socialism”), the higher the positive impact of trust. In sum, trust between CEO and
divisional managers helps to reduce the negative effects of the misallocation of resources due to
the “socialist” tendencies in conglomerates.
We cross-check these findings by appealing to the geography of investment literature. The
effectiveness of trust in improving information transmission between managers and CEO
should be higher if the CEO and the divisional managers are geographically distant. Indeed,
the farther the CEO and the divisional managers are, the more the CEO should trust the
distant — and therefore less controllable — managers to allocate them resources and the more
the distant managers presiding good segments should trust the CEO in order not to implement
defensive investment schemes (Rajan et al., 2000, Scharfstein and Stein, 2000). In line with
our working hypothesis, we find that connected managers are better able to reduce the
sensitivity of their investment to their cash flows the more distant they are from the firm’s
headquarters. Moreover, the connection gap reduces the transfer to the bad segments and
increases firm value more the more distant the subsidiaries are from the headquarters.
One possible issue is that the existence of connections may be endogenous. That is, the
selection of the divisional managers may be due to their connections with the CEO, as he
chooses to appoint connected managers to the helm of the best/bigger segments. We control
for endogeneity by using an instrumental variable technique. The instruments are related to
the climate characteristics of the localities where the segments are located. We argue that
weather affects the attractiveness of a particular location to a manager while being at the same
5
time uncorrelated with segment performance. The instrumented results support our working
hypothesis.
Our findings relate to different streams of research. Coase (1937) first suggested that the
main difference between decisions taken inside a hierarchy and those taken in a market place is
the fact that the former are affected by control considerations rather than relative prices.
3
More recently, a growing literature has studied the role of manager specific characteristics on
firm value and investment (Bertrand and Schoar, 2003; Malmendier and Tate 2005a;
Malmendier and Tate 2005b; Guner et al. 2007), as well as the links between social networks
and diversity in the board (Adams and Ferreira 2004; Adams et al.
,
2005; Landier et al., 2005;
Subrahmanyam, 2006). We contribute to both strands of literature by showing how the
existence of a power network between CEO and divisional managers affects resource allocation
and the value of the firm.
We also directly relate to the literature on internal capital markets. This has shown that
less efficient resource allocation in conglomerates is associated with lower firm value (Lang and
Stulz, 1994; Berger and Ofek, 1995; Lamont, 1997; Shin and Stulz, 1998; Denis, Denis and
Yost, 2002; Lamont and Polk, 2002).
4
A number of theories have been put forward to explain
this result. At the firm level, the existence of private benefits of control and/or manager’s
aversion to idiosyncratic risk might imply overinvestment through wasteful diversification (e.g.
Jensen, 1986; Stulz, 1990; Matsusaka and Nanda, 1997; Denis et al., 1997; Aggarwal and
Samwick, 2003). In Meyer, Milgrom and Roberts’ (1992) influence cost model, managers of
segments with poor prospects lobby the headquarters for more resources. A conglomerate
discount therefore appears due to the ultimately useless but wasteful lobbying effort of the
segments. Wulf (2005) shows that in a context where divisional managers can distort the
information about the prospects of the segment as perceived by headquarters, the size of the
investment distortion is positively related to managers’ ability to exert such influence
activities. We relate to this literature by building on the models that show that, in order to
induce cooperation between segments in ex-post bargaining, headquarters transfer resources
from segments with good opportunities to segments that with poor investment opportunities
(e.g., Scharfstein and Stein, 2000, Rajan et al., 2000). We show how trust between CEO and
divisional managers helps to reduce this “socialist” reallocation of resources.
3
Gertner et al., (1994) analyze the importance of control rights in the context of internal capital markets. For
broader analyses in corporate finance, see e.g. Rajan and Zingales (1998) and Zingales (2000).
4
This result is not consensual and has been strongly called into question, due to econometric problems (e.g Whited,
2001; Campa and Kedia, 2002; Villalonga 2004a), data issues (e.g. Villalonga 2004b), and differences in productivity
(Maksimovic and Philips, 2002), valuation (Graham et al., 2002) and capital structure (Mansi and Reeb, 2002)
between single-segment and multi-segment firms.
6
Finally, our work is closely related to the recent literature showing that connections among
economic agents (e.g., Cohen et. 2007, Kuhnen, 2005) and trust (e.g., Guiso et al., 2008) affect
investment behavior. In the corporate context, Xuan (2006) looks at evidence of reverse-
favoritism by newly-appointed CEOs that are chosen from within the ranks of the firm’s
segments. We complement this literature by focusing on the commonality fostered by common
background and similar career paths.
The remainder of the paper is articulated as follows. In Section 2, we describe the data and
the way we constructed our proxies for trust and commonality. Section 3 reports the segment-
level findings about the relation between connection and investment, as well as with the cash-
flow sensitivity. Section 4 looks at the allocation of resources within the firm and the
implications in terms of firm’s excess value. Section 5 investigates how the effect of connections
is related to the distance of divisional managers from headquarters. Section 6 addresses the
issue of endogeneity. A brief conclusion follows.
2. Data description
2.1 Sample construction
Our primary data sources are the annual CRSP-Compustat Merged files (containing firm-level
accounting data) and the Compustat Segment files (containing segment assets, investment and
cash-flow data). We keep in the sample multiple-segment firms that have non-missing segment
SIC codes, CRSP share codes equal to 10 or 11, positive values for book equity and sales higher
or equal to 20 million USD (Berger and Ofek, 1995). We remove firms whose sum of segment
sales is more than 1 percent away from total firm sales reported in Compustat, as well as
financial companies and utilities (Berger and Ofek, 1995). After winsorizing all variables at the
1% level, a total of 43,090 segment-year observations remain.
We proceed to add biographical data on company managers, as follows. First, we obtain
information on the subsidiaries of each company through Dun & Bradstreet’s (D&B) Million
Dollar Database. This data source contains information on the identity of each subsidiary, its
position in the corporate structure, its number of employees, its SIC code and its location.
5
We
then create an algorithm that allocates company subsidiaries to each Compustat segment. The
algorithm allocates subsidiaries first by matching the segment’s SIC code and secondly (on the
basis of a text-matching score) by matching the business description of both the segment and
subsidiary. The algorithm finds matching subsidiaries for 47% (60%) of the total number (asset
value) of Compustat segments.
5
D&B also contains some accounting information on subsidiaries, but it is to be too sparse for effective use.
7
Second, we collect biographical information on each company’s CEO and middle-managers,
also using D&B as the source. This data contains information on the year of birth and year of
joining the company of about 31,959 middle managers and CEOs working in D&B identified
subsidiaries. We then use keywords to parse each manager’s biography in search of his career
and educational background (e.g. Malmendier and Tate, 2005). Once this data is merged with
Compustat, we are left with CEO biographical information for 23,465 segment-years and
middle-manager age information for 16,982 observations.
6
The final sample results in 10,459 segment-years that have non-missing values for our main
variables: beginning-of-year segment assets and sales, firm beginning-of-year assets, market
capitalization, segment investment, Tobin’s Q, firm age, Diversity (Rajan
et al., 2000) and
Connection (defined below). A detailed description of the merge of the datasets and the
statistics comparing our final usable sample with the Compustat base sample are reported in
the Appendix.
2.2 Proxies for Connection
The final sample contains biographical profiles on the company’s CEO and the middle-level
managers working in every subsidiary (”divisional managers”). We construct proxies for
Connection by conducting pairwise comparisons between the CEO’s profile and the divisional
managers’'profiles as follows. Each pairwise comparison consists of creating a dummy variable
that assumes the value 1 if a certain characteristic is common to both the CEO and a
divisional manager and zero otherwise. The 4 characteristics are: whether the CEO and the
divisional manager joined the company at the same time; if they belong to the same age
cohort; if they attained their current position at around the same time; and if they share the
same career background (i.e., Finance, Marketing, or Engineering).
7
The dummy variables D (one for each of the 4 characteristics, denoted d = 1,..,4) are
constructed for every manager (indexed by m) of every subsidiary k belonging to given segment
i of firm j. To get measures of connection at the segment level, we sum the dummy variables
across all managers and divide by the total number of managers working in the subsidiaries of
the segment. Denoting the dummy variables by D
d
m,k,i,j
, we have
∑∑
∑∑
∈∈
∈∈
×=
ikkm
d
j,i,k,m
ikkm
d
j,i
DConnection
1
1
6
These figures refer to the cases where we are able to identify the age and time of joining of at least one manager of
one subsidiary belonging to a given segment.
7
The career path is inferred from each manager’s biography using a keyword search (e.g. Malmendier and Tate,
2005) on their educational degrees and previous job positions.
8
Hence, Connection by Time of Joining is the segment’s proportion of CEO-manager pairs
for which the CEO and the divisional manager entered the firm within 2 years of each other;
Connection by Age Cohort is the proportion of CEO- manager pairs for which the CEO’s and
the manager’s age are within 4 years of each other; Connection by Time of Appointment is the
proportion of pairs for which the CEO and the divisional manager reached their current
management position within 2 years of each other; and Connection by Same Career is the
proportion of pairs for which the CEO and the divisional manager have similar career
backgrounds. Our index of Connection for segment i is the segment average of the four
Connection variables:
∑
×=
d
d
j,ij,i
ConnectionConnection
4
1
In most of our analysis, we also use a measure of ‘excess’ Connection defined as the difference
between the segment’s connection and the average Connection for the firm
×−=
∑
i
j,i
j
j,i
Net
j,i
Connection
N
ConnectionConnection
1
,
where N
j
is the number of segments belonging to firm j. We do this to accommodate the
possibility that some firms are by nature of their business more connected than others.
8
Henceforth, we use Connection as shorthand for ‘net’ Connection, unless specified otherwise.
2.3 Other Variables
We employ as control for alternative hypotheses the set of other variables used in previous
papers. We refer to the caption of Table 1 for precise definitions. Concerning segment level
variables, we use total capital expenditures I
i,j
of segment i (Investment), segment i’s cash-flow
(Same Segment Cash-Flow), and the sum of the cash-flows accruing to all the firm segments
apart from segment i (Other Segment Cash-Flow). All of the former are normalized by the
beginning-of-period total assets A
j
. We also employ the percentage of segment assets relative to
total firm assets (Weight of Segment) and a measure of the resources reallocated by the
internal capital market to a given segment (Transfer), defined (Rajan
et al., 2000) as
∑
∈
−−
−=
j
Ni
ss
j,i
j,i
j,i
j,i
j,i
ss
j,i
j,i
j,i
j,i
A
I
A
I
w
A
I
A
I
Transfer
where N
j
is now shorthand for the set of segments belonging to firm j, A
i,j
represents segment-
level assets, and w
i,j
is the weight of segment i on firm j’s total assets.
8
The formula for excess Connection attributes equal weight to every segment in the firm in calculating the firm-
level Connection. Results using value weights (segment’s assets or sales) are similar and available upon request.
9
Concerning firm-level variables, the ones most worth mentioning are: the percentage
growth in firm sales in the past year (Sales Growth), the firm’s number of business lines
reported in Compustat segment data (Number of Segments), the ratio between market value
and book value of assets (Tobin’s Q, defined as in Villalonga, 2004) and a measure of diversity
in investment opportunities between segments (Diversity), defined (Rajan
et al., 2000) as
(
)
∑
∑
∈
∈
−−
=
j
j
Ni
j
ss
i
Ni
j
ss
ij,i
ss
ij,i
NQ
NQwQw 1
Diversity
2
.
We also calculate the firm’s weighted sum of the Qs of the individual segments (Imputed Q)
by imputing to each segment the average Tobin’s Q of all single-segment firms in segment i’s
SIC code (denoted Q
i
ss
):
9
∑
∈
=
j
Ni
ss
ij,i
QwQ
ˆ
The difference between the firm’s Q and the Imputed Q captures the diversification discount.
We call it Excess Value. Since one of our objectives is to see how the existence of connections
affects the investment allocation process depending on the relative attractiveness of segments,
we define a high (low) attractiveness segment — or “High Q” (“Low Q”) segment for short — a
segment that is above (below) the firm’s imputed Q. We then compute the High (Low) Q
Segment Connection variable as the average Connection of the firm’s High (Low) Q segments.
10
2.4 Summary statistics
Table 1 presents summary statistics for our variables. The upper panel displays summary
statistics for segment-level variables. The average (median) segment represents 34% (28%) of
firm’s assets and makes investments of around 2% (1%) of firm assets per year. The median
transfer to each segment (
–0.0003) is not statistically different from zero.
11
The lower panel of
Table 1 displays summary statistics for firm-level variables. The average (median) firm in our
sample has more than 3.3 $Bn. (640$M) in revenues and operates in 2.9 (3.0) business
segments. The conglomerate discount is patent in the sample, with multi-segment firms
exhibiting an average (median) discount of 19% (19%) relative to single-segment firms. The
average and median sample values of Diversity are similar to the ones reported by Rajan, et
9
To calculate the average Q of single-segment firms, we use 4-digit SIC codes as long as there are at least 3 single-
segment firms in that SIC code; if not, we use 3-digit, and so forth.
10
For example, the averaging for the High segments is calculated assuming a zero weight on Low segments, and
vice-versa. This is akin to the way total transfers to High and Low segments are calculated in Rajan
et al
., (2000).
11
For clarity, Table 1 presents all values for Transfer multiplied by 100, just as Rajan
et al
., (2000).
10
al., (2000). The average (median) firm age is 22 (17) years, revealing that firms in our sample
are rather mature, established firms.
Concerning our variables of interest, the table indicates that an average 14% of the middle-
level managers identified in D&B share a common characteristic with the company’s CEO.
12
There is some dispersion in the Connection variable: the standard deviation is 17%, while the
inter-quartile range is 22%. The table also indicates the values of the individual Connection
variables: the most common connection is to belong to the same age cohort (18%), followed by
similar time of appointment (15%) and by similar time of joining (15%).
The median value of the individual connection variables, however, is zero, indicating that
the individual measures are probably less reliable than the Connection index. The data in the
Table also reports that each segment has on average 2.3 subsidiaries and a total of 8.4
managers in those subsidiaries.
13
Going to the firm-level, we find that the average (median) Connection is 14% (8%).
Interestingly, Connection is not related to the attractiveness of the segments: High Q segments
(which represent on average 42% of firm assets) display the same average and median levels of
Connection (respectively 8% and 3%) than Low Q Segments (which represent on average 45%
of firm assets).
14
This is very important as it suggests that there is no systematic bias or direct
relationship between the connection of the divisional managers and some firm characteristics.
15
3. Connection and Segment Resource Allocation
3.1 Connections and Segment Investment
The first question we address is whether connections affect the way resources are allocated to
individual segments. We estimate
t,j,it,jt,j,it,j,it,j,i
ZXCONNECTIONI εδµβα +++×+= (1)
where
I represents the normalized segment investment, and the matrices X and Z refer
respectively to a set of segment-level control variables and firm-level control variables. Base-
specification variables in
X include Same-Segment Cash-Flow, Other Segment Cash-Flow,
Segment Q and Weight of Segment, while firm-level variables in
Z include (the Log of) firm
12
These values refer to ‘gross’ connection. The ‘net’ connection has zero median and negligible mean by
construction, and a range between –0.6 and 0.5.
13
The latter number is the denominator over which all the Connection pairings are summed.
14
“Mid” segments refers to cases where segment Q and imputed Q are equal. These are firms that are “pseudo-
conglomerates” (Sanzhar, 2004), that is, firms whose reported segments all belong to the same 4-digit SIC code.
15
We interpret this as indicating that connections are the “accidental” outcome of an process developing over time
and can hence be treated as predetermined in econometric terms. We deal explicitly with this issue in section 5.
11
Sales and Sales Growth (Shin and Stulz, 1998). All regression specifications include firm fixed
effects and industry dummies. Standard errors are clustered by year.
16
We add two other specifications to the base specification (all of these will be used
consistently throughout the paper). The second specification adds firm-level controls for Firm
Age, Diversity, Imputed Q, Long-term Debt, Cash and Number of Segments.
17
The third
specification adds two groups of variables. The first group consists of Connection variables
meant to control for the fact that our results are not due to specific career or education
backgrounds by the manager and the CEO.
18
The second group consists of Intra-Manager
Connection variables that are constructed in a similar way as the original Connection
variables, with the difference that the proportion is calculated over the number of possible
pairs between managers of a segment (excluding the CEO). Indeed, if we observe that a
segment run by an old manager that has the same age as the CEO is channeled more
resources, this may be due to the fact that older divisional managers as a group have more
power, as opposed to the fact that there is a link between the managers and the CEO. By
explicitly adding variables that proxy for the amount of links between divisional managers, we
can control for this spurious correlation.
The results are reported in columns 1 though 3 of Table 2. The coefficients of the control
variables (Same Segment Cash-Flow, Other Segment Cash-Flow and Sales Growth) display
signs, magnitudes and significance levels similar to the ones reported by Shin and Stulz (1998).
The exception is the variable Segment Q which exhibits quite lower coefficients in all
specifications.
19
The R-squared above 40% (relatively high for a cross-sectional regression)
indicates a satisfactory overall fit of the model due to the use of firm fixed-effects.
16
We cluster by year to prevent the possibility of a time effect biasing inferences. Some of the readers might wonder
if clustering by firm is not necessary to have unbiased standard errors. Petersen’s (2007) results show that having
firm fixed effects is enough to ensure unbiased standard errors, under the assumption that the fixed effect is
permanent and not temporary. The short length of our panel makes us comfortable in assuming that the former is
the case in our data. In any case, results with clustering by firm produce similar-sized standard errors.
17
Diversity and (the inverse of) Imputed Q are used by Rajan
et al.,
(2000). The importance of controlling for
capital structure is pointed out by Mansi and Reeb (2002). Firm age and Cash holdings are proxies for the existence
of agency problems. Number of Segments control for the fact that Shin and Stulz (1998) suggest differences in
investment behavior between more and less diversified conglomerates.
18
‘Connection by Engineering/Marketing/Finance Career’ is the proportion of CEO-manager pairs for which the
CEO and the manager have both respectively an Engineering/Marketing/Finance background. ‘Connection by
MBA/PhD’ is the proportion of CEO-Subsidiary manager pairs for which the CEO and the manager have an MBA
or a PhD degree. Connection by Ivy League is the proportion of CEO-Subsidiary manager pairs for which the CEO
and the manager have a degree from an Ivy League university.
19
One possible reason might be our different sample period and the fact that we control for firm size (log of Sales)..
12
The coefficient of the Connection variable is positive (0.012 to 0.015, depending on the
specification) and statistically significant at the 1% level. This result is robust across all the
different specifications. The magnitude of the coefficient is economically meaningful: a 0.013
coefficient for Connection implies that a one-standard deviation increase in Connection implies
a 4.4% increase in segment investment relative to the mean.
20
Column 3 also shows that the
Intra-manager Connection variables are not significant, indicating that links between the
managers and the CEO matter as opposed to just the characteristics of the managers per se.
Columns 4 through 6 explore this result further by controlling for quality of corporate
governance. It may indeed be the case that connections are just another proxy for the overall
quality of governance of the firm. To address this issue, we include there alternative measures
of governance: the fraction of institutional ownership, the fraction of blockholders and the
index of corporate governance of Gompers, Ishii and Metrick (2003). The first two measures
are based on the literature arguing that institutional investors and blockholders provide a
measure of “external governance” (Maug, 1998, Kahn and Winton, 1998, Bolton and Von
Thadden, 1998, Noe 2002, and Faure-Grimaud and Gromb, 2004). The Gompers, Ishii and
Metrick’s Index is calculated by adding one point for each of 24 provisions, compiled by the
Investor Responsibility Research Center (IRRC), that restrict shareholder rights. Higher levels
of the index represent weaker shareholder rights. The results show that our previous findings
are robust to the alternative measures of governance we use. It is interesting to note that
governance in general not significant, with the exception of the specification based on the
fraction of institutional shareholders.
Overall, these results indicate that connections do indeed play a role in the investment
allocation process. This is consistent with our working hypothesis that trust favors a transfer of
resources to the more connected divisional managers. The next section complements this
evidence by looking at investment cash-flow sensitivities.
3.2 Connections and Investment-Cash Flow Sensitivity
We now investigate the degree to which connections influence the sensitivity of segment
investment to cash-flow. Previous papers (e.g. Shin and Stulz, 1998) showed that investment in
a given segment depends on the cash-flow generated by that segment as well as on the cash-
flow made available by the other segments of the firm (the sensitivity coefficients providing
clues on how the internal capital markets work). If connections affect the resource allocation,
we expect that connected segments bear less the brunt of their own cash-flow shortfalls, while
20
The standard deviation of Excess Connection is 0.068. Hence a one-standard deviation increase implies a 0.013 ×
0.068 = 0.088% increase in segment investment, a 0.088%/2% = 4.4% rise relative to the average segment
investment of 2% reported in Table 1.
13
at the same time being on the receiving end if the rest of the firm is performing well. To test
this intuition, we estimate
t,j,it,jt,j,it,j,it,j,it,j,it,j,it,j,i
ZXCSSCFCOSCFSSCFI εδµββγα +++×+++
ϕ
+=
1
, (2)
where investment is regressed on Same-Segment Cash-Flow (SSCF), Other Segment Cash-Flow
(OSCF, the sum of all other segment’s cash-flow apart from segment
i), Connection (denoted
as
C for short) and an interaction variable between SSCF and Connection (as well as a set of
control variables). The interaction coefficient
β
1
determines whether connections impact the
dependence of a segment on its own cash-flow. In addition, columns 4 through 6 repeat the
same analysis, but decomposing the segment cash-flow in cases when this cash-flow is positive
and when it is negative. That is, we estimate (dropping the subscripts)
εδµββ
βγα
+++×+×+
ϕ+ϕ+++=
−−++
−−++
ZXCSSCFCSSCF
SSCFSSCFCOSCFI
11
, (3)
where SSCF
+
(SSCF
—
) is a interaction variable that takes a value equal to SSCF if SSCF>0
(<0), and zero otherwise. These two variables are themselves interacted with Connection. The
coefficients
β
1
+
and β
1
—
give information on the case in which connections play a major role.
Each of these tests is repeated for each specification using firm fixed-effects. In the interest of
brevity, the coefficients of the control variables in the longer specifications are not shown.
The results in columns 1 through 3 of Panel A of Table 3 show that the interaction
coefficient
β
1
is statistically significant and negative. This result is robust across the three
specifications, with similar magnitudes and significance levels. This indicates that the impact of
cash-flow shortfalls is lower for segments that exhibit higher degrees of connection. In addition,
columns 4 to 6 show that this effect is statistically stronger when the segment has a negative
cash-flow. We interpret this as evidence that segments run by trusted divisional managers are
more likely to be ‘protected’ during bad times and hence not see their investment allocations
decreased.
To which extent does trust affect the investment sensitivity to the cash-flow of other
segments? Columns 1 through 3 of Panel B of table 3 present the results of estimating
εδµββ
γγβα
+++×+×+
++++ϕ+=
−−++
−−++
ZXCOSCFCOSCF
OSCFOSCFCSSCFI
22
(4)
where OSCF
+
(OSCF
—
) is a interaction variable that takes a value equal to OSCF if OSCF>0
(<0), and zero otherwise. These two variables are in turn interacted with Connection.
The results show that in general, segment investment is sensitive to the cash-flows
generated by the other segments of the firm, irrespective of whether these are positive or
14
negative (both coefficients γ
+
and γ
—
are positive and statistically significant). In contrast,
segments with connections are more likely to ‘receive’ allocations when the remaining of the
firm is enjoying good times. This can be seen by the coefficient
β
2
+
being positive and
significant, while the coefficient
β
2
—
is slightly negative but not statistically different from zero.
Finally, columns 4 to 6 of Panel B show the results of a similar specification,
εδµββ
γγβα
+++×+×+
++++ϕ+=
−−++
−−++
ZXCOSCFCOSCF
OSCFOSCFCSSCFI
cc
cccc
33
, (5)
where the interaction variables OSCF
+
and OSCF
—
are respectively replaced by variables that
take a value equal to OSCF if
Same-Segment Cash-Flow is positive (negative), and zero
otherwise. That is, these variables determine the investment sensitivity to the cash-flow of the
rest of the firm, but
conditional on whether the segment i itself is performing well.
The results show that the increased sensitivity of segment investment to OSCF for
connected segments is especially strong in the cases in which the segment is performing badly
(
β
3
+
positive and significant). In other words, when a segment is enduring bad times, it tends
to be sensitive to the cash-flow generated by the other segments, but more so if it enjoys the
trust of the CEO.
To summarize, we find that connectedness — trust — of managers of a segment tends to
decrease the impact of cash-flow shortfalls on their investment allocations. This decrease is
particularly important when the segment is going through ‘bad times’. Segments run by
trusted managers are more dependent on other segments’ cash-flow, particularly in ‘bad times’.
3.3 Connections and Transfer to Segments
As a final robustness check for our results, we directly look at the transfer of resources within
the firm. Rajan et al.
,
(2000) use the variable Transfer to determine to which extent the
segment is being attributed resources by the corporate headquarters. Given our focus on the
segment-specific connections, we replicate our regression analysis of section 4.1 with the
Transfer attributed to a segment (instead of the segment’s investment) as dependent variable:
t,j,it,jt,j,it,j,it,j,i
ZXCONNECTIONT εδµβα +++×+= . (6)
The results are reported in Table 4 (for readability all the coefficients have been multiplied
by 100). Transfer is negatively related to the importance of the segment for the firm and to the
firm’s age. The coefficient of the segment’s Q is insignificant, which indicates that the flow of
transfers is not consistently related to the investment opportunities of a segment. Connection is
again positive and significant, with coefficients ranging in magnitude between 0.024 and 0.029.
These coefficients imply that a one-standard deviation change in Connection induces a 9%
15
change in Transfers made available to the segment.
21
These results confirm that Connections
play a role in the internal capital market process.
4. The Connection Gap and Firm Performance
As a next step, we study the impact on firm value. We argued that the presence of trust
between the CEO and the managers of good segments may help to reduce the misallocation
due to “socialism” — i.e., the need to subsidize bad segments in order to lower the incentives of
the managers at the helm of the good segments to engage in defensive investments. This is due
to the higher ability of trusted CEOs to commit on ex-post reallocations as well as to the
higher ability of trusted divisional managers to stop the flow of resources to less productive
segments. Therefore, we expect that the more connected managers preside over good segments,
the lower is the transfer to the bad segments. This would have positive wealth implications.
We therefore first build a proxy for the relative position of connected managers at the top of
good and bad segments, then we study the transfer and finally we look at the value
implications at the firm level.
4.1 The Connection Gap
We want to see how connections affect the investment allocation process as a function of the
relative attractiveness of the segments where power is located. To investigate the importance of
differences in power between High Q and Low Q segments and how they impact the overall
firm, we define a new variable: the Connection Gap. This is the difference between the High Q
Segment Connection and the Low Q Segment Connection:
{}
{}
{}
{}
×−
×=
∑∑
<
<
>
>
Q
~
Q:i
j,i
Q
~
Q:ij
Q
~
Q:i
j,i
Q
~
Q:ij
j
ss
ss
ss
ss
Connection
N
Connection
N
GapConnection
11
Connection Gap measures the extent to which segments with valuable investment
opportunities are run by managers more trusted by the CEO relative to segments with weak
investment opportunities. A positive value of the Connection Gap means that High Q segments
have ‘the upper hand’ compared to Low Q segments.
Given that the impact on the firm of the differences in power between the two groups may
not be symmetric, we allow the Connection Gap to have a differential impact on firm
investment allocation and performance depending on who holds power — managers in the high
Q segments or managers in the low Q segments. We divide the Connection Gap into two
variables. Connection Gap
+
is equal to the difference between the High Q Segment Connection
21
Analogously, a one-standard deviation increase in Connection implies a 0.03 × 0.06 = 0.018% increase in Transfer,
a 0.018%/0.2% = 9% increase relative to the average segment Transfer of 0.2% reported in Table 1.
16
and the Low Q Segment Connection when this difference is positive, and zero otherwise.
Connection Gap
—
is equal to the negative of the difference between the High Q Segment
Connection and the Low Q Segment Connection when this difference is negative, and zero
otherwise:
22
0
>
+
×=
GapConnectionjj
GapConnectionGapConnection 1
0
<
−
×−=
GapConnectionjj
GapConnectionGapConnection 1
If connections reduce the misallocation of resources, we expect Connection Gap
+
to be
negatively related to the transfer to the bad segments.
4.2 Connection Gap and Internal Transfers
We regress the total transfers to Low Q segments on our measures of connection gap and a set
of control variables:
23
t,jt,jjt,j
WGapConnectionT ε+µ+β+α=
−
4
(7)
and
t,jt,jjjt,j
WGapConnectionGapConnectionT ε+µ+γ++β+α=
−+−
55
(8)
where T
–
j,t
is the (segment asset-weighted) sum of the transfers to the segments that operate in
segments with a Q lower than then the firm overall Imputed Q (Rajan et al.
,
2000). The set of
control variables is the same as before and it includes the measure of Diversity as defined in
Section 3.2. All the specifications contain firm-fixed effects specifications, industry dummies
and time-clustered robust standard errors.
The results are reported in Table 5. Columns 1 to 3 report the estimation results using the
linear Connection Gap variable while columns 4 to 6 use the Connection Gap breakdown into
positive and negative components.
Concerning our variables of interest, two main features are worth mentioning. The first is
that the transfers to the Low Q segments are related to the Connection Gap. The coefficient β
4
is positive and significant (with values ranging between 0.019 to 0.020, significant at 1%
level).
24
When we decompose this effect using Connection Gap
+
and Connection Gap
–
, we see
that most of the explanatory power comes from Connection Gap
+
. The coefficient β
5
is
negative and significant (
–0.026 coefficient, t-stats from 2.23 to 2.46), while the coefficient γ
5
is
positive but of lower magnitude and with lower levels of significance (t-values between 1.67
22
The switch of the sign of Connection Gap
—
allows a direct, intuitive interpretation of the coefficients.
23
This is done without any loss in generality since the total Transfers to High Q segments are the symmetric of the
total Transfers to Low Q segments.
24
Again all parameter values in the Table have been multiplied by 100 for clarity.
17
and 1.86). In other words, we find that companies where the connections of High-Q segments
are relatively stronger show lower transfers to Low-Q segments. These are firms in which a
positive connection gap displays a more efficient allocation of resources as the less productive
segments receive less transfers. This result is robust across the different specifications and
controlling for different sets of variables. The results are not only statistically significant but
also economically relevant. An increase of one standard deviation in the positive connection
gap within the firm corresponds to 0.2% (
–0.1%) increase (decrease) in the level of transfers to
the good (bad) segments.
25
Among the other variables, transfers to Low-Q segments are a positive function of the
number of segments and the amount of debt in the capital structure, and a negative function
of growth and firm age. The coefficient on diversity is positive but not statistically significant.
These findings support our working hypothesis that trust reduces the misallocation of
resources. We now move on the see whether this has value implications.
4.3 Connection Gap and Excess Value
We now test how connections affect the firm value. We regress the firm excess value on the
connections gap and a set of control variables
t,it,jt,it,it,j
WGapConnectionGapConnectionXV εµγβα
66
++++=
−+
, (9)
where excess value (XV
j,t
) is the difference between the firm’s Q and Imputed Q. We report the
results in Table 6. We use two definitions of excess value, the first (columns 1 to 3) based on
Villalonga (2004) and the second (columns 4 to 6) based on Lang and Stulz (1994).
26
The results show a strong positive correlation between Connection Gap
+
and firm value.
An increase of one standard deviation in the connection gap within the firm corresponds to a
2.3% absolute decrease in the average conglomerate discount, equivalent to a 13% decrease for
both Villalonga’s and Lang and Stulz’s definition of excess value. The coefficient of Connection
25
The standard deviation of Connection Gap
+
and Connection Gap
—
is 2.8% and 2.5%, respectively. Multiplying the
first of these by the coefficients reported in Table 4, leads to the 0.2% and –0.1% changes that are sizable given the
magnitude of the Transfer variable.
26
In this case, the excess value is constructed as follows. The numerator of Q is equal to liabilities (Compustat item
181) minus deferred taxes and investment credit (item 35) plus preferred stock (item 10, or 56, or 130, in that
order) plus market value of equity (item 25 times item 199). The denominator of Q is the sum of replacement value
of property, plant and equipment (PPE), inventory (INV), and other assets. The replacement value of PPE is equal
to the change in item 8 plus lag PPE×(1 −depreciation rate) × (1 + inflation rate). The depreciation rate is set at 5%
and inflation is given by the CPI series of the Federal Reserve’s FRED database. The replacement value of INV
depends on the accounting method: item 3 + (change in item 3) × (1 + inflation × 0.5) if LIFO; item 3 if FIFO;
(item 3) × (1 + inflation × 0.5) if average cost method is used; item 3 for all other methods. The accounting method
comes from item 59. The replacement value of other assets is the sum of items 1, 2, 31, 32, 33, 68 and 69.
18
Gap
—
is positive and sizeable, but not statistically significant. We interpret this finding as
evidence that trust between the CEO and the division managers at the helm of the less
productive segments helps to reduce the waste of resources. Most of the effect seems however
concentrated in the segments with good investment opportunities. These findings are robust
across different specifications and controlling for different sets of variables. Regarding the other
variables, Firm Size and Long-term Debt are negatively related to excess value, the latter
result being consistent with Mansi and Reeb (2002).
In Table 6, Panel B, we provide specifications in which we interact both Connection Gap
+
and Connection Gap
-
with the degree of diversity of the firm’s investment opportunities
(Diversity). We know that the wasteful allocation of resources due to the need of engaging in
subsidization of bad segments to prevents the managers at the top of good segments from
engaging in defensive investment is higher the higher the difference in investment opportunities
of the firm (Rajan, et al., 2000). If trust reduces the need for such value reducing activity, we
expect that the positive value impact of connections should be detectable only for firms with a
high degree of diversity. Therefore, the interaction between the degree of diversity of the firm’s
investment opportunities and connection gap should be significant only for firms with a high
degree of diversity. And indeed, this is the case. Across all the specifications it appears that the
interaction terms are statistically significant, in particular for the interaction with Connection
Gap
+
which is economically stronger than that of the interaction with the Connection Gap
—
.
Overall, the evidence presented in this section and in the previous one show that firms in
which High-Q managers are at the helm of good segments the transfers to the bad segments
required to prevent them from engaging in defensive investment are lower. To the extent that
the relationship between headquarters and segments is characterized by (at least some)
asymmetric information, these results would also be consistent with trust ameliorating the flow
of such information and improving the “congruence” (Dessein, 2002) between CEO and
divisional managers.
5. Further Evidence: Connections and Distance to Headquarters
As we mentioned above, the effectiveness of trust should be higher if the CEO and the
divisional managers are geographically distant. Indeed, the farther the CEO and the divisional
managers are, the more the CEO should trust the distant — and therefore less controllable —
managers to allocate them resources and the more the distant managers presiding good
segments should trust the CEO in order not to implement defensive investment schemes. We
now bring this intuition to the data.
We measure the distance as the log of the weighted average kilometric distance between
the headquarters and the subsidiaries belonging to the segment, where the weights are the
19
number of employees in each subsidiary. We create a High (Low) Distance dummy variable
that takes the value 1 if the segment is above (below) median in terms of distance and zero
otherwise. For the sake of brevity, we focus on our most important test, the test on segment
investment (equation 1) and that on excess value (equation 9).
Table 7 reports the results of re-estimating equation 1 splitting the Connection coefficient
into two interaction terms, where Connection multiplies the High Distance and Low Distance
dummies. The results show that connections have a positive impact on segment investment,
but that this impact is statistically stronger for segments far-away from headquarters. The
coefficients are also quite different, with the coefficient of the far-away segments’ interaction
with connection being about 4 times as large as the one of the low-distance interaction. It is
also worth mentioning that the distance variable is not significant by itself.
Table 8 presents the results of re-estimating equation 9. We split the connection gap terms
by interacting them with high and low distance dummies. These dummies now determine
which of the two types of segments (High-Q or Low-Q) are closer to the headquarters. That is,
the High (Low) Distance dummy takes the value 1 if the High Q (Low Q) segments are further
away than Low Q (High Q) segments to corporate headquarters, and zero otherwise. The
distance for a group of segments is the log of the weighted average distance in km between the
headquarters and the company’s subsidiaries, where the weights are the size of the segments to
which the subsidiaries belong times the weight of the subsidiary in terms of number of
employees within the segment.
The table shows that firms with positive connection gap have lower conglomerate
discounts, particularly in the case where the High-Q segments are further away from
headquarters than Low-Q segments. These findings support our working hypothesis. Indeed,
they consistently show that trust is more effective if the divisional managers are located further
from the CEO.
6 Controlling for Potential Endogeneity
The previous results may be subject to the objection that the location of the connected
managers can be endogenous. That is, CEOs may choose to appoint connected managers
members of their social network to the helm of the best/bigger segments. To address this issue,
we use an instrumental variable methodology.
We rely on a set of exogenous variables that explain the CEO’s appointment decision of a
divisional manager but are unrelated to the company or segment performance. We use
variables related to the weather conditions of the area in which the segments are located. Our
assumption is that when a CEO is considering filling a position within the organization, he will
(all else equal) appoint his ‘protégés’ to locations that have a higher quality of living. To the
20
extent that climate conditions affect quality of life, they should affect appointment decisions
while being uncorrelated with investment opportunities in a particular location.
We therefore collect a set of weather measures such as: the degree of humidity, the wind
speed, and the number of sunshine days in the year.
27
For the degree of humidity and wind
speed, we first define the average for each year across locations and then we calculate the
absolute difference with respect to it. The working hypothesis is that managers would prefer to
go to areas in which the weather conditions are milder. We augment this set of instruments
with a dummy taking the value of 1 is the regional headquarters is located in a remote city
and zero otherwise. We define a remote city a city that is more than 100 km miles away from
one of the top 20 major (in terms of population) cities in the US. The working hypothesis is
that managers would prefer to live closer to big cities. We also include a set of variables that
capture these characteristics (e.g., wind, humidity, days of sunshine, remote city) for the
corporate headquarters.
In the case of firm-level regressions, for each firm we instrument the difference in
connections between good and bad segments with the average differences in climate
characteristics of the areas in which the segments are located. Regarding the quality of the
instruments, (unreported) regressions show a strong correlation between segment connections
and climate characteristics.
28
Moreover, in terms of orthogonality, in each regression, the
Hansen’s J test of over-identification provides evidence of the lack of residual correlation of the
instruments with the second-stage residuals.
We use these instruments to re-do our most important tests.
29
We re-estimate the relation
between connections and segment investment (equation 1) and between connections and
segment transfers (equation 6), whose results are reported in Table 9, Panel A. They confirm
the previous findings, showing a strong positive correlation between a segment’s connection
and its investment. In Panel B, we report the results of the instrumental variable regression
concerning excess value. Also in this case, there is a strong positive correlation between a
positive connection gap and firm value. The magnitude of the coefficients is similar to the ones
in the previous tables. Overall, these findings show that the relationship between connections
and transfer and firm value is not due to spurious correlation induced by endogeneity. It is
instead the case that connections do indeed induce reallocations of resources within the firm
27
See http://www.census.gov/compendia/statab/geography_environment/weather_events_and_climate/.
28
The F-statistic values of regressions of Firm Connection or the Connection Gap on the instrument set range from
19 to 26, therefore above the rule-of-thumb of a F-statistic of 10 as an indicator of weak instruments.
29
Unfortunately our large sample makes it extremely cumbersome (bordering on impossible) to run a fixed-effects
IV. We therefore opted for a standard IV with time and industry dummies and standard errors clustered by firm.
21
that are value increasing only if the more connected segments happen to be the ones operating
in the more successful segments.
7. Conclusion
We study the role of commonalities between the CEO and divisional managers on the resource
allocation process. We focus on the informal links, or “connections”, that relate the CEO and
the divisional managers when they have the same educational and career background, or share
characteristics due to affiliation with the same generation inside the firm (they are in the same
age group, or they joined the firm at the same time, or got promoted to their current position
at the same time).
We argue that these features make it more likely that CEO and divisional managers share
similar beliefs tend to interpret information in a similar way as well as to belong to the same
power group within the firm. This creates trust that leads the CEO to channel more resources
to the segments run by the “similar” managers. We argue that trust may have a positive
impact on the value of the firm when it reduces the misallocation of resources induced by the
lobbying efforts of managers and the “socialist” allocation of capital across divisions (Rajan et
al, 2000).
We study this phenomenon by using a sample of US corporations for which we collected
biographical information on the cohort the CEOs and the divisional managers. We create a
connection index that is a sum of several characteristics that the CEO and divisional managers
may have in common: age, time they joined the company, type of education and career, time
they got appointed to the position they hold. We show that the segments run by connected
managers receive more investment and exhibit lower sensitivity to their own cash short-falls
(and more sensitivity to the other segments’ cash-flow) during bad times.
At the firm-level, having more connected managers presiding over high Q segments
improves the transfer of resources by preventing wasteful socialistic reallocation. This impact is
stronger in the case the pressure for a socialistic allocation is higher — i.e., firms characterized
by higher diversity in investment opportunities across divisions. The impact of trust is stronger
for segments located relatively far from the corporate headquarters.
These findings contribute to the literature on corporate governance, the theory of the firm
and the literature on internal capital markets. They provide a first look on the distribution of
power and connections within an organization and support Coase’s (1937) original intuition on
the role of power in affecting the decisions taken inside a hierarchy.
22
Appendix A: The merge of the datasets
To construct our proxy of Connection between the CEO and the divisional-managers, we
aggregate data from several sources. The first source is the result of the merge of the annual
CRSP-Compustat Merged (CCM) database files containing firm-level accounting data, and the
Compustat Segment files. The second source is Dun & Bradstreet’s (D&B) Million Dollar
Database, which contains information on location, number of employees, and manager identity
for more than 23 million U.S. companies and their subsidiaries. Observations in D&B are at
the “subsidiary” level. D&B also contains information on their Parents and Ultimate Parents
(headquarters). We collect data on the top subsidiaries per sales, approximately 33,000
observations per year.
To link the D&B subsidiaries to Compustat segments, we proceed as follows. First, we
match Compustat firms with all Ultimate Parents in D&B using a name-recognition algorithm.
Each Ultimate Parent will have several subsidiaries as “children”. Second, we match
Compustat segments with the children, or subsidiaries, of each Ultimate Parent. This match is
done sequentially by: 4-digit SIC code of the segment and of the subsidiary; 3-digit SIC code of
the segment and of the subsidiary; a keyword match between the segment’s Compustat name
and the subsidiary’s D&B name and business description. We repeat these two matching
procedures iteratively after checking manually for unaccounted parent-subsidiary relationships,
unmatched large firms due to differences in designation, etc. The quality of the matching is
very good: matching subsidiaries were found for 47% of the total number of Compustat
segments, representing 60% of segment asset value. Out of the remaining, about 26% of the
number of segments were “unusable” (they refer to corporate headquarters, have missing or
zero sales, or have missing segment SIC codes); 11% refer to companies with no information in
D&B; and 17% of segments could not be matched due to ambiguous or missing segment
business descriptions. In terms of asset value, these numbers are 32%, 0% and 9% respectively.
The second step of the merge links D&B manager biography information to Compustat
segments. We consider that a manager exhibits “biographical data” if we have at least
information on his year of birth and year of joining the company. Using this criterion, we have
data on about 31,959 individuals (middle managers and CEOs) working in D&B subsidiaries.
Then, we conduct pairwise comparisons between the Ultimate Parent CEO and the subsidiary
managers in terms of their biographical information. Each pairwise comparison consists of
creating a dummy variable that assumes the value 1 if a certain characteristic is common to
both CEO and a subsidiary manager (e.g. age group). Once the pairwise comparisons are done
at the subsidiary level, we aggregate them twice: first, we aggregate across all managers within
a given subsidiary; second, across all subsidiaries belonging to the same segment. The resulting
23
aggregate figures at the segment level (Connection by Age Cohort, Connection by Time of
Joining, etc.) are then averaged into an index of Connection as described in the main text.
Some remarks. First, it is sometimes the case that the headquarters themselves are
allocated to a given segment if there is evidence that the segment’s operations are effectively
taking place at the headquarters. This evidence is usually given by a large number of
employees working at the headquarters’ location as well as a perfect match between the
segment and the headquarters SIC code. Secondly, it is sometimes the case that the CEO is
referred to as part of the management team of the subsidiary. These cases are obviously not
taken into account when calculating the pairwise comparisons. This diminishes our sample
somewhat (by about one-fifth) because there are many cases in D&B where the subsidiary only
contains info about the (Ultimate Parent) CEO in its manager biography section.
24
References
Adams, Rene, and Daniel Ferreira, 2004, Gender diversity in the boardroom. ECGI Finance Working
Paper No. 57/2004.
Adams, Rene, Almeida Heitor and Daniel Ferreira, 2005, Powerful CEOs and their impact on corporate
performance, The Review of Financial Studies, 18, 1403-1432.
Aggarwal, R. and A. Samwick, 2003, “Why do managers diversify their firms? Agency reconsidered”,
Journal of Finance 58, 71-188.
Asquith, P., and D.W. Mullins, 1986, Equity issues and offering dilution, Journal of Financial Economics,
15, 61-89.
Berger, Allen N., and Gregory F. Udell, 1995, Relationship lending and lines of credit in small firm finance,
Journal of Business, 68, 351—382.
Berger, Phillip and Eli Ofek, 1995, Diversifications effect on firm value, Journal of Financial Economics
37, 39-65.
Bertrand, Marianne, and Antoinette Schoar, 2003, The Managing with Style: The Effect of Managers on
Corporate Policy, The Quarterly Journal of Economics, 118, 1169-1208.
Bolton, Patrick, and David Scharfstein, 1996, Optimal Debt Structure and the Number of Creditors, Journal
of Political
Economy, 104 (1), 1-25.
Boot, Arnoud, and Anjan Thakor. 2004. Disagreement and Flexibility: A Theory of Optimal Security
Issuance and Capital Structure. Mimeo.
Bower, J.,1970, “Managing the Resource Allocation Process.” Harvard Business School Press.
Campa, Jose Manuel, and Simi Kedia, 2002, Explaining the diversification discount, Journal of Finance
57 1731-1762.
Coase Ronald, 1937, The nature of the firm, Economica 4, 386-405.
Cohen, Lauren, Frazzini, Andrea. and Christopher Malloy, 2007, The small world of investing: board
connections and mutual fund returns, Working Paper.
Coval, Joshua D. and Tobias Moskowitz, 1999, Home bias at home: Local equity preference in domestic
portfolios, Journal of Finance 54, 2045-2073.
Coval, Joshua D. and Tobias Moskowitz, 2001, The geography of investment: Informed trading and asset
prices, Journal of Political Economy 109, 811-841.
Denis, David J., Diane Denis, and Atula Sarin 1997, Agency problems, equity ownership, and corporate
diversification, Journal of Finance 52, 135-160.
Denis, David J., Diane Denis, and K. Yost 2002, Global diversification, industrial diversification and
firm value, Journal of Finance 57, 1951-1979.
Dessein, Wouter 2002. Authority and Communication in Organizations. Review of Economic Studies 69,
811—838.
Faure-Grimaud, A., Gromb, D., 2004. Public trading and private incentives. Review of Financial Studies
17, 985-1014.
Gaspar, José-Miguel and Massimo Massa, 2007, Local ownership as private information: evidence on the
liquidity-monitoring trade-off, Journal of Financial Economics 83, 751-792.
Gertner, R., D. Scharfstein and J. Stein, 1994, Internal versus external capital markets, Quarterly
Journal of Economics 109, 1211-1230.
Gompers, Paul, Joy Ishii, and Andrew Metrick, 2003, Corporate governance and equity prices, Quarterly
Journal of
Economics 118, 107-155.
Graham, J., M. Lemmon and J. Wolf, 2002, “Does corporate diversification destroy value?” Journal of
Finance 57, 695-720.
25
Guiso Luigi, Sapienza, Paola, and Luigi Zingales, 2008, Trusting the Stock Market, Journal of Finance,
forthcoming.
Guner B., U. Malmendier and G. Tate, 2007, Financial expertise of directors, Journal of Financial
Economics, forthcoming.
Jensen, Michael, 1986, Agency costs of free cash flow, corporate finance and takeovers, American
Economic Review 76 659-665.
Kahn, Charles, and Andrew Winton, 1998, Ownership Structure, Speculation, and Shareholder Intervention,
Journal
of Finance, 53 (1), 99-129.
Khanna, N., and Tice, S., 2001, The bright side of internal capital markets, Journal of Finance 56, 1489-
1528.
Kuhnen, Camelia, M., 2005, Social Networks, Corporate Governance and Contracting in the Mutual
Fund Industry”, Working Paper.
Lamont, O.A. and C. Polk, 2002, “Does diversification destroy value? evidence from the industry
shocks”, Journal of Financial Economics 63, 51-77.
Lamont, Owen, 1997, Cash flow and investment: Evidence from internal capital markets, Journal of
Finance 52: 83-109.
Landier, Augustin, David Thesmar and David Sraer, 2004, Bottom-up corporate governance, Mimeo.
Lang, Larry, and Renee Stulz 1994, Tobin's q, corporate diversification, and firm performance, Journal
of Political Economy 102, 1248-1291.
Lindenberg, Eric, and Stephen Ross, 1981, Tobin’s q and industrial organization, Journal of Business 54,
1-32.
Maksimovic, Vojislav and Gordon Phillips, 2002, Do conglomerate firms allocate resources inefficiently
across industries? theory and evidence, Journal of Finance 57 721 — 767.
Malloy, Christopher J., 2005, The geography of equity analysis, Journal of Finance 60, 719-755.
Malmendier, U. and G. Tate, 2005a, CEO Overconfidence and Corporate Investment, Journal of Finance
60 (6), 2661-2700.
Malmendier, U. and G. Tate, 2005b, Does overconfidence affect corporate investment? Ceo
overconfidence measures revisited, European Financial Management 11, 649-659.
Mansi, Sattar A. and David Reeb, 2002, Corporate diversification: what gets discounted? Journal of
Finance 57, 2167 — 2183.
Matsusaka, John and, Vickram Nanda, 1997, Internal capital markets and corporate refocusing,
Working paper, University of Southern California.
Maug, E., 1998. Large shareholders as monitors: is there a trade-off between liquidity and control?.
Journal of Finance 53, 65-98.
Meyer, Margaret, Milgrom, Paul, and John Roberts, 1992, Organizational prospects, influence costs, and
ownership changes, Journal of Economics and Management Strategy 1, 9-35.
Myers, Stewart, 1977 Determinants of corporate borrowing, Journal of Financial Economics 5, 147-175.
Myers, S.C., and N.S. Majluf, 1984, Corporate financing and investment decisions when firms have
information that investors do not have, Journal of Financial Economics, 13, 187-221.
Noe, T., 2002. Investor activism and financial market structure. Review of Financial Studies 15, 289-318.
Ozbas, Oguzhan and David S. Scharfstein, 2007, Evidence on the Dark Side of Internal Capital Markets,
Working Paper.
Petersen, Mitchell, 2007, Estimating standard errors in finance panel data sets: comparing approaches,
Northwestern University mimeo.
Rajan, Raghuram, and Luigi Zingales, 1995, Power struggles and inefficiency, Working paper, University
of Chicago.
26
Rajan, Raghuram and Luigi Zingales, 1998, Power in a theory of the firm, Quarterly Journal of
Economics 113, 387-432.
Rajan, Raghuram, Henri Servaes, and, Luigi Zingales, 2000, The cost of diversity: The diversification
discount and inefficient investment, Journal of Finance, 40, 25-80.
Scharfstein David S., 1997, The dark side of internal capital markets II, Working paper, MIT.
Scharfstein, David S. and J. Stein, 2000, The dark side of internal capital markets: segment rent-seeking
and inefficient investment, Journal of Finance 55, 2537-2564.
Shin, Hyun-Han, and Renee Stulz, 1998, Are internal capital markets efficient?, Quarterly Journal of
Economics 113, 531-553.
Stein, Jeremy, 1997, Internal capital market and the competition for corporate resources, Journal of
Finance 52, 111-133.
Stulz, René, 1990, Managerial discretion and optimal financing policies, Journal of Financial Economics
26: 3-27.
Subrahmanyam, Avanidhar, 2006, Social Networks and Corporate Governance, UCLA working paper.
Whited, Toni, 2001, Is it inefficient investment that causes the diversification discount, Journal of
Finance 56, 19667-1691.
Villalonga, Belén, 2004a, Does Diversification Cause the "Diversification Discount"?, Financial
Management 33, 5-24.
Villalonga, Belén, 2004b, Diversification discount or premium? new evidence from the business
information tracking series, Journal of Finance 59, 479-506.
Wulf, Julie, 2002, Internal capital markets and firm-level compensation incentives for divisional
managers, Journal of Labor Economics 20, 219-262.
Wulf, Julie, 2005, Influence and inefficiency in the internal capital market, Wharton School Working
paper.
Xuan, Yuhai, 2006, Empire-building or bridge-building? Evidence from new CEOs' internal capital
allocation decisions, Harvard Business School working paper.
Zingales, Luigi, 2000, In search of new foundations, Journal of Finance 55, 1623-1653.
Zhu, Ning, 2002, The local bias of individual investors, Yale SOM working paper 02-30.
27
Table 1
Sample Characteristics
This table presents summary statistics for the sample used in this study. The sample includes multiple-segment
firms in the merged CRSP-Compustat Segments universe for the period 1996-2004 that have non-missing segment
SIC codes, and non-missing values for market capitalization, segment sales, segment assets, diversity variables, and
manager biographical data. We remove securities with share codes different from 10 or 11, as well as financial
companies and utilities. We also remove firms with negative book equity, with sales lower than 20 million USD, and
whose sum of segment sales are more than 1 percent away from total firm sales reported in Compustat. The firm-
level variables can be described as follows. Sales growth is the percentage growth in sales (Compustat item 12) from
the past year. Number of Segments is the number of business lines reported in Compustat segment data. Tobin’s Q
is defined as the ratio between market value and book value of assets (e.g. Villalonga, 2004). The numerator of
Tobin’s Q (Fama and French, 2002) is equal to liabilities (item 181) minus deferred taxes and investment credit
(item 35) plus preferred stock (item 10, or 56, or 130, in that order) plus market value of equity (item 25 times item
199). The denominator is total assets (item 6). Segment Q, denoted Q
i
ss
, is the average Q of all single-segment firms
in segment i’s SIC code. We use 4-digit SIC codes as long as there are at least 3 single-segment firms in that SIC
code; if not, we use 3-digit, and so forth. Imputed Q is the weighted sum of the Q of the individual segments
∑
∈
=≡
j
Ni
ss
ij,i
QwQ
ˆ
Q Imputed
where Nj is the set of segments belonging to firm j, w
i,j
is the weight of segment i on the firm j total assets, and Q
i
ss
is the average Q of all single-segment firms in segment i’s SIC code. Excess Value is the difference between the
firm’s Q and Imputed Q. Diversity (Rajan, Servaes and Zingales, 2000) is defined as
(
)
∑
∑
∈
∈
−−
=
j
j
Ni
j
ss
i
Ni
j
ss
ij,i
ss
ij,i
NQ
NQwQw 1
Diversity
2
where we used the shorthand N
j
to denote the number of segments of firm j. Long-Term Debt and Cash are
respectively, long-term liabilities (item 9) and cash (item 1) over total assets. Firm Age is the number of years since
the company entered the CRSP database. The Weights of High Q (Low Q) Segments is the sum of weights w
i,j
for
segments that have Q
i
ss
respectively above and below Imputed Q. Investment (at the segment level) is segment
investment divided by the firm beginning-of-period total assets. Transfer (at the segment level) is defined as
∑
∈
−−
−=
j
Ni
ss
j,i
j,i
j,i
j,i
j,i
ss
j,i
j,i
j,i
j,i
A
I
A
I
w
A
I
A
I
Transfer
where I
i,j
refers to segment investment and A
i,j
to the segment’s assets. Transfer (at the firm level) is the segment
asset-weighted sum of the Transfers to High Q and Low Q segments. Connection (at the segment level) is the
segment average of the four Connection variables, described as follows. Connection by Time of Joining (at the
segment level) is the proportion of CEO-Subsidiary manager pairs in a given segment for which the CEO and the
manager entered the firm within 2 years of each other. Connection by Age Cohort (at the segment level) is the
proportion of CEO-Subsidiary manager pairs for which the CEO’s and the manager’s age are within 4 years of each
other. Connection by Time of Appointment (at the segment level) is the proportion of pairs for which the CEO and
the manager reached their current management position within 2 years of each other. Connection by Same Career
(at the segment level) is the proportion of pairs for which the CEO and the manager have a similar career
background (in Finance, Marketing, or Engineering). Firm-level Connection is the asset-weighted average
Connection of the firm’s segments. Connection of High Q (Low Q) Segments is the average Connection of segments
that have Q
i
ss
above (below) Imputed Q. Same Segment Cash-Flow is segment i’s cash-flow divided by firm
beginning-of-period total assets. Other Segment Cash-Flow is sum of cashflow accruing to all segments apart from
segment i, again divided by beginning-of-period total assets. Number of Subsidiaries is the number of D&B
subsidiaries matched to a given Compustat segment. Number of Managers is the number of managers identified in
the D&B database and matched to a given segment; it is therefore the denominator over which all the Connection
pairings were summed.
28
Table 1
Sample Characteristics (cont.)
Variable N Mean
Std.
Dev. Q1 Median Q3
Segment-level Assets 10,459 952 2,736 50 196 710
Sales 10,459 1,017 2,844 63 236 822
Weight of Segment 10,459 0.34 0.26 0.13 0.28 0.52
Investment 10,459 0.02 0.04 0.00 0.01 0.02
Transfer × 100 10,459 0.20 6.75 -1.82 -0.03 1.62
Same Segment Cash-Flow 10,459 0.05 0.07 0.01 0.03 0.08
Other Segment Cash-Flow 10,459 0.08 0.08 0.04 0.08 0.12
Segment Q 10,459 1.78 1.00 1.13 1.49 2.06
Connection 10,459 0.14 0.17 0.00 0.08 0.22
Connection by Time of Joining 10,459 0.15 0.24 0.00 0.01 0.29
Connection by Age Cohort 10,459 0.18 0.23 0.00 0.06 0.33
Connection by Time of Appointment' 10,459 0.15 0.23 0.00 0.01 0.25
Connection by Same Career 10,459 0.06 0.12 0.00 0.01 0.10
Number of Subsidiaries 10,459 2.38 3.02 1.00 1.00 2.00
Number of Managers 10,459 8.38 9.81 3.00 6.00 9.00
Firm-level Sales 5,224 3,291 11,728 209 640 2,108
Sales growth 5,224 0.16 1.02 -0.02 0.07 0.20
Number of Segments 5,224 2.98 1.15 2.00 3.00 3.00
Tobin’s Q 5,224 1.60 0.99 1.03 1.31 1.81
Imputed Q 5,224 1.78 0.93 1.18 1.52 2.08
Excess Value 5,224 -0.19 1.13 -0.61 -0.19 0.21
Diversity 5,224 0.27 0.19 0.13 0.22 0.38
Long-term Debt 5,224 0.21 0.17 0.06 0.20 0.32
Cash 5,224 0.09 0.13 0.01 0.04 0.12
Firm age 5,224 21.8 18.3 7.0 17.0 31.0
Weight of Segments
Segment Q > Inputed Q 5,224 0.42 0.34 0.08 0.37 0.72
Segment Q = Inputed Q 5,224 0.13 0.34 0.00 0.00 0.00
Segment Q < Inputed Q 5,224 0.45 0.35 0.10 0.45 0.78
Transfer to Segments × 100
Segment Q > Inputed Q 5,224 -0.01 1.78 -0.21 0.00 0.13
Segment Q < Inputed Q 5,224 0.01 1.81 -0.13 0.00 0.22
Average Connection 5,224 0.14 0.17 0.00 0.08 0.21
Connection of segments
Segment Q > Inputed Q 5,224 0.08 0.11 0.00 0.03 0.12
Segment Q < Inputed Q 5,224 0.08 0.11 0.00 0.03 0.12
29
Table 2
Investment and Connection
This table presents least-squares estimates of the relation between Segment Investment and Connection level. The left-hand side is segment investment normalized by beginning-of-
period firm assets. The right-hand side variables can be described as follows. Same Segment Cash-Flow is segment i’s cash-flow divided by beginning-of-period firm assets. Other
Segment Cash-Flow is the total cash-flow accruing to all segments of the firm apart from segment i, divided by beginning-of-period total assets. Segment Q is the average Q of all
single-segment firms in segment i’s SIC code, where Q is the ratio between market value and book value of assets (e.g. Villalonga, 2004). The numerator (Fama and French, 2002) is
equal to liabilities (item 181) minus deferred taxes and investment credit (item 35) plus preferred stock (item 10, or 56, or 130, in that order) plus market value of equity (item 25
times item 199). The denominator is total assets (item 6). Weight of Segment is the weight of segment i on the firm’s total assets. Log Sales is the natural logarithm of Computat sales
(data item 12). Sales growth is the percentage growth in sales from the past year. Connection is the segment average of the four individual Connection variables: Connection by Time
of Joining (the proportion of CEO-Subsidiary manager pairs in a given segment for which the CEO and the manager entered the firm within 2 years of each other); Connection by Age
Cohort (the proportion of CEO-Subsidiary manager pairs for which the CEO’s and the manager’s age are within 4 years of each other); Connection by Time of Appointment (the
proportion of pairs for which the CEO and the manager reached their current management position within 2 years of each other); and Connection by Same Career (the proportion of
pairs for which the CEO and the manager have a similar career background in Finance, Marketing, or Engineering). Connection is interacted with Same Segment Cash-Flow+ and
Same Segment Cash-Flow— . Firm Age is the number of years since the company entered the CRSP database. Diversity (Rajan, Servaes and Zingales, 2000) is defined as
(
)
∑
∑
∈
∈
−−
=
j
j
Ni
j
ss
i
Ni
j
ss
ij,i
ss
ij,i
NQ
NQwQw 1
Diversity
2
where we used the shorthand N
j
to denote the number of segments of firm j. Imputed Q is the weighted sum of the Q of the individual segments
∑
∈
=≡
j
Ni
ss
ij,i
QwQ
ˆ
Q Imputed
where Nj is the set of segments belonging to firm j, w
i,j
is the weight of segment i on the firm j total assets, and Q
i
ss
is the average Q of all single-segment firms in segment i’s SIC code.
Long-Term Debt and Cash are respectively, long-term liabilities (item 9) and cash (item 1) over total assets. Number of Segments is the number of business lines reported in
Compustat segment data. Connection Controls are meant to control for education and specific career background of managers. Connection by Engineering/Marketing/Finance Career
is the proportion of CEO-Subsidiary manager pairs in a given segment for which the CEO and the manager have both respectively an Engineering/Marketing/Finance background.
Connection by MBA/PhD is the proportion of CEO-Subsidiary manager pairs for which the CEO and the manager have both an MBA or a PhD degree. Connection by Ivy League is
the proportion of CEO-Subsidiary manager pairs for which the CEO and the manager have a degree from an Ivy League University. Intra-Manager Connections are calculated in a
similar way as the original Connection variables, with the difference that the proportion is calculated over the number of possible pairs between managers of a segment (therefore
excluding the CEO). In columns 4, 5, and 6, we add control variables related to quality of governance. Institutional ownership (Block holdings) is the percentage of the firm held by
institutional investors (institutional investors holding blocks of stock of 5% or more), as reported in Spectrum 13F. Governance Index is the index of shareholder rights from Gompers,
Ishii and Metrick (2003), where higher levels of the index denote worse corporate governance. All specifications contain industry dummies and firm fixed effects. T-statistics are
calculated using robust clustered (by year) standard errors. The symbols ***,**,* denote significance levels of 1%, 5% and 10%, respectively, for the two-tailed hypothesis test that the
coefficient equals zero.
30
Table 2
Investment and Connection (cont.)
Dependent Variable: Segment Investment
(1) (2) (3) (4) (5) (6)
Coeff. t-stat. Coeff. t-stat. Coeff. t-stat. Coeff. t-stat. Coeff. t-stat. Coeff. t-stat.
Intercept 0.120 5.34 *** 0.113 4.69 *** 0.115 4.35 *** 0.028 0.86 0.034 1.06 0.09 8.60 ***
Same Segment Cash-Flow 0.115 7.14 *** 0.111 7.09 *** 0.111 7.14 *** 0.111 7.07 *** 0.111 6.95 *** 0.124 8.36 ***
Other Segment Cash-Flow 0.058 5.86 *** 0.055 5.70 *** 0.055 5.71 *** 0.055 6.75 *** 0.055 5.54 *** 0.015 1.49
Segment Q 0.002 5.88 *** 0.001 4.28 *** 0.001 4.44 *** 0.001 4.39 *** 0.001 4.86 *** 0 1.46
Weight of Segment 0.050 8.94 *** 0.050 8.93 *** 0.050 8.89 *** 0.051 9.21 *** 0.05 8.85 *** 0.041 9.55 ***
Log Sales -0.009 -4.89 *** -0.007 -4.05 *** -0.007 -4.02 *** -0.006 -3.82 *** -0.007 -4.08 *** -0.003 -3.37 ***
Sales growth 0.001 2.71 *** 0.001 2.11 ** 0.001 2.14 ** 0.001 2.07 ** 0.001 2.12 ** 0.006 3.38 ***
Connection 0.012 2.99 *** 0.012 2.80 *** 0.015 3.33 *** 0.014 2.88 *** 0.015 3.33 *** 0.011 3.53 ***
Institutional Ownership -0.01 -3.19 ***
Block holdings 00.03
Governance Index 0 -0.31
Firm age 0.000 -1.87 * 0.000 -1.77 * 0 -1.67 * 0 -1.78 * 0 -2.90 ***
Diversity 0.002 0.63 0.001 0.43 0.001 0.38 0.001 0.45 -0.002 -0.96
Imputed Q 0.002 3.26 *** 0.003 3.52 *** 0.003 3.90 *** 0.003 3.49 *** 0.002 4.43 ***
Long-term Debt -0.019 -3.19 *** -0.020 -3.28 *** -0.02 -3.48 *** -0.02 -3.25 *** -0.021 -4.77 ***
Cash 0.007 0.90 0.006 0.88 0.01 1.42 0.006 0.88 0 -0.08
Number of Segments 0.000 0.19 0.000 0.24 0 0.24 0 0.27 0 -0.29
Connection Controls
by Engineering Career -0.003 -0.87 -0.003 -0.97 -0.003 -0.89 -0.002 -0.66
by MBA 0.001 0.50 0.002 0.70 0.001 0.51 0.003 1.01
by Marketing Career 0.002 1.00 0.003 1.10 0.002 1.02 0.002 0.49
by PhD 0.010 2.87 *** 0.01 2.87 *** 0.01 2.77 *** 0.006 0.75
by Finance Career -0.002 -0.80 -0.002 -0.76 -0.002 -0.81 -0.001 -0.48
by Ivy League -0.010 -1.49 -0.009 -1.46 -0.01 -1.49 -0.006 -1.32
Intra Manager Connections
by Time of Joining -0.006 -1.48 -0.007 -1.57 -0.006 -1.48 -0.006 -2.85 ***
by Age Cohort 0.006 0.67 0.005 0.62 0.006 0.67 0.004 0.79
Time of Appointment' 0.003 0.44 0.005 0.76 0.003 0.45 0.006 1.98 **
by Same Career -0.001 -0.42 -0.001 -0.39 -0.001 -0.42 -0.001 -0.43
N 10459 10459 10459 10247 10439 6369
Adjusted R2 0.44 0.44 0.45 0.44 0.44 0.51
31
Table 3
Investment Cash-Flow Sensitivities and Connection
This table presents least-squares estimates of Investment-Cash-Flow sensitivities conditional on Connection level and segment performance. The left-hand side is segment investment
normalized by beginning-of-period firm assets. Right-hand side variables are similar to the ones used in Table 2 (please refer to its caption for a complete description of all the control
variables), apart from the following modifications. In columns 1 to 3 of Panel A, Connection is interacted with Same-Segment Cash-Flow (SSCF). In columns 4 to 6 of Panel A, we
remove SSCF from the specification and introduce SSCF
+
(SSCF
—
), interaction variable that takes a value equal to SSCF if SSCF>0 (<0), and zero otherwise. These two variables are
themselves interacted with Connection. In columns 1 to 3 of Panel B, we remove Other Segment Cash-Flow (OSCF) from the specification and replace it with OSCF
+
(OSCF
—
), an
interaction variable that takes a value equal to OSCF if OSCF>0 (<0), and zero otherwise. These two variables are in turn interacted with Connection. In columns 4 to 6 of Panel B,
we remove Other Segment Cash-Flow (OSCF) from the specification and replace it with OSCF
+
(OSCF
—
), an interaction variable that takes a value equal to OSCF if SSCF>0 (<0),
and zero otherwise. These two variables are in turn interacted with Connection. All specifications contain industry dummies and firm fixed-effects. T-statistics are calculated using
robust clustered (by year) standard errors. The symbols ***,**,* denote significance levels of 1%, 5% and 10%, respectively, for the two-tailed hypothesis test that the coefficient
equals zero.
32
Table 3
Investment Cash-Flow Sensitivities and Connection (cont.)
Panel A.
Dependent Variable: Segment Investment
(1) (2) (3) (4) (5) (6)
Coeff. t-stat. Coeff. t-stat. Coeff. t-stat. Coeff. t-stat. Coeff. t-stat. Coeff. t-stat.
Intercept 0.119 5.24 *** 0.112 4.58 *** 0.114 4.28 *** 0.117 5.44 *** 0.110 4.67 *** 0.112 4.33 ***
Same Segment Cash-Flow (SSCF) 0.137 8.15 *** 0.134 7.88 *** 0.134 7.96 ***
Other Segment Cash-Flow (OSCF) 0.058 5.70 *** 0.056 5.56 *** 0.056 5.56 *** 0.050 4.82 *** 0.048 4.79 *** 0.048 4.82 ***
Connection 0.006 1.78 * 0.006 2.00 ** 0.007 1.82 * 0.000 0.03 0.000 0.05 0.000 -0.05
SSCF x Connection -0.135 -2.25 ** -0.135 -2.20 ** -0.134 -2.17 **
SSCF
+
0.176 10.45 *** 0.173 10.21 *** 0.172 10.30 ***
SSCF
-
0.029 0.90 0.028 0.88 0.028 0.90
SSCF
+
x Connection -0.062 -1.02 -0.060 -0.99 -0.058 -0.97
SSCF
-
x Connection -0.429 -2.19 ** -0.433 -2.19 ** -0.434 -2.21 **
Firm Characteristic Controls No Yes Yes No Yes Yes
Connection Controls No No Yes No No Yes
N 10459 10459 10459 10459 10459 10459
Adjusted R2 0.44 0.44 0.44 0.45 0.45 0.45
Panel B.
Dependent Variable: Segment Investment
(1) (2) (3) (4) (5) (6)
Coeff. t-stat. Coeff. t-stat. Coeff. t-stat. Coeff. t-stat. Coeff. t-stat. Coeff. t-stat.
Intercept 0.128 5.90 *** 0.121 5.22 *** 0.123 4.85 *** 0.124 5.63 *** 0.118 4.98 *** 0.120 4.63 ***
Same Segment Cash-Flow 0.116 7.69 *** 0.113 7.65 *** 0.113 7.73 *** 0.122 7.57 *** 0.120 7.50 *** 0.120 7.59 ***
Connection 0.002 6.42 *** 0.001 4.52 *** 0.001 4.66 *** 0.002 6.23 *** 0.001 4.51 *** 0.001 4.65 ***
OSCF
+
0.026 2.01 ** 0.022 1.80 * 0.022 1.78 *
OSCF
-
0.088 3.24 *** 0.086 3.14 *** 0.086 3.16 ***
OSCF
+
x Connection 0.168 2.07 ** 0.170 2.10 ** 0.173 2.09 **
OSCF
-
x Connection -0.024 -0.25 -0.015 -0.17 -0.014 -0.15
OSCF conditional on SSCF
+
(OSCFc
+
) 0.037 3.37 *** 0.034 3.18 *** 0.034 3.18 ***
OSCF conditional on SSCF
-
(OSCFc
-
) 0.053 2.99 *** 0.051 2.96 *** 0.051 2.96 ***
(OSCFc
+
) x Connection 0.066 1.63 0.069 1.66 * 0.072 1.69 *
(OSCFc
-
) x Connection 0.216 1.98 ** 0.220 2.05 ** 0.220 2.05 **
Firm Characteristic Controls No Yes Yes No Yes Yes
Connection Controls No No Yes No No Yes
N 10459 10459 10459 10459 10459 10459
Adjusted R2 0.44 0.44 0.46 0.44 0.44 0.47
33
Table 4
Transfer and Connection
This table presents least-squares estimates of the relation between Segment Transfer and Connection level. For purposes of readability all coefficients have been multiplied by 100. The
left-hand side is defined as
∑
∈
−−
−=
j
Ni
ss
j,i
j,i
j,i
j,i
j,i
ss
j,i
j,i
j,i
j,i
A
I
A
I
w
A
I
A
I
Transfer
where I
i,j
refers to segment investment and A
i,j
to the segment’s assets. Right-hand side variables are similar to the ones used in Table 2 (please refer to its caption for a complete
description of all the control variables). All specifications contain industry dummies and firm fixed effects. T-statistics are calculated using robust clustered (by year) standard errors.
The symbols ***,**,* denote significance levels of 1%, 5% and 10%, respectively, for the two-tailed hypothesis test that the coefficient equals zero.
34
Table 4
Transfer and Connection (cont.)
Dependent Variable: Segment Transfer
(1) (2) (3)
Coeff. t-stat. Coeff. t-stat. Coeff. t-stat.
Intercept -1.31 -0.43 -1.35 -0.5 -1.10 -0.40
Same Segment Cash-Flow -2.30 -1.05 -2.50 -1.14 -2.50 -1.14
Other Segment Cash-Flow -1.95 -1.25 -2.01 -1.24 -2.04 -1.24
Segment Q -0.10 -0.52 -0.20 -0.82 -0.27 -0.82
Weight of Segment -1.40 -4.03 *** -1.30 -3.87 *** -1.40 -3.96 ***
Log Sales -0.28 -0.75 0.00 -0.03 0.00 -0.09
Sales growth -0.10 -0.77 -0.20 -1.52 -0.10 -1.39
Connection 2.38 3.213 *** 2.29 3.099 *** 2.31 2.948 ***
Firm age -0.10 -2.56 ** -0.10 -2.69 ***
Diversity 0.60 0.92 0.50 0.86
Imputed Q 0.30 1.37 0.30 1.38
Long-term Debt 1.20 1.29 1.20 1.28
Cash 0.81 0.90 0.83 0.84
Number of Segments 0.10 0.63 0.10 0.63
Connection Controls
by Engineering Career -0.60 -1.41
by MBA 0.00 0.01
by Marketing Career 0.42 0.60
by PhD 1.90 1.15
by Finance Career 0.20 0.35
by Ivy League 0.40 0.32
Intra Manager Connections
by Time of Joining -0.50 -0.26
by Age Cohort -0.60 -0.46
Time of Appointment' 1.52 3.609 ***
by Same Career 0.10 0.171
N 10457 10457 10457
Adjusted R2 0.04 0.04 0.06
35
Table 5
Transfers to Segments and Connection Gap
This table presents least-squares estimates of the relation between the Transfers to Low Q segments and the Connection Gap between High Q and Low Q segments. For purposes of
readability all coefficients have been multiplied by 100. The left-hand side is the sum of the Transfers to High Q (Low Q) segments, where transfer is defined as
∑
∈
−−
−=
j
Ni
ss
j,i
j,i
j,i
j,i
j,i
ss
j,i
j,i
j,i
j,i
A
I
A
I
w
A
I
A
I
Transfer
where I
i,j
refers to segment investment and A
j
is the firm beginning-of-period total assets. Connection Gap is the difference in the average Connection between High Q and Low Q
segments. Connection Gap
+
(Connection Gap
—
) is equal to the (negative of the) difference in the average Connection of High Q and Low Q segments when this difference is positive
(negative), and zero otherwise. For definitions of all the remaining variables, please see the captions to Tables 1 and 2. All specifications contain industry dummies and firm-fixed
effects. T-statistics are calculated using robust clustered (by year) standard errors. The symbols ***,**,* denote significance levels of 1%, 5% and 10%, respectively, for the two-tailed
hypothesis test that the coefficient equals zero.
36
Table 5
Transfers to Segments and Connection Gap (cont.)
Dependent Variable: Transfer to Low Q Segments
(1) (2) (3) (4) (5) (6)
Coeff. t-stat. Coeff. t-stat. Coeff. t-stat. Coeff. t-stat. Coeff. t-stat. Coeff. t-stat.
Intercept -2.18 -0.84 0.69 0.37 -2.47 -0.92 -2.20 -0.84 0.64 0.34 -2.48 -0.92
Log Sales 0.04 0.22 0.00 0.02 0.01 0.07 0.05 0.23 0.01 0.03 0.02 0.09
Sales growth -0.04 -1.78 * -0.04 -1.81 * -0.05 -1.93 * -0.04 -1.77 * -0.04 -1.80 * -0.05 -1.92 *
Firm age -0.01 -1.82 * -0.02 -2.14 ** -0.02 -2.09 ** -0.01 -1.78 * -0.02 -2.12 ** -0.02 -2.07 **
Diversity 0.29 1.00 0.36 1.21 0.35 1.15 0.29 1.02 0.36 1.23 0.35 1.18
Imputed Q 0.06 0.74 0.06 0.71 0.05 0.61 0.06 0.76 0.06 0.73 0.05 0.64
Weight of High Q Segments 0.05 0.64 0.04 0.64 0.05 0.74 0.04 0.61 0.04 0.60 0.05 0.71
Firm-Level Connection 0.13 1.33 0.11 0.98 -0.03 -0.16 0.14 1.31 0.11 0.99 -0.01 -0.05
Connection Gap -1.97 -3.08 *** -1.94 -3.05 *** -2.01 -3.21 ***
Connection Gap
+
-2.53 -2.23 ** -2.54 -2.35 ** -2.67 -2.46 **
Connection Gap
—
1.52 1.74 * 1.46 1.67 * 1.49 1.86 *
Long-term Debt 0.76 2.40 ** 0.73 2.47 ** 0.76 2.41 ** 0.73 2.48 **
Cash -0.23 -0.34 -0.24 -0.36 -0.24 -0.35 -0.25 -0.37
Number of Segments 0.08 1.96 * 0.09 2.02 ** 0.08 1.98 ** 0.09 2.04 **
Connection Controls
by Engineering Career -0.49 -2.06 ** -0.50 -2.02 **
by MBA 0.30 0.85 0.32 0.92
by Marketing Career 0.40 0.86 0.39 0.85
by PhD 1.17 0.47 1.16 0.47
by Finance Career 0.22 0.76 0.23 0.80
by Ivy League 0.75 0.93 0.75 0.93
Intra Manager Connections
by Time of Joining 0.26 0.41 0.26 0.42
by Age Cohort 1.35 1.25 1.34 1.24
Time of Appointment' -1.05 -2.39 ** -1.07 -2.41 **
by Same Career -0.07 -0.15 -0.07 -0.15
N 5224 5224 5224 5224 5224 5224
Adjusted R2 0.09 0.09 0.11 0.09 0.09 0.12
37
Table 6
Excess Value and Connection Gap
This table presents least-squares estimates of the relation between Excess Value and the Connection Gap between High Q and Low Q segments. The left-hand side is Excess value, the
difference between the firm’s Q and its imputed Q. Connection Gap
+
(Connection Gap
—
) is equal to the (negative of the) difference in the average Connection of High Q and Low Q
segments when this difference is positive (negative), and zero otherwise. For details and definitions of all the remaining variables, please refer to the captions to Tables 1 and 2. The
symbols ***,**,* denote significance levels of 1%, 5% and 10%, respectively, for the two-tailed hypothesis test that the coefficient equals zero. For columns 1 to 3, the excess value is
constructed as in Villalonga (2004). In columns 4 to 6, the Excess Value is calculated using the Q as in Lang and Stulz (1994). Panel B is similar to Panel A except that we interact
Connection Gap
+
and Connection Gap
—
with Diversity. All specifications contain industry dummies and firm-fixed effects. T-statistics are calculated using robust clustered (by year)
standard errors. The symbols ***,**,* denote significance levels of 1%, 5% and 10%, respectively, for the two-tailed hypothesis test that the coefficient equals zero.
38
Table 6
Excess Value and Connection Gap (cont.)
Panel A.
Dependent Variable: Excess value LS Excess Value
(1) (2) (3) (4) (5) (6)
Coeff. t-stat. Coeff. t-stat. Coeff. t-stat. Coeff. t-stat. Coeff. t-stat. Coeff. t-stat.
Intercept 2.218 2.94 *** 2.263 2.96 *** 2.074 2.66 *** 2.260 2.76 *** 2.304 2.77 *** 2.109 2.50 **
Log Sales -0.294 -4.60 *** -0.286 -4.06 *** -0.276 -3.99 *** -0.319 -4.47 *** -0.310 -4.00 *** -0.299 -3.97 ***
Sales growth 0.001 0.09 0.001 0.06 0.001 0.07 0.005 0.46 0.004 0.44 0.004 0.45
Firm age -0.008 -0.64 -0.007 -0.55 -0.006 -0.52 -0.007 -0.55 -0.005 -0.44 -0.005 -0.42
Diversity -0.121 -0.65 -0.141 -0.71 -0.141 -0.72 -0.125 -0.65 -0.147 -0.72 -0.147 -0.73
Imputed Q -0.163 -1.63 -0.163 -1.62 -0.168 -1.68 * -0.166 -1.68 * -0.167 -1.67 * -0.172 -1.74 *
Weight of High Q Segments -0.031 -0.54 -0.032 -0.54 -0.029 -0.47 -0.028 -0.49 -0.030 -0.50 -0.026 -0.42
Firm-Level Connection 0.257 1.05 0.268 1.09 0.075 0.33 0.270 1.05 0.282 1.09 0.072 0.32
Connection Gap
+
0.842 1.97 ** 0.840 1.92 * 0.929 2.37 ** 0.870 1.98 ** 0.867 1.92 * 0.955 2.40 **
Connection Gap
—
0.919 1.10 0.931 1.11 1.025 1.26 0.974 1.09 0.987 1.10 1.085 1.25
Long-term Debt -0.172 -1.68 * -0.191 -1.92 * -0.198 -1.98 ** -0.218 -2.27 **
Cash -0.110 -0.44 -0.093 -0.38 -0.123 -0.46 -0.106 -0.40
Number of Segments -0.029 -1.16 -0.029 -1.20 -0.033 -1.34 -0.033 -1.38
Connection Controls
by Engineering Career 0.259 1.30 0.262 1.34
by MBA -0.076 -0.34 -0.069 -0.29
by Marketing Career -0.071 -1.07 -0.064 -0.89
by PhD -0.063 -0.17 -0.126 -0.35
by Finance Career 0.089 0.42 0.101 0.46
by Ivy League 0.829 10.88 *** 0.881 11.33 ***
Intra Manager Connections
by Time of Joining -0.531 -1.80 * -0.513 -1.51
by Age Cohort 0.664 2.46 ** 0.699 2.64 ***
Time of Appointment' -0.274 -1.23 -0.272 -1.30
by Same Career -0.076 -0.74 -0.084 -0.74
N 5224 5224 5224 5203 5203 5203
Adjusted R2 0.57 0.57 0.57 0.57 0.57 0.57
39
Table 6
Excess Value and Connection Gap (cont.)
Panel B.
Dependent Variable: Excess value LS Excess Value
(1) (2) (3) (4) (5) (6)
Coeff. t-stat. Coeff. t-stat. Coeff. t-stat. Coeff. t-stat. Coeff. t-stat. Coeff. t-stat.
Intercept 2.237 3.05 *** 1.869 2.50 ** 2.093 2.75 *** 2.276 2.84 *** 2.318 2.84 *** 2.125 2.57 **
Log Sales -0.291 -4.72 *** -0.284 -4.18 *** -0.273 -4.11 *** -0.316 -4.58 *** -0.307 -4.10 *** -0.296 -4.08 ***
Sales growth 0.000 0.05 0.000 0.03 0.000 0.03 0.004 0.42 0.004 0.40 0.004 0.42
Firm age -0.008 -0.67 -0.007 -0.58 -0.007 -0.55 -0.007 -0.58 -0.006 -0.48 -0.006 -0.45
Diversity -0.219 -1.28 -0.237 -1.28 -0.239 -1.31 -0.221 -1.25 -0.241 -1.27 -0.243 -1.29
Imputed Q -0.168 -1.71 * -0.168 -1.70 * -0.173 -1.77 * -0.171 -1.76 * -0.172 -1.75 * -0.178 -1.83 *
Weight of High Q Segments -0.035 -0.63 -0.036 -0.63 -0.032 -0.55 -0.032 -0.57 -0.033 -0.58 -0.029 -0.50
Firm-Level Connection 0.270 1.14 0.280 1.17 0.078 0.35 0.283 1.13 0.294 1.17 0.076 0.34
Connection Gap
+
-0.961 -1.15 -0.960 -1.14 -0.906 -1.09 -0.938 -1.16 -0.938 -1.15 -0.888 -1.09
Connection Gap
—
-0.625 -0.95 -0.583 -0.91 -0.540 -0.91 -0.487 -0.72 -0.439 -0.66 -0.394 -0.64
Connection Gap
+
x Diversity 6.173 2.51 ** 6.161 2.49 ** 6.286 2.70 *** 6.188 2.53 ** 6.172 2.50 ** 6.316 2.73 ***
Connection Gap
—
x Diversity 5.605 1.87 * 5.488 1.84 * 5.675 1.87 * 5.290 1.74 * 5.155 1.71 * 5.351 1.74 *
Long-term Debt -0.174 -1.77 * -0.192 -2.01 ** -0.199 -2.11 ** -0.219 -2.40 **
Cash -0.103 -0.41 -0.086 -0.35 -0.117 -0.44 -0.100 -0.38
Number of Segments -0.027 -1.09 -0.027 -1.12 -0.031 -1.27 -0.031 -1.30
Connection Controls
by Engineering Career 0.266 1.35 0.270 1.39
by MBA -0.069 -0.31 -0.062 -0.27
by Marketing Career -0.078 -1.14 -0.071 -0.94
by PhD -0.062 -0.17 -0.125 -0.35
by Finance Career 0.116 0.58 0.128 0.61
by Ivy League 0.821 12.04 *** 0.872 12.48 ***
Intra Manager Connections
by Time of Joining -0.533 -1.87 * -0.517 -1.56
by Age Cohort 0.674 2.52 ** 0.709 2.69 ***
Time of Appointment' -0.275 -1.21 -0.272 -1.27
by Same Career -0.078 -0.74 -0.085 -0.74
N 5224 5224 5224 5203 5203 5203
Adjusted R2 0.57 0.57 0.57 0.57 0.57 0.57
40
Table 7
Distance to Headquarters, Connection and Investment
This table presents least-squares estimates of Investment-Connection relation conditional on geographical distance between the segment subsidiaries and the company’s headquarters.
The left-hand side is segment investment normalized by beginning-of-period firm assets. Right-hand side variables are similar to the ones used in Table 2 (please refer to its caption for
a complete description of all the control variables), apart from the following modifications. We create High (Low) Distance dummy variables that take the value 1 if the segment is
above (below) median in terms of distance from the headquarters and zero otherwise. Connection × High (Low) Distance is an interaction variable that multiplies Connection and the
High (Low) Distance dummies. All specifications contain industry dummies and firm-fixed effects. T-statistics are calculated using robust clustered (by year) standard errors. The
symbols ***,**,* denote significance levels of 1%, 5% and 10%, respectively, for the two-tailed hypothesis test that the coefficient equals zero.
41
Table 7
Distance to Headquarters, Connection and Investment (cont.)
Dependent Variable: Segment Investment
(1) (2) (3)
Coeff. t-stat. Coeff. t-stat. Coeff. t-stat.
Intercept -0.189 -14.62 *** 0.024 0.747 -0.193 -12.97 ***
Same Segment Cash-Flow 0.116 7.10 *** 0.113 7.07 *** 0.113 7.11 ***
Other Segment Cash-Flow 0.058 5.19 *** 0.055 5.04 *** 0.055 5.05 ***
Segment Q 0.002 5.94 *** 0.001 4.00 *** 0.001 4.21 ***
Weight of Segment 0.050 8.94 *** 0.050 8.95 *** 0.050 8.91 ***
Log Sales -0.008 -4.77 *** -0.007 -3.93 *** -0.007 -3.91 ***
Sales growth 0.001 2.54 ** 0.001 1.97 ** 0.001 2.00 **
Connection x High Distance 0.016 4.09 *** 0.014 3.57 *** 0.017 3.96 ***
Connection x Low Distance 0.004 0.76 0.006 1.05 0.011 1.91 *
Distance -0.001 -0.72 -0.001 -0.87 -0.001 -0.96
Firm age 0.000 -1.75 * 0.000 -1.65 *
Diversity 0.000 0.16 0.000 -0.02
Imputed Q 0.003 3.78 *** 0.003 4.07 ***
Long-term Debt -0.020 -3.02 *** -0.020