Beruflich Dokumente
Kultur Dokumente
Robert Day School of Economics and Finance, Claremont McKenna College, 500 E 9th St, Claremont, CA 91711, USA
Fisher College of Business, The Ohio State University, 2100 Neil Ave, Columbus, OH 43210, USA
Kelley School of Business, Indiana University, 1309 East Tenth Street, Bloomington, IN 47405, USA
a r t i c l e in f o
abstract
Article history:
Received 23 August 2010
Received in revised form
11 February 2011
Accepted 1 March 2011
Available online 16 August 2011
We nd that rms behave consistently with how their CEOs behave personally in the
context of leverage choices. Analyzing data on CEOs leverage in their most recent
primary home purchases, we nd a positive, economically relevant, robust relation
between corporate and personal leverage in the cross-section and when examining CEO
turnovers. The results are consistent with an endogenous matching of CEOs to rms
based on preferences, as well as with CEOs imprinting their personal preferences on the
rms they manage, particularly when governance is weaker. Besides enhancing our
understanding of the determinants of corporate capital structures, the broader contribution of the paper is to show that CEOs personal behavior can, in part, explain
corporate nancial behavior of the rms they manage.
& 2011 Elsevier B.V. All rights reserved.
JEL classication:
G30
G32
G34
Keywords:
Corporate nance
Behavioral consistency theory
CEO personal leverage
Corporate leverage
1. Introduction
Since the start of modern capital structure research
with the seminal work of Modigliani and Miller (1958),
$
We are thankful for comments by Sumit Agarwal, Brent Ambrose, Zahi Ben-David, Tim Burch, Jeff Coles, Harry DeAngelo, Werner DeBondt, Rudi
Fahlenbrach, Mark Flannery, David Funder, Nikolay Halov, Eric Helland, Eric Hughson, Danling Jiang, Kose John, Andrew Karolyi, Jon Karpoff, Sandy Klasa,
Alok Kumar, Tim Loughran, Angie Low, Lisa Meulbroek, Enrico Perotti, Richard Rosen, Tony Sanders, Bill Schwert (the editor), Hersh Shefrin, Janet Smith,
0304-405X/$ - see front matter & 2011 Elsevier B.V. All rights reserved.
doi:10.1016/j.jneco.2011.08.005
21
1
Borghans, Duckworth, Heckman, and Weel (2008) is a very
informative overview of the economics of personal characteristics, and
they conclude: There is a lot of room for cooperation and exchange of
ndings and methods between personality psychology and economics.
(p. 84). We view our paper as an attempt to engage in such exchange of
methods and ndings.
2
Later in this paper, we review related empirical studies. But, it is
worth pointing out that there also exists theoretical research which
(footnote continued)
incorporates heterogeneity in CEOs personal characteristics into models
of corporate capital structure decisions. For example, Cadenillas,
Cvitanic, and Zapatero (2004) model the relation between managerial
risk aversion and leverage, while Hackbarth (2008) models the relation
with optimism or overcondence.
3
Seminal references include Allport (1937, 1966), Epstein (1979,
1980), and Funder and Colvin (1991).
22
4
A positive relation may not be supported if behaviors are situation-specic (e.g., Mischel, 1968; Slovic, 1972a,b; Endler and
Magnusson, 1976).
5
Liu and Yermack (2007) nd that rm performance deteriorates
when CEOs acquire very large mansions, but unlike our paper, they do
not examine the relation between personal leverage and corporate
capital structures.
6
We measure corporate leverage in year 2004 while personal home
leverage is measured at the time of the most recent primary home
purchase, which at the median is ve years earlier, reducing concerns
that an omitted contemporaneous variable such as mortgage/interest
rates jointly explains both personal and corporate leverage.
23
24
50
40
30
20
10
0
Corporate leverage
measured
11
Following the approach employed in other work on capital
structure (e.g., Lemmon, Roberts, and Zender, 2008), we winsorize and
truncate leverage at 1.0.
25
Table 1
Summary statistics.
The table reports summary statistics for a sample of 605 S&P 1,500 rms. Panels A and B display statistics relating to the nancing of primary residences by
CEOs and Panels C and D display statistics of rm-level leverage measures and control variables. In Panel A, HomeLev is determined at the time of the CEO
home purchase, and is computed as the mortgage divided by the purchase price of the home. If a mortgage is found but if any one of the mortgage amount,
purchase price, or the improvement cost (if the home is new construction) is unavailable, then HomeLev is set to missing. Statistics for HomeLev are for the
unconditional sample, and for HomeLev9Mort the reported sample statistics are conditional on the CEO using a mortgage to nance the home. Mortgage
amount is the sum of the rst and second mortgages at the time of the CEOs home purchase. Panel B reports the percent of the sample and number of
observations that use mortgage nance in the purchase of their primary residence (Mortgage usage at purchase), that use either a mortgage at the time of the
purchase or debt nancing on their home at some point in time (Home leverage usage), and for which there is no public record that the CEO ever used debt
(Never use leverage). The U.S. median data are tabulated from 2005 data provided by the Federal Housing Finance BoardPeriodic Summary Tables and the
2005 American Community Survey Subject Tables. Mortgage amounts are displayed in 2005 home price dollars. Values are adjusted using the Ofce of Federal
Housing Enterprise Oversights National Home Price Index. In Panel C, the corporate leverage variables are total debt to market value of assets (TDM) and total
debt to book value of assets (TDA). All debt measures are computed as of the end of the calendar year 2004. In Panel D, the control variables are market-to-book
ratio (Mktbk), the log of total assets (Assets), protability (Prot), tangibility of assets (Tang), and median industry leverage (IndusLev). All control variables are
computed as of the end of the calendar year 2003, i.e., with a lag of one year compared to the corporate leverage measures. ExecuComp MEAN and MEDIAN
values are calculated from 1,351 U.S.-based, non-nancial, and non-utility rms covered by ExecuComp in 2004. Stars in the ExecuComp MEAN and MEDIAN
columns denote signicance levels from testing the difference between the sample mean (median) and the ExecuComp sample mean (median) using a t-test
(Wilcoxon Mann-Whitney test). Signicance levels are denoted by n, nn, nnn, which correspond to 10%, 5%, and 1% levels, respectively. All leverage and control
variables are computed from S&Ps Compustat database. Detailed denitions are found in Appendix B.
Panel A
MED
HomeLev
HomeLev9Mort
Mortgage amount ($1000s)
MEAN
STD
MIN
MAX
0.47
0.66
0.40
0.63
0.35
0.21
0.00
0.01
1.00
1.00
1,047.00
1,233.00
973.00
54.00
8,626.00
N
608
385
430.00
U.S. MED
0.75
212.00
Panel B
66.0
73.8
22.0
642
642
642
Panel C
TDM
TDA
MEAN
STD
10th
Percentile
50th
90th
MEAN
0.179
0.205
0.198
0.188
0.000
0.000
0.126
0.185
0.425
0.415
605
605
0.178
0.205
1.769
7.119
0.113
0.247
0.154
1.241
1.606
0.139
0.183
0.153
0.710
5.164
0.009
0.050
0.004
1.428
6.950
0.121
0.201
0.104
3.134
9.349
0.240
0.515
0.369
605
605
605
605
605
1.796
7.140
0.112
0.268nn
0.166n
ExecuComp
MEDIAN
0.130
0.187
Panel D
Mktbk
Assets
Prot
Tang
IndusLev
xed effects to account for any differences across purchase years in legislation and market conditions not
picked up by mortgage rates and market returns.
Can differences in personal taxes explain the variation
in personal home leverage? This does not seem to be the
case for several reasons. First, the tax code in the U.S.
allows married (single) taxpayers to deduct interest on
home mortgages up to $1 million ($500,000) on up to two
homes. Out of the mortgages in our database, only 9.6%
are exactly $1 million. Only 11.7% of the CEOs have 100%
HomeLev if their home purchase price is below $1 million
or a $1 million mortgage if it is above the tax deductability threshold. Second, in column 8 of Table 2 we
control for the ratio of a CEOs total compensation which
is not tax deferrable (TaxIncRatio), i.e., salary and other
cash compensation (e.g., bonus) divided by total compensation. CEOs with a larger proportion of their
1.366
6.961
0.124
0.210
0.134nn
14
The number of observations is reduced in columns 8 and 9
because we require data on the CEOs compensation at the time of the
home purchase. In several cases, such data are missing because the CEO
purchased the home at a time when the CEO was not a top-executive
covered by ExecuComp or before the start date of the ExecuComp
database.
26
Table 2
Determinants of personal leverage.
The table reports the coefcients and standard errors from regressing HomeLev on determinants of personal leverage. The sample is non-nancial S&P
1,500 rms during 2004. HomeLev is the ratio of mortgage value to purchase price used by the rms CEO in his most recent primary home purchase.
PurAge is the age of the CEO at the time of his home purchase. PurAfterCeo is a dummy variable that is equal to one if the home was purchased after the
purchaser became CEO. LnMedHmVal is the natural logarithm of the median home value in the county in which the CEOs primary residence is located.
County-level median home value data are obtained from the 2005 American Community Survey. MortRate30 is the prevailing 30-year conventional xed
mortgage rate in the month and year of the CEOs home purchase. Data on monthly mortgage rates are obtained from the Federal Reserve Economic
Database series MORTG. MktRet 5yr is the ve-year annualized return of the value-weighted Center for Research on Security Prices (CRSP) index ending on
the last day of the month prior to the CEOs home purchase. TaxIncRatio is the ratio of CEO compensation for which the CEO cannot defer the tax liability.
LnCashComp is the natural logarithm of the total cash compensation (ExecuComp data item TOTAL CURR) of the CEO in the year of the home purchase
adjusted to 2005 dollars. This compensation includes salary plus bonuses. It is computed as the CEOs salary plus bonus divided by total compensation in
the year of the home purchase (ExecuComp items TOTAL CURR / TDC1). The table reports White (1980) heteroskedasticity-consistent standard errors.
Signicance levels are denoted by n, nn, nnn, which correspond to 10%, 5%, and 1% levels, respectively.
(1)
PurAge
(2)
(3)
(4)
(5)
0.0001
(0.0017)
PurAfterCeo
0.0444
(0.0311)
0.0679nnn
(0.0254)
LnMedHmVal
0.1355
(0.1664)
MktRet 5yr
(7)
(8)
0.0048
(0.0022)
0.0663nn
(0.0337)
0.0671n
(0.0355)
0.0672nn
(0.0261)
6.0716nnn
(0.7285)
4.8663
(4.3941)
0.4475nn
(0.1792)
0.2542
(0.6674)
TaxIncRatio
0.0053
(0.0638)
0.0344n
(0.0193)
LnCashComp
Intercept
AdjR2
N
Fixed effects
(9)
nn
0.0032
(0.0020)
0.0746nnn
(0.0247)
3.9221nnn
(0.6008)
MortRate30
(6)
0.3981nnn
(0.0848)
0.4138nnn
(0.0166)
0.0017
605
No
0.0018
605
No
0.4558
(0.3231)
0.0102
605
No
0.7008nnn
(0.0496)
0.3844nnn
(0.0240)
0.0402
605
No
0.0006
605
No
4. Empirical evidence
In this section, we examine whether personal leverage is
related to corporate leverage, i.e., we test the empirical
predictionsbehavioral consistency versus hedgingfrom
Section 2.
0.0463
(0.3441)
0.0785
605
No
0.2323
(0.4758)
0.1010
605
PurYear
0.4248nnnn
(0.0314)
0.6552nnn
(0.1321)
0.0027
364
No
0.0053
392
No
In column 2, we include lagged rm-level characteristics as control variables: the market-to-book ratio
(Mktbk) as a measure of growth opportunities, the log of
total assets (Assets) measuring rm size, protability
(Prot), and the tangibility of the rms assets (Tang) as
a measure of collateral. We choose this set of controls to
follow Frank and Goyal (2009a).16 In column 3, we control
for industry leverage by including IndusLev, the median
total debt to market value of assets ratio in the rms
industry, following Frank and Goyal (2009a). In column 4,
which we label our baseline model for corporate leverage, we include all these controls at the same time. The
rm-level control variables have the expected signs. Most
importantly, we nd that the estimated coefcient on
HomeLev is still positive (0.0718) and statistically signicant at the 1%-level.
16
Frank and Goyal (2009a) explore the relative importance of a very
large set of potential determinants of corporate leverage. We include the
controls that they conclude are the most reliable determinants of
corporate leverage. In untabulated regressions, we also checked that
our results are robust to the inclusion of other controls. For example, we
included Sales instead of Assets, a different collateral measure (inventory
plus net property, plant, and equipment scaled by assets) instead of
Tang, and we included Z-score by Altman (1968).
27
4.1.3. Non-linearity
A non-linear relation may mask support for the hedging hypothesis. It may be that only CEOs with the highest
home leverage choose to countervail their personal leverage through corporate capital structure decisions. That is,
we may nd an inverse relation between personal and
corporate debt, but only for the CEOs who are the most
highly levered. We choose an 80% cutoff to dene high
personal leverage because of the standard in the U.S.
mortgage industry related to down payments. Specically, we dene HL80 to be an indicator variable that is
one if HomeLev 40:80, and zero otherwise. However,
column 6 shows that there is no evidence of CEOs offsetting their personal leverage by changing their rms
leverage in a countervailing way, not even among the
CEOs who are the most highly levered.
19
28
Table 3
CEO personal and corporate leverage.
Panel A of the table reports coefcients and standard errors from regressing the total debt to market value of assets of the rm in 2004 (TDM) on
determinants of capital structure, using OLS estimation. Control variables are constructed using 2003 data and dened as in Appendix B. The sample is
non-nancial S&P 1,500 rms. HomeLev is dened as the ratio of mortgage value to purchase price used by the rms CEO in his most recent primary
home purchase. Column 5 includes industry xed effects by two-digit SIC code. HL80 is a dummy variable that equals one if HomeLev4 0:80, and zero
otherwise. HomeLevPredict and HomeLevRes are the predicted and residual series from regression 6 in Table 2. In column 8 we estimate the baseline
regression using home leverage data for all previous CEOs during three years prior to 2004, in total 84 observations. HomeLevPrev is the home leverage of
these previous CEOs. The dependent variable in this regression is TDM during the last year of tenure of the previous CEO and control variables are lagged
by one year. Panel B of the table displays the incremental increase in adjusted-R-square caused by adding each of the determinants of capital structure to
a model which includes only industry xed effects by two-digit SIC code. The dependent variable in Panel B is TDM. The table reports White (1980)
heteroskedasticity-consistent standard errors. Signicance levels are denoted by n, nn, nnn, which correspond to 10%, 5%, and 1% levels, respectively.
Panel A
(1)
HomeLev
0.0632nnn
(0.0226)
(2)
(3)
0.0781nnn
(0.0192)
0.0666nnn
(0.0192)
(4)
0.0718nnn
(0.0179)
(5)
(6)
0.0784nnn
(0.0180)
HL80
(7)
(8)
0.0631nnn
(0.0202)
0.0662
(0.4122)
HL80 HomeLev
0.0963
(0.4388)
HomeLevPredict
0.0419
(0.0576)
0.0828nnn
(0.0190)
HomeLevRes
0.1285nn
(0.0592)
HomeLevPrev
Mktbk
0.0454nnn
(0.0076)
0.0281nnn
(0.0058)
0.0346nnn
(0.0072)
0.0279nnn
(0.0058)
0.0273nnn
(0.0057)
Assets
0.0284nnn
(0.0049)
0.0199nnn
(0.0045)
0.0202nnn
(0.0047)
0.0201nnn
(0.0046)
0.0204nnn
(0.0045)
Prot
0.2616nnn
(0.0846)
0.2691nnn
(0.0683)
0.2011nn
(0.0887)
0.2727nnn
(0.0687)
0.2721nnn
(0.0664)
0.2668nn
(0.1265)
Tang
0.2458nnn
(0.0452)
0.0659
(0.0468)
0.1374nn
(0.0598)
0.0659
(0.0468)
0.0624
(0.0469)
0.0138
(0.1294)
0.5604nnn
(0.0675)
0.5562nnn
(0.0666)
0.5934nnn
(0.1453)
0.7247nnn
(0.0555)
IndusLev
Intercept
AdjR2
N
Fixed effects
0.1535nnn
(0.0102)
0.0106
605
No
0.0409nnn
(0.0109)
0.0054
(0.0409)
0.2890
605
No
0.3230
605
No
0.5607nnn
(0.0670)
0.0136
(0.0352)
0.4160
605
No
0.0388
(0.0975)
0.0134
(0.0354)
0.4220
605
Indus
0.0286
(0.0408)
0.4150
605
No
0.4190
605
No
0.0265n
(0.0134)
0.0425nnn
(0.0153)
0.1375
(0.1004)
0.4460
84
No
Panel B
(1)
(2)
(3)
(4)
(5)
0.0742nnn
(0.0198)
HomeLev
0.0462nnn
(0.0066)
Mktbk
0.0190nnn
(0.0051)
Assets
0.2094nnn
(0.0665)
Prot
0.1624nnn
(0.0627)
Tang
Intercept
AdjR2
0.1401
(0.1036)
0.1123
(0.0835)
0.1709
(0.1069)
0.0051
(0.1057)
0.1501
(0.1112)
0.0946
(0.1078)
0.3000
0.3160
0.0160
0.3760
0.0760
0.3190
0.0190
0.3200
0.0200
0.3100
0.0100
605
Indus
605
Indus
605
Indus
605
Indus
605
Indus
605
Indus
DAdjR2
N
Fixed effects
(6)
146
150
103
105
100
Frequency
29
67
50
48 47
54
18
11
0
0.00
0.01
0.02
0.03
0.04
0.05
0
0.06
30
Table 4
Effects of persistence in corporate leverage.
The table reports coefcients and standard errors from regressing the
total debt to market value of assets of the rm in 2004 (TDM) on
determinants of capital structure, using OLS estimation. Control variables are constructed using 2003 data and dened as in Table 1 of the
paper. The sample is non-nancial S&P 1,500 rms. HomeLev is dened
as the ratio of mortgage value to purchase price used by the rms CEO
in his most recent primary home purchase. FirmLeveragei is rm
leverage lagged by i years. The table reports White (1980) heteroskedasticity-consistent standard errors. Signicance levels are denoted by
n nn nnn
, ,
, which correspond to 10%, 5%, and 1% levels, respectively.
(1)
HomeLev
0.0838nnn
(0.0241)
FirmLeverage10
0.3262nnn
(0.0725)
(2)
0.0697nnn
(0.0175)
(3)
0.0245n
(0.0144)
0.4159nnn
(0.0512)
FirmLeverage5
0.7218nnn
(0.0842)
FirmLeverage1
Mktbk
0.0316nnn
(0.0082)
0.0138nnn
(0.0053)
0.0013
(0.0028)
Assets
0.0092
(0.0061)
0.0185nnn
(0.0044)
0.0013
(0.0029)
Prot
0.3074nnn
(0.0912)
0.3213nnn
(0.0702)
0.0781nn
(0.0360)
Tang
0.0872
(0.0568)
0.0590
(0.0449)
0.0089
(0.0230)
IndusLev
0.4012nnn
(0.0861)
0.3522nnn
(0.0669)
0.1307nn
(0.0595)
Intercept
0.0432
(0.0518)
AdjR2
N
0.4920
322
0.0683n
(0.0369)
0.5540
504
0.0111
(0.0212)
0.7880
605
22
31
Table 5
Zero personal leverage and alternative measures.
The table reports coefcients and standard errors from regressing the total debt to market value of assets of the rm in 2004 (TDM) on determinants of
capital structure, using OLS estimation. Control variables are constructed using 2003 data and dened as in Appendix B. The sample is non-nancial S&P
1,500 rms. ZeroPersLev is an indicator variable that equals one if there is no public record that the CEO ever used debt, and zero otherwise. Mort is a
dummy variable that takes a value of one if the CEO uses a mortgage to nance the purchase of his home and takes a value of zero otherwise. MortRe is
an indicator variable that equals one if there is evidence that the CEO uses a mortgage at the time of purchase or some time other than the time of
purchase for his primary residence, and zero otherwise. LnMortAmt is the natural logarithm of the real value of the total mortgage amount used by the
CEO in his most recent home purchase. LnPurPrice is the natural logarithm of the real purchase price of the CEOs most recent primary home purchase.
Real mortgage values and purchase prices are computed in 2005 home price dollars using the Ofce of Federal Housing Enterprise Oversights National
Home Price Index. TotHomeLev is the value-weighted average of the LTV ratios of all homes owned by the CEO simultaneously to his primary residence
(including the primary residence) and was calculated for a random sample of 100 CEOs from the 2004 sample. We are able to compute TotHomeLev for 92
of the 100 random CEOs because some own homes in states where data are not complete. The table reports White (1980) heteroskedasticity-consistent
standard errors. Signicance levels are denoted by n, nn, nnn, which correspond to 10%, 5%, and 1% levels, respectively.
(1)
ZeroPersLev
(2)
(3)
(4)
0.0494nnn
(0.0138)
0.0464nnn
(0.0124)
Mort
0.0512nnn
(0.0131)
MortRe
0.0033nnn
(0.0009)
0.0080
(0.0074)
LnMortAmt
LnPurPrice
0.0281nnn
(0.0059)
0.0177nnn
(0.0045)
0.2712nnn
(0.0690)
0.0713
(0.0475)
0.5729nnn
(0.0672)
0.0381
(0.0334)
0.0288nnn
(0.0059)
0.0184nnn
(0.0045)
0.2656nnn
(0.0694)
0.0671
(0.0472)
0.5657nnn
(0.0666)
0.0050
(0.0355)
0.0283nnn
(0.0059)
0.0180nnn
(0.0044)
0.2708nnn
(0.0690)
0.0700
(0.0474)
0.5711nnn
(0.0669)
0.0115
(0.0356)
0.0282nnn
(0.0059)
0.0194nnn
(0.0045)
0.2683nnn
(0.0691)
0.0643
(0.0476)
0.5641nnn
(0.0669)
0.1035
(0.1043)
0.0983n
(0.0518)
0.0158
(0.0158)
0.0265nnn
(0.0091)
0.1046
(0.0715)
0.1131
(0.0905)
0.6867nnn
(0.1694)
0.1348n
(0.0757)
0.4120
605
0.4130
605
0.4140
605
0.4110
605
0.5220
92
TotHomeLev
Mktbk
Assets
Prot
Tang
IndusLev
Intercept
AdjR2
N
(5)
32
Table 6
Timing of personal leverage choices.
The table reports coefcients and standard errors from regressing
different measures of corporate leverage computed in 2004 on determinants of capital structure, using OLS estimation. Control variables are
constructed using 2003 data and dened as in Appendix B. The sample is
non-nancial S&P 1,500 rms. HomeLev is dened as the ratio of
mortgage value to purchase price used by the rms CEO in his most
recent primary home purchase. Column 1 reports regression results
using observations in which the CEOs most recent home purchase was
made prior to becoming CEO. Column 2 reports regression results using
observations in which the CEOs most recent home was purchased prior
to 1999. Column 3 reports regression results using observations in
which the CEOs most recent home was purchased after 1998. The table
reports White (1980) heteroskedasticity-consistent standard errors.
Signicance levels are denoted by n, nn, nnn, which correspond to 10%,
5%, and 1% levels, respectively.
Prior to
being CEO
(1)
Earlier
leverage choices
(2)
Recent
leverage choices
(3)
0.0647nnn
(0.0226)
0.0471n
(0.0263)
0.0882nnn
(0.0260)
Mktbk
0.0275nnn
(0.0070)
0.0196nnn
(0.0067)
0.0450nnn
(0.0095)
Assets
0.0162nnn
(0.0054)
0.0188nnn
(0.0058)
0.0210nnn
(0.0069)
Prot
0.2623nnn
(0.0796)
0.3217nnn
(0.0771)
0.1311
(0.0817)
Tang
0.0885
(0.0563)
0.0082
(0.0617)
0.1233n
(0.0736)
IndusLev
0.6258nnn
(0.0780)
0.6550nnn
(0.0966)
0.4565nnn
(0.0915)
Intercept
0.0008
(0.0438)
HomeLev
AdjR2
N
0.4240
427
0.0097
(0.0474)
0.4180
296
0.0101
(0.0524)
0.4170
309
23
Because we do not have access to Malmendier, Tate, and Yans
(2010) Longholder data, in our analysis we use a classication of CEOs
based on media articles. We review articles on our sample CEOs for the
three years prior to 2004 in The New York Times, Business Week, The
Economist, and The Wall Street Journal. Articles in which the words
condent, condence, optimistic, and optimism were used in
association with the CEO were classied to imply a condent CEO. Along
with articles negating overcondence, articles with cautious, conservative, reliable, practical, frugal, and steady were classied
to imply a cautious CEO. We dene Condent as an indicator variable
with value one if the number of articles implying a condent CEO
exceeds the number implying a cautious CEO, and zero otherwise.
Cautious is one if the number of cautious articles exceed the number
of condent articles, and zero otherwise.
Panel A
EqOwn
Founder
Age
Tenure
DepBaby
Cohort
Military
MBA
PriorCFO
FinBack
Condent
Cautious
Mean
St. dev.
8.3824
0.0463
55.0000
7.1372
0.0033
1945.0000
0.0615
0.3743
0.1229
0.1453
0.0645
0.0099
2.1632
0.2103
6.8385
6.5432
0.0574
7.2919
0.2405
0.4846
0.3288
0.3528
0.2458
0.0992
535
605
605
605
605
605
358
358
358
358
605
605
Corr. with
HomeLev
0.2006nnn
0.0707n
0.1694nnn
0.1312nnn
0.0667
0.1615nnn
0.0652
0.0153
0.0642
0.0265
0.0445
0.1159nnn
Table 7
CEO characteristics.
The table reports the relationship between HomeLev and various other CEO characteristics. In Panel A, summary statistics and correlations with HomeLev are reported for each of the CEO characteristics. In
Panel B, the coefcients and standard errors are reported for various CEO characteristics as determinants of corporate leverage. The estimates are obtained from regressing the total debt to market value of
assets of the rm in 2004 (TDM) on the baseline model of capital structure reported in Table 3 Panel A column 4 with HomeLev omitted plus the noted characteristic, using OLS estimation. Coefcient estimates
and standard errors for the control variables are not reported, but are included in all regressions in Panels B, C, and D. In Panel C, HomeLev is added as a determinant of capital structure for the regressions in
Panel B. Coefcient estimates and standard errors are reported for both HomeLev and the specic CEO characteristic being tested. In Panel D, we test several characteristics jointly. HomeLev is dened as the
ratio of mortgage value to purchase price used by the rms CEO in his most recent primary home purchase. EqOwn is the log of the market value of the CEOs equity ownership in the rm. Ownership data are
from ExecuComp and price data are from CRSP. Founder is an indicator variable equal to one if the CEO is the founder of the company, and zero otherwise. The data on founder CEOs are from Fahlenbrach
(2009). Age is the age of the CEO in 2004. Tenure is the number of years the CEO held the CEO position as of 2004. DepBaby is a dummy variable that is equal to one if the CEO was born during the period 1920 to
1929. Cohort is the decade in which the CEO was born (i.e., if the CEO was born in 1945, Cohort is 1940). Military is a dummy variable that is equal to one if the CEO has military experience and is zero otherwise.
MBA is a dummy variable that is equal to one if the CEO has an MBA and is zero otherwise. PriorCFO is a dummy variable that is equal to one if the CEO was ever CFO of a company and is zero otherwise. FinBack
is a dummy variable that is equal to one if PriorCFO equals one or the CEO has a degree in the area of nance and is zero otherwise. The data on CEOs career paths, educational background, and military history
are hand-collected from Marquis Whos Who database and the NNDB online database. We are able to identify 358 (59.1%) of the 605 CEOs in the sample using these sources. Condent is a dummy variable that
is equal to one if in the years 2001 through 2003 more news articles use condent adjectives than cautious adjectives and is zero otherwise. Cautious is a dummy variable that is equal to one if in the years 2001
through 2003 more news articles use cautious adjectives than condent adjectives and is zero otherwise. In constructing both the Cautious and Condent variables, we follow the methodology outlined in
Malmendier, Tate, and Yan (2010). The table reports White (1980) heteroskedasticity-consistent standard errors. Signicance levels are denoted by n, nn, nnn, which correspond to 10%, 5%, and 1% levels,
respectively.
Panel B
EqOwn
(1)
CEO characteristic
0.373
535
0.0411n
(0.0234)
0.403
605
Age
(3)
0.0000
(0.0008)
0.401
605
Tenure
(4)
0.0011
(0.0009)
0.402
605
DepBaby
(5)
0.0694nnn
(0.0223)
0.401
605
Cohort
(6)
0.0003
(0.0008)
0.401
605
Military
(7)
0.0604
(0.0512)
0.395
358
MBA
(8)
0.0092
(0.0180)
0.390
358
PriorCFO
(9)
FinBack
(10)
0.0016
(0.0219)
0.0025
(0.0205)
0.389
358
0.389
358
Condent
(11)
0.0023
(0.0306)
0.401
605
Cautious
(12)
0.0277
(0.0498)
0.401
605
33
AdjR2
N
0.0080nnn
(0.0031)
Founder
(2)
34
Table 7 (continued)
Panel C
EqOwn
(1)
CEO characteristic
HomeLev
0.0323
(0.0231)
Age
(3)
0.0006
(0.0008)
Tenure
(4)
DepBaby
(5)
0.0005
(0.0009)
0.0408n
(0.0227)
Cohort
(6)
0.0008
(0.0008)
Military
(7)
MBA
(8)
0.0661
(0.0505)
0.0076
(0.0178)
PriorCFO
(9)
FinBack
(10)
0.0034
(0.0217)
0.0001
(0.0199)
Condent
(11)
0.0001
(0.0300)
0.0667nnn
(0.0188)
0.0703nnn
(0.0179)
0.0737nnn
(0.0180)
0.0702nnn
(0.0181)
0.0713nnn
(0.0180)
0.0745nnn
(0.0180)
0.0728nnn
(0.0240)
0.0699nnn
(0.0236)
0.0704nnn
(0.0238)
0.0702nnn
(0.0237)
0.0718nnn
(0.0179)
0.388
535
0.416
605
0.416
605
0.416
605
0.415
605
0.416
605
0.409
358
0.403
358
0.402
358
0.402
358
0.415
605
Cautious
(12)
0.0017
(0.0509)
0.0717nnn
(0.0180)
0.415
605
Panel D
(1)
(2)
0.0104nn
(0.0041)
0.0255
(0.0345)
0.0015
(0.0019)
0.0989n
(0.0540)
0.0010
(0.0011)
0.0013
(0.0341)
0.0169
(0.0181)
0.0170
(0.0231)
0.0195
(0.0271)
0.0720nn
(0.0318)
Yes
0.364
319
0.0602nn
(0.0252)
0.0092nn
(0.0041)
0.0293
(0.0340)
0.0018
(0.0019)
0.0854
(0.0539)
0.0013
(0.0011)
0.0020
(0.0340)
0.0141
(0.0182)
0.0168
(0.0226)
0.0196
(0.0266)
0.0567
(0.0349)
Yes
0.375
319
HomeLev
EqOwn
Founder
Tenure
DepBaby
Cohort
Military
MBA
FinBack
Condent
Cautious
Baseline controls
AdjR2
N
AdjR2
N
0.0061nn
(0.0030)
Founder
(2)
35
Table 8
CEO turnover and changes in corporate leverage.
Panel A of this table reports summary statistics for changes in CEO HomeLev for 84 S&P 1,500 non-nancial rms. HomeLev is dened as the ratio of
mortgage value to purchase price used by the rms CEO in his most recent primary home purchase and HomeLevPrev is the home leverage of the rms
previous CEO. Panel B reports regression results using the 84 sample rms for which changes in HomeLev are calculated. Column 1 of Panel B reports
coefcients and standard errors from regressing HomeLev on HomeLevPrev. Columns 2 and 3 of Panel B report coefcients and standard errors from
regressing the change in the total debt to market value of assets of the rm (TDMChg) on changes in the determinants of capital structure, using OLS
estimation. Control variables are constructed using one-year lagged data and are dened as in Appendix B. HomeLevChg is dened as HomeLev
HomeLevPrev. HomeLevDecr is a dummy variable that takes a value of one if the HomeLev of the incumbent CEO is less than the HomeLev of the previous
CEO. HomeLevIncr is a dummy variable that takes a value of one if the HomeLev of the incumbent CEO is greater than the HomeLev of the previous CEO.
TDM0 is the year-end TDM of the last full year of the previous CEOs tenure. Columns 2 and 3 include xed effects for the rst year of tenure of the current
CEO. The table reports White (1980) heteroskedasticity-consistent standard errors. Signicance levels are denoted by n, nn, nnn, which correspond to 10%,
5%, and 1% levels, respectively.
Panel A
Increases in HomeLev
No change in HomeLev
Decreases in HomeLev
Changes in HomeLev
MED
MEAN
STD
MIN
MAX
0.35
0.00
0.29
0.00
0.41
0.00
0.36
0.06
0.27
0.00
0.25
0.42
0.03
0.00
0.93
0.93
0.95
0.00
0.06
0.95
N
39
15
30
84
Panel B
Dependent variable
Intercept
HomeLevPrev
HomeLev
(1)
TDMChg
(2)
0.3435nnn
(0.0512)
0.2319nn
(0.1007)
0.0674n
(0.0346)
0.056
84
0.1609nn
(0.0712)
0.0142
(0.0104)
0.0065
(0.0312)
0.3103nn
(0.1292)
0.1166
(0.1755)
0.0139
(0.2132)
0.094
83
0.1152nnn
(0.0418)
0.0302
(0.0345)
0.1599nn
(0.0626)
0.0123
(0.0089)
0.0186
(0.0292)
0.2985nn
(0.1321)
0.0392
(0.1687)
0.0276
(0.1794)
0.168
83
No
Time
Time
HomeLevDecr
HomeLevIncr
TDM0
MktbkChg
AssetsChg
ProtChg
TangChg
IndusLevChg
Fixed effects
0.0145
(0.0487)
0.0622n
(0.0329)
HomeLevChg
AdjR2
N
TDMChg
(3)
CEO, 30 observations with HomeLevChg o 0, and 15 observations with no change (often zero or 80% home leverage).
We construct indicator variables for a leverage increase
(HomeLevIncr) and decrease (HomeLevDecr). As can be
seen in the table, the mean (median) increase in personal
home leverage is 0.41 (0.35), while the mean (median)
decrease is 0.36 (0.29).
We report two results from the CEO turnover analysis.
First, in column 1 of Panel B, we regress the current CEOs
personal home leverage on the previous CEOs leverage. We
nd a positive (0.2319) and statistically signicant, at the
5%-level, relation between the home leverage of the current
and previous CEOs. That is, if the previous CEO of a rm had
relatively low personal leverage, the current CEO also tends
to have low personal leverage. Second, in column 2, we
regress changes in corporate leverage on HomeLevChg,
36
Table 9
Effects of corporate governance.
The table reports coefcients and standard errors from regressing
different measures of corporate leverage computed in 2004 on determinants of capital structure, using OLS estimation. Unreported control
variables are constructed using 2003 data and include those in the
baseline model dened in column 4 of Table 3, Panel A. The sample is
non-nancial S&P 1,500 rms. HomeLev is dened as the ratio of
mortgage value to purchase price used by the rms CEO in his most
recent primary home purchase. IncentPay is the ratio of CEO incentive
compensation to total compensation in 2003. It is computed as the CEOs
total compensation minus salary and deferred compensation divided by
his total compensation (ExecuComp items (TDC1 SALARY DEFER RPT
AS INC TOT)/TDC1). SmallBoard is an indicator variable that is one if the
number of directors on the rms board is less than or equal to the
median number of board members per rm in the sample in 2004 (nine
directors or less), and is zero otherwise. GoodGov is an indicator variable
that equals one if the 2004 governance index of Gompers, Ishii, and
Metrick (2003) is less than or equal to six, and zero otherwise. The data
on the governance index and board size are from RiskMetrics. The table
reports White (1980) heteroskedasticity-consistent standard errors.
Signicance levels are denoted by n, nn, nnn, which correspond to 10%,
5%, and 1% levels, respectively.
(1)
(2)
nnn
HomeLev
0.1280
(0.0445)
IncentPay
0.0030
(0.0356)
HomeLev IncentPay
(3)
nnn
0.0928
(0.0234)
0.0614nnn
(0.0194)
0.1163
(0.0749)
SmallBoard
0.0172
(0.0173)
0.0711nn
(0.0324)
HomeLev SmallBoard
GoodGov
0.0341n
(0.0187)
HomeLev GoodGov
0.0253
(0.0397)
Baseline controls
AdjR2
N
Yes
0.4010
580
Yes
0.4480
483
Yes
0.4320
535
27
We do not have sufcient cross-sectional variation in our 2004
sample to study board independence (because of the Sarbanes-Oxley Act
and the new listing rules by NYSE and Nasdaq).
37
Table 10
Corporate valuation effects.
The table reports coefcients and standard errors from regressing rm value (Q) on various determinants of rm value, using OLS estimation. Control
variables are constructed using 2003 data and Q is dened as 2004 Mktbk dened as in Table 1. AbsDev is the absolute value of the difference between the
predicted values from a standard corporate leverage model specication with and without HomeLev included. Fig. 2 displays a histogram of this variable.
AbsDevQ4 is an indicator variable which is one if the rm is in the quartile with the largest absolute deviation. Assets is the natural logarithm of total rm
assets. EBIT is a measure of protability and is dened as EBIT/Sales. CAPEX is a measure of capital expenditures and is dened as CAPEX/Sales. SPDum is a
dummy variable that is one if the rm was a member of the S&P 500 in 2004 and zero otherwise. Lev is 2003 TDM as dened in Appendix B. IndusQ is the
median Q value for the rms four-digit SIC industry for the universe of Compustat rms. Denitions for SmallBoard and GoodGov are found in Table 9. The
sample is non-nancial S&P 1,500 rms. The table reports White (1980) heteroskedasticity-consistent standard errors. Signicance levels are denoted by
n nn nnn
, ,
, which correspond to 10%, 5%, and 1% levels, respectively.
(1)
AbsDev
(2)
(3)
(4)
0.2618nn
(0.1305)
0.4112nnn
(0.1084)
0.0713
(0.1054)
0.3541n
(0.1957)
0.2754nnn
(0.1046)
7.1425
(6.2225)
AbsDevQ4
SmallBoard
AbsDevQ 4 SmallBoard
GoodGov
0.3638nnn
(0.0720)
0.0037
(0.0053)
0.7382
(0.6352)
1.1106nnn
(0.1887)
0.0126nnn
(0.0038)
0.5387nnn
(0.1209)
3.3817nnn
(0.6643)
0.3775nnn
(0.0742)
0.0035
(0.0052)
0.7704
(0.6178)
1.1220nnn
(0.1910)
0.0135nnn
(0.0036)
0.5327nnn
(0.1205)
3.3910nnn
(0.6170)
0.1896nnn
(0.0535)
0.9289nn
(0.3903)
0.7850n
(0.4147)
0.7640nnn
(0.1639)
1.7203nnn
(0.2934)
0.3908nnn
(0.1196)
2.4401nnn
(0.3784)
0.1468
(0.1158)
0.3909n
(0.2135)
0.1756nnn
(0.0462)
0.5119nnn
(0.1735)
1.0423nn
(0.4123)
0.7173nnn
(0.1573)
1.5329nnn
(0.2477)
0.3709nnn
(0.1069)
2.4421nnn
(0.3491)
0.2490
605
0.2520
605
0.2890
483
0.2850
535
AbsDevQ 4 GoodGov
Assets
EBIT
CAPEX
SPDum
Lev
IndusQ
Intercept
AdjR2
N
(footnote continued)
leverage. If the G-index is not a measure of good governance, but rather
a measure of young growth rms, we nd that our results are robust to
controlling for young growth rms as evidenced by the positive and
statistically signicant coefcient on HomeLev in column 3 of Table 9.
31
Recall that AbsDev is the absolute value of the difference between
the predicted values from a standard corporate leverage model specication with and without HomeLev included.
32
See, e.g., Korteweg (2010) for estimates of the markets valuation
of corporate leverage deviations.
38
39
Table A1
Summary statistics: CEOs primary homes.
The table reports summary statistics for characteristics of primary homes of CEOs for a sample of S&P 1,500 rms. Data on CEO home characteristics
were collected primarily from the Lexis-Nexis public documents database, which includes national coverage of mortgage records, deed transfers, and tax
assessor records. The U.S. median data are tabulated from 2005 data provided by the Federal Housing Finance BoardPeriodic Summary Tables and the
2005 American Community Survey Subject Tables. Purchase prices are reported in 2005 home price dollars, and adjusted using the Ofce of Federal
Housing Enterprise Oversights National Home Price Index.
MED
MEAN
STD
MIN
MAX
5,154
1.1
1989
10.0
4.0
5.0
1,585
5,658
3.4
1975
10.9
4.5
5.1
2,155
2,852
9.7
34
3.5
1.4
2.0
1,929
785
0.1
1740
5.0
0.0
1.0
114
22,371
140.0
2008
36.0
16.0
17.0
14,643
647
604
676
396
520
622
641
U.S. MED
1975
5.0
3.0
40