You are on page 1of 35

American Finance Association

Back to the Beginning: Persistence and the Cross-Section of Corporate Capital Structure
Author(s): Michael L. Lemmon, Michael R. Roberts and Jaime F. Zender
Source: The Journal of Finance, Vol. 63, No. 4 (Aug., 2008), pp. 1575-1608
Published by: Wiley for the American Finance Association
Stable URL: http://www.jstor.org/stable/25094484
Accessed: 26-02-2016 14:08 UTC

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at http://www.jstor.org/page/
info/about/policies/terms.jsp
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content
in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship.
For more information about JSTOR, please contact support@jstor.org.

American Finance Association and Wiley are collaborating with JSTOR to digitize, preserve and extend access to The Journal
of Finance.

http://www.jstor.org

This content downloaded from 210.56.13.7 on Fri, 26 Feb 2016 14:08:45 UTC
All use subject to JSTOR Terms and Conditions

THE JOURNAL OF FINANCE

VOL. LXIII, NO. 4

AUGUST 2008

Persistence
Back to the Beginning:
of Corporate
Cross-Section
Capital

and the
Structure

L. LEMMON, MICHAEL R. ROBERTS,


and JAIME F. ZENDER*

MICHAEL

ABSTRACT
the majority
of variation
in leverage
ratios
is driven
by an unobserved
effect that generates
stable capital
structures:
time-invariant
(low)
surprisingly
High
as such for over two decades.
levered firms
tend to remain
This
of leverage
feature
identified
to firm exit,
is largely unexplained
is robust
determinants,
by previously

We

find

that

to the IPO, suggesting


that variation
in capital
structures
and is present
is
prior
determined
stable for long periods
of time. We then
that remain
by factors
primarily
show that these results have
for empirical
important
implications
analysis
attempting
structure
to understand
capital
heterogeneity.

in financial
is:How do firms choose their cap
economics
A fundamental
question
ital structures?
is at the heart of the "capital structure
Indeed, this question
address. Attempts
(1984) in his AFA presidential
puzzle" put forward by Myers
a great deal of discussion
to answer this question have generated
in the finance
literature. Many
both before and after Myers's
iden
studies,
pronouncement,
to explain variation
of factors that purport
in corporate
tify a number
capital
structures.
after decades
of research,
how much do we really know?
However,
how much
More precisely,
closer have previously
and
identified
determinants
existing empirical models moved us toward solving the capital structure puzzle?
how can we move
still closer to ultimately
And, given this progress,
providing
a more complete understanding
of capital structure
decisions?
we quan
these questions.
The goal of this paper is to address
Specifically,
to which
determinants
cross-sectional
and
govern
tify the extent
existing
*
Michael
from

Lemmon

the Wharton

is from

Business,
University
and associate
editor. We
Al?n

Brav,

Mark

School

of Business,
of Utah.
Michael
Roberts
is
University
of Pennsylvania.
Jamie
Zender
is from the Leeds
School
of
are especially
for helpful
comments
from our referee
grateful

thank

Franklin

Murray
Andrew

Frank,

also

Flannery,
Mark Leary,

John

Eccles

School, University
of Colorado. We

Graham,
Ivo Welch,
tersen, Rob Stambaugh,
of Arizona,
Babson
Boston
College,
sity, University

Carolina,
conference

from

Metrick,
Toni Whited,
Bilge Yilmaz;
Cornell University,
College,
of Maryland,
University
University

of Colorado,
of Pennsylvania,

University

participants
and Technology
Finance
ern Finance
Association
a Rodney

L. White

Yakov Amihud,
Lincoln
Allen, Heitor Almeida,
Berger,
Sara Ghafurian,
William
Vidhan
Goetzmann,
Goyal,
Roni Michaely,
Pe
Palia, Mitchell
Vinay Nair, Darius

at the 2005

Queens
Five-Star

University,

seminar
Drexel

participants
University,

of Michigan,
of North
University
the University
of Western
and
Ontario;
2005 Hong Kong University
of Science

Symposium,
for helpful

Conference,
NBER
Finance
Corporate
Conference,
Roberts
discussions.
gratefully
acknowledges

Grant

an NYSE

and

2006

Research

at University
Harvard
Univer

and
financial

Fellowship.

1575
This content downloaded from 210.56.13.7 on Fri, 26 Feb 2016 14:08:45 UTC
All use subject to JSTOR Terms and Conditions

2006 West
support

1576

The Journal

of Finance

structures
in observed
time-series
variation
the evolu
capital
by examining
tion of corporate
the
leverage ratios. In doing so, we are not only able to assess
we are also able to
of existing
progress
empirical work, but more
importantly,
characterize
what existing determinants
appear to miss?the
gap in our under
of
in
what
determines
structure.
Our analysis,
capital
standing
heterogeneity
on
some
new
to
while
several
also
issues,
presents
shedding
light
challenges
how firms choose their capital structures.
understanding
We begin by showing that leverage ratios exhibit two prominent
features
that
are unexplained
identified
determinants
by previously
(e.g., size, profitability,
in sample composition
market-to-book,
industry, etc.) or changes
(e.g., firm exit).
are illustrated
1 (see Section
in Figure
which
shows the
These
features
II),
future
evolution
of leverage
ratios for four portfolios
constructed
by sorting
in
firms according
to their current
feature
ratios. The first notable
leverage
of convergence
is that leverage
amount
the figure
ratios exhibit a significant
over time; firms with relatively
toward more
tend to move
high (low) leverage
this convergence,
levels of leverage. The second feature is that, despite
stable over time; firms with
ratios are remarkably
relatively
high
leverage
tend to maintain
(low) leverage
relatively
high (low) leverage for over 20 years.
and a permanent
ratios are characterized
Thus,
by both a transitory
leverage
as
to
have
discussed
be
identified.
that,
below,
yet
component
are these components?
The adjusted ?-squares
from tradi
How
important
identified
determinants
tional leverage regressions
range from
using previously
on the specification.
In contrast,
18% to 29%, depending
the adjusted
i?-square
"stand-ins"
for the
of leverage on firm fixed effects (statistical
from a regression
of
varia
is
that
of
the
60%, implying
component
majority
permanent
leverage)
in a panel of firms is time invariant
and is largely unexplained
tion in leverage
determinants.
identified
by previously
em
is that commonly
for these findings
One possible
employed
explanation
are
more
are
with
concerned
because
models
managers
misspecified
pirical
as opposed
in long-run or equilibrium
levels of leverage determinants,
variation
a distributed
to short-run
We test this hypothesis
fluctuations.
by estimating

moderate

of leverage. Two facts emerge from this exercise. First, the responses
lag model
in its determinants
often differ,
to short-run
of leverage
and long-run variation
for lagged effects in empir
of accounting
the potential
importance
highlighting
even after allowing
that
for an aggregate
ical specifications.
response
Second,
to
still struggles
occurs over 8 years, variation
determinants
in the traditional
a
deviation
one-standard
For
structures.
in
variation
example,
explain
capital
level of a firm's industry median
in the long-run equilibrium
leverage,
change
results
determinant
of book leverage,
observable
influential
the single most
to
the
uncondi
relative
in expected
small
fraction
in a 6% change
leverage?a
of whether
21%. Thus, regardless
of book leverage,
deviation
tional standard
or a long-run perspective,
of cap
one takes a short-run
determinants
existing
in
a
of
the
variation
small
fraction
to
ital structure
appear
relatively
explain
leverage

ratios.

We then examine
in capital structure.

the
We

for empirical
of these findings
implications
between
associations
find that the estimated

This content downloaded from 210.56.13.7 on Fri, 26 Feb 2016 14:08:45 UTC
All use subject to JSTOR Terms and Conditions

research
leverage

Back

1577

to the Beginning

are highly sensitive


to changes
in model
determinants
identified
and previously
on
identified
determinants
The
estimates
coefficient
previously
specification.
an average
in magnitude
of 86% (65%) in the book (mar
decrease
experience
firm
fixed effects
after
ket) leverage
(i.e., accounting
incorporating
regression
errors (i.e., ac
and
of
for the permanent
leverage)
component
serially correlated
of leverage). Given the importance
of this
component
counting for the transitory
in leverage, parameter
that do not account
estimates
unobserved
heterogeneity
structural
effect (via within-transformation,
for the firm-specific
differencing,
natural
etc.) and serial correlation
(via lagged depen
estimation,
experiments,
correlated
dent variables,
errors, etc.) are suspect. That is, it is simply
serially
in models
of the
to draw causal inferences
untenable
ignoring these components
of the parameters
is
identification
of interest
data generating
process because
(2003) and Hsiao
(2003)).
(Arellano
questionable
lies behind the unidentified
Next, we turn to the question: What
components
com
1? Focusing
first on the transitory
of capital structure
revealed
by Figure
we
1
of
show
that
the
ratios
revealed
convergence
ponent,
leverage
by Figure
state variable
in
is due, at least in part, to the role of leverage as an important
An analysis
of security
firms' net issuance decisions.
issuance
behavior
identi
fies debt policy as an important mechanism
for controlling
corporate
leverage,
while
role. This finding
that active
suggests
equity policy plays a secondary
of leverage
ratios is at least partially
for the mean
management
responsible
in leverage ratios?findings
consistent
with a large recent literature
reversion
this issue.1
examining
we also show that this dynamic
is directed
toward a
Moreover,
rebalancing
the unobserved
of
target, namely,
largely time-invariant
permanent
component
is at odds with recent conclusions
and Ran
leverage. This finding
by Flannery
that firms
gan (2006) and Hovakimian,
(2001), who suggest
Opler, and Titman
toward time-varying
appear to be adjusting
targets. We show that accounting
for time-varying
factors in the target specification
has a negligible
effect on both
a
the model fit and estimated
of
to
relative
time
speed
adjustment
firm-specific,
our
invariant
for
This
result
fits
with
other
target leverage.
specification
nicely
as opposed
the relative
of cross-sectional,
to
findings
importance
highlighting
in
variation
structures
and
the
limited
time-series,
power
capital
explanatory
of previously
identified
leverage determinants.
of leverage,
Finally, we turn to the permanent
component
portrayed
by the
1.While
factors
what
long-lived
stability of leverage ratios in Figure
identifying
are behind
this feature of the leverage data generating
is beyond
the
process
that persistent
scope of this paper, we take a step toward this goal by showing
in leverage
differences
ratios exist among a sample of privately
in
held firms
the United
for
which we are able to obtain data. We also find that,
Kingdom
these differences
back in time, predating
the
among domestic
firms,
persist
IPO. In other words, high (low) levered private firms remain so even after going
1
Studies
and

by Flannery

Roberts

response

and Rangan
(2005)

and Liu
(2005),
to various
shocks.

(2006), Hovakimian
that firms
find

(2006),
gradually

Kayhan
adjust

and Titman
their

capital

This content downloaded from 210.56.13.7 on Fri, 26 Feb 2016 14:08:45 UTC
All use subject to JSTOR Terms and Conditions

(2007), Leary
structure
in

1578

The Journal

of Finance

in the information
the corresponding
environment,
public and despite
changes
the distribution
of control, and the access to capital. Thus, any explanation
of
structure
must
with
be
reconciled
this
of
capital
long-lived
stability
leverage
ratios.2

as follows. Section
The remainder
I discusses
the
of the paper is organized
In Section
data and sample selection.
the dynamic
behavior
of
II, we examine
from several perspectives.
III examines
the implica
Section
capital structure
of leverage
tions of our findings
for existing
determinants
ratios. Section
IV
for the empirical
examines
the implications
of our findings
of capi
analysis
tal structure.
what drives the convergence
of leverage
Section V investigates
Section VI investigates
what
ratios over time (i.e., the transitory
component).
drives the long-lived
of leverage
ratios over time (i.e., the permanent
stability
Section VII concludes.
component).

I. Data

and

Sample

Selection

in
consists
of all nonfinancial
observations
Our primary
firm-year
sample
1965 and 2003.3 We require that all
the annual Compustat
database
between
anal
have nonmissing
data for book assets, while
all multivariate
firm-years
for
the
relevant
We
variables.
data
require
ysis implicitly
requires nonmissing
lie in the closed unit interval. All other
book and market?to
leverage?both
to mitigate
the effect
ratios are trimmed at the upper and lower one-percentiles
we also
errors in the data. For some of our analysis,
of outliers
and eradicate
an
used
of
all
of
the
variables
IPO
The
construction
identifiable
date.4
require
in the Appendix.
in this study is detailed
for all of our firms, as well as a subsample
statistics
Table I presents
summary
at least 20 years of nonmissing
data on book leverage. We refer
of firms having
on at least
is predicated
to this latter sample as "Survivors,"
since selection
mo
for survivorship
The potential
bias in our analysis
20 years of existence.
as a ro
our examination
in all subsequent
of this subsample
tivates
analysis
of the
and the similarity
due to space considerations
bustness
check; however,
we sometimes
these results.
suppress
findings,
dif
several unsurprising
reveals
the samples
between
A quick comparison
more
fewer
and
have
to
tend
be
Survivors
ferences.
growth
profitable,
larger,
to
assets
relative
but more tangible
(i.e., lower market-to-book),
opportunities
to
have
also
tend
Survivors
the general
leverage,
especially
higher
population.
that firm exits due to buyouts and acquisi
in market
value terms. This suggests
as those due to bankruptcy.
as important
the
tions are potentially
Alternatively,
an
effects?survivor
of
be
artifact
of
survivors
may
confounding
leverage
higher
2
Strebulaev
bility are also
3
We choose
Compustat;
analysis
4
The

and Yang
(2006) show in a recent working
paper
zero leverage
common
ratios.
to firms with
concerns
selection
of sample
because
this horizon

however,

none

1950
examining
IPO information

of our results
to 2003

and

is obtained

are sensitive
1971
from

to 2003
SDC

to this

that

associated

time

frame,
horizons.

sample
and Jay Ritter,

this

whom

stability
with

as suggested
we

kindly

and

pre-1965

inexplica
data

by unreported
thank.

This content downloaded from 210.56.13.7 on Fri, 26 Feb 2016 14:08:45 UTC
All use subject to JSTOR Terms and Conditions

in

Table

Statistics

Summary
The

consists

sample

in the Compustat
from 1965 to 2003. The
database
and standard
deviations
(SD) for the entire
(in brackets),
to survive
of firms
for at least 20 years
the subsample
required
are provided
in the Appendix.
firms

of all nonfinancial

variable
averages,
presents
as well
as
(All Firms),
sample
Variable
definitions
(Survivors).
table

medians

All
Mean

Variable
Book

[0.24]
0.28

leverage

Firms

[Median]
0.27

leverage

Market

1579

to the Beginning

Back

Log( Sales)

[0.23]
4.43

Market-to-book

[4.54]
1.59

Profitability

[1.00]
0.05

Tangibility

[0.12]
0.34

Survivors

(SD)

flow

0.10

vol.
book

industry

leverage

0.24

0.27

(0.18)

(0.26)

(0.25)

(2.52)

[0.28]
5.49

(2.25)

(1.87)

[5.59]
1.23

(1.19)

(0.26)

[0.89]
0.12

(0.13)

[0.13]
(0.25)

Intangible

0.39

payer

[0.00]
0.05

assets

(0.14)

(0.25)

0.07

(0.08)

[0.05]
(0.13)

0.26

(0.12)

(0.49)

[0.25]
0.63

(0.48)

(0.10)

[1.00]
0.04

(0.08)

[0.00]
Obs.

0.39

[0.33]

[0.24]
Dividend

(SD)

[0.26]
0.32

[0.06]
Median

[Median]

(0.21)

[0.28]
Cash

Mean

[0.00]
92,306

225,839

firms are larger, and larger firms tend to have higher leverage
(Titman and Wes
are
note that these summary
sels (1988)). At this point, we merely
statistics
in
consistent
with
those
studies
and
with
intuition.
found
broadly
previous

II. The

Evolution

of Leverage

We begin our analysis


the evolution
of leverage
for our cross
by studying
1 presents
section of firms. Figure
the average
of
four
ratios
leverage
portfolios
in "event time." The figure
in the following manner.
is constructed
Each cal
endar year, we sort firms into quartiles
to their
(i.e., four portfolios)
according
we
which
denote:
and
Low.
The
ratios,
leverage
Very High, High, Medium,
formation
event year 0. We then compute
the average
year is denoted
portfolio
for
each
in
each
of
the
20
the port
years, holding
leverage
portfolio
subsequent
folio composition
constant
(but for firms that exit the sample). We repeat these
two steps of sorting and averaging
for every year in the sample period. This
one for each calendar year in
39 sets of event-time
process generates
averages,
our sample. We then compute
the average
leverage of each portfolio across the

This content downloaded from 210.56.13.7 on Fri, 26 Feb 2016 14:08:45 UTC
All use subject to JSTOR Terms and Conditions

1580

of Finance

The Journal
Panel A: Book Leverage Portfolios

1. Average
Figure
of all nonfinancial
average
leverage
That
is, for each

Panel

B: Book Leverage Portfolios (Survivors)

of actual
in event
consists
time. The sample
leverage
leverage
portfolios
in the Compustat
firms
database
from 1965 to 2003. Each panel
the
presents
in event
of four portfolios
formation
year zero is the portfolio
time, where
period.
on their actual
calendar
firms
based
year, we form four portfolios
by ranking

the average
the portfolios
fixed
for the next 20 years, we compute
leverage.
Holding
leverage
in 1975 we sort firms
each portfolio.
For example,
ratios.
into four groups based on their leverage
1975 to 1994, we compute
each year from
the average
for each of these four portfolios.
leverage

for
For
We

this process
of sorting
and averaging
for every year in our sample horizon.
After performing
repeat
this sorting
and averaging
for each year from
1965 to 2003, we then average
the average
leverages
across
"event time" to obtain
the bold lines in the figure.
The surrounding
dashed
lines represent
are presented
95% confidence
intervals.
in Panels
A and
for book and market
The results
leverage
is defined

book
(market)
C, where
leverage
debt and market
Panels
equity).
but for a subsample
respectively,
can only

the portfolio

perform

as

the

B and D present
of firms
required

formation

through

ratio

of total

debt

similar

results

to exist

for at least

1984

for this

5
The

confidence

Estimating
underestimate
conservative
each

interval

error

standard
the

error

standard
the

estimate

is defined
as

is estimated
true

standard
that

as a two-standard
the

average
the standard

using
error

ignores

the

because
effects

error

standard
error
of the

assets

of total

(sum

and market

leverage,
we
(consequently,

20 years

sample).

39 sets within
each event year. We perform
this exercise
are presented
and market
the results
of which
leverage,
els A and C, respectively.
The dashed
lines surrounding
95% confidence
intervals.5
represent

The

to total

for book

interval
error

of the

overlapping
across
of averaging

for both book leverage


as solid lines in Pan
the portfolio
averages

around

across

the

across
average
observations.
the 39

sets,

mean.

the estimated
39

sets
the

of averages.
39 sets would

Thus,
effectively

set as redundant.

This content downloaded from 210.56.13.7 on Fri, 26 Feb 2016 14:08:45 UTC
All use subject to JSTOR Terms and Conditions

we

use

treating

Back

to the Beginning

1581

are worth
is a great
of the graphs
there
features
Several
First,
noting.
in
initial
the
formation
deal of cross-sectional
dispersion
portfolio
period. The
is
52%
is notice
of
book
there
(60%).
(market)
range
average
Second,
leverage
over time. After 20 years,
able convergence
among the four portfolio
averages
from 55% to 35%, whereas
the Very High book leverage portfolio has declined
from 3% to 19%. (The market
the Low portfolio has increased
leverage
port
a similar
occurs
in
of the convergence
folios display
Third, most
pattern.)
as evidenced
the first few years after the formation
period,
by the flatten
the
ing slope over time in the Low and Very High portfolios.
Finally,
despite
the average
20 years later remains
convergence,
leverage across the portfolios
and economically.
both statistically
different,
significantly
and Low
The average book leverage ratios in the Very High, High, Medium,
an average
after 20 years are 35%, 30%, 25%, and 19%, respectively,
portfolios
to the average within-firm
differential
of over 5%. When
standard
compared
this
of book leverage
differential
is
deviation
(12.9%),
large. There
economically
a
of
ratios
of a
examination
the
presence
fore,
leverage
suggests
preliminary
or
a
leads
to
short-run
in
that
lever
convergence
component
transitory
gradual
or long-run component
that leads to highly
age ratios, as well as a permanent
in leverage.
cross-sectional
differences
persistent
concern with interpreting
1 is the effect of survivorship
One potential
Figure
bias. First, as we progress
further away from the portfolio
formation
period,
out
firms will naturally
of
the
to
due
exit
drop
sample
through
bankruptcy,
or buyouts. Second, from 1984 onward, the length of time for which
acquisitions,
we can follow each portfolio
is censored because we only have data through 2003.
To address
this issue, we repeat the analysis
described
above for our sample of
for this subset of firms are presented
Survivors.
The results
in Panels B and
D of Figure
differences
between
the survivors
and the
1, and reveal negligible
in terms of the evolution
of leverage.
general population
we examine
the event-time
evolution
of leverage among the sub
Additionally,
we compute,
in similar event-time
fashion,
sample of exiting firms. Specifically,
the average of the last nonmissing
for each firm that exits
leverage observation
on the last observation
the sample, conditional
last
occurring before 2003?the
so
our
In
far as the last observable
of
year
sample.
leverage ratio is a reasonable
exam
for an additional
proxy for future leverage ratios, this analysis
provides
ination of the sample
are nearly
selection
issue. The figures,
not presented,
to those found in Figure
1 but for the larger standard
identical
errors due to
in the degrees
the significant
reduction
of freedom
10% of firms
(approximately
exit the sample each period). That is, the leverage
ratios of firms just prior to
the sample
look similar to the leverage
ratios of the broader
exiting
sample of
firms. Thus, firm exit does not appear
to be driving
the patterns
in
observed
l.6
Figure
6
A more

thorough

investigation

an

identification
appropriate
the capital
structure
decision.

potentially

fruitful

area

of the

for future

survivorship

issue would

entail

to disentangle
instrument(s),
is beyond
the scope
investigation
research.

strategy,
Such an

i.e.,

a model
the

of firm
exit

of this

exit

decision
study,

This content downloaded from 210.56.13.7 on Fri, 26 Feb 2016 14:08:45 UTC
All use subject to JSTOR Terms and Conditions

and
from

though

1582

The Journal

of Finance

concern with
A second potential
is the effect of the
the figure
interpreting
on the
In
bounded
of
because
other
is defined
words,
support
leverage.
leverage
a
will
unit interval,
have
to
reflect
natural
average
away
leverage
tendency
from the extremes
of zero and one (Chang and Dasgupta
(2006)). To exam
we transform
ine this possibility,
which
leverage using a logit transformation,
we
onto
from
the
unit
interval
the
whole
real
line.
maps
leverage
Specifically,
construct

. /t
Logit{Leverageit)

? /

Leverage^
\1
Leverageit

In-^_i?_

on the open unit


is that it is only defined
transformation
it
to
that
cannot
be
values
of
interval,
exactly equal
implying
applied
leverage
we perform
to zero or one. To address
this limitation,
two separate
analyses.
The first excludes
values
leverage values
equal to zero or one. Null
leverage
on
book
10.6%
the
of
observations
(market)
(8.4%)
leverage, while
comprise

limitation

of this

unit values
of leverage
less than 0.001%
of the
(less than 0.001%)
comprise
on book (market) leverage. The second analysis
adds 0.001 to each
observations
that are equal to one. Plots
value of leverage and excludes
the few observations
are
series
but for scale, to
of both transformed
identical,
virtually
leverage
in Figure
1 and, as such, are not presented.
those presented
Thus, the patterns
1 do not appear to be an artifact
of
of the bounded
domain
in Figure
observed
leverage.

concern with
is that the sorting
A final potential
the figure
interpreting
in
variation
cross-sectional
of firms
by leverage may
simply be capturing
in leverage
variation
with
factors associated
cross-sectional
(e.g.,
underlying
research
costs, agency costs, etc.). For example,
previous
(e.g., Tit
bankruptcy
man and Wessels
is positively
correlated with firm
(1988)) finds that leverage
to large
of the Very High portfolio may simply correspond
size, so that members
to small firms. To address
of the Low portfolio correspond
firms, while members
we modify
the sorting procedure.
this possibility,
a cross-sectional
we
of
Each calendar
year,
regression
begin by estimating
on
the
identified
have
been
that
factors
by
previously
1-year lagged
leverage
as being relevant
of capital
structure
determinants
literature
(e.g., Titman
and Phillips
and Wessels
(2005),
(1995), Mackay
(1988), Rajan and Zingales
on
tan
we
firm
and others).7
size,
regress
leverage
profitability,
Specifically,
variables
indicator
and industry
(Fama and French
gibility, market-to-book,
based on
into four portfolios
We then sort firms
classification).8
38-industry
the residuals
then

7
We
8
We

track

which we term "unexpected


from this regression,
over
of each portfolio
the average
actual
leverage

also

examine

also

examine

the effect

of using

an alternative

contemporaneous

determinants.
by Frank
an indicator

The

and
leverage,"
the subsequent

results

and Goyal
for whether

are unchanged.
(2004)

consisting
the firm paid a
assets,
collateral,
intangible
The results
indicators.
and Schallheim
dividend
(1998)), and year and industry
Lemmon,
(Graham,
are largely unchanged
and, as such, are not presented.
by these modifications
of firm

size, market-to-book,

specification

suggested

This content downloaded from 210.56.13.7 on Fri, 26 Feb 2016 14:08:45 UTC
All use subject to JSTOR Terms and Conditions

Panel A: Unexpected Book Leverage PortfoU?e

Panel C: Unexpected Market Leverage Portfolios

Pand B: Unexpected Book Leverage Portfolios (Survivors)

'
;,
ygy

. Panel ?>:
Urffiipected Nfarket Leverage Portfolios (Survivors)

2.

of unexpected
Average
leverage
in the Compustat
of all nonfinancial
firms

Figure
consists
the

average
That
period.
unexpected
the average

of four portfolios
leverage
year,
is, for each calendar
leverage

(defined

1583

to the Beginning

Back

leverage
database

portfolios
from 1965

in event
to 2003.

time.
Each

The

panel

year zero is the portfolio


time, where
firms
based
four portfolios
by ranking
the portfolios
fixed for the next 20 years, we

in event
we

form

below).

Holding
in 1975 we sort firms
for each portfolio.
For example,
into four groups
leverage
their unexpected
ratios. For each year from
1975 to 1994, we compute
the average
leverage
of sorting
for every
for each of these four portfolios.
We repeat
this process
and averaging
our sample
1965
for each year from
horizon.
After
this sorting
and averaging
performing

we

the average
average
leverages
dashed
lines represent
surrounding
are presented
in Panels
A and
leverage
then

The

total

debt

to total

assets

(sum

of total

across

"event

95% confidence
book
C, where
debt and market

time"

to obtain

intervals.
(market)

the bold

The

leverage
Panels
equity).
for a subsample

results

lines
for book

is defined

sample

presents
formation
on

their

compute
based on
leverage
year in
to 2003,

in the figure.
and market

as the ratio

B and D present
of firms
required

of

similar

to exist
but
leverage,
respectively,
we can only perform
1984 for
the portfolio
formation
through
as the residuals
is defined
from a cross-sectional
of
this sample).
Unexpected
leverage
regression
on firm size, profitability,
and tangibility,
where
all independent
variables
market-to-book,
leverage
are lagged 1 year. Also included
are industry
in the regression
indicator
variables
(Fama and French
results

for at

for book

least

and market

20 years

(consequently,

38-industry

classifications).

20 years.

An

regressions

Variable

definitions

attractive
feature
each year, we allow

are provided

in the Appendix.

is that, by estimating
of this approach
the marginal
effect of each factor to vary

the
over

time.

a linear sense?the
This approach
respect
homogenizes?in
sample with
to previously
of leverage.
The consequence
is that
identified
determinants
firms
that are uncorrelated
the
observable
each portfolio
contains
along
This content downloaded from 210.56.13.7 on Fri, 26 Feb 2016 14:08:45 UTC
All use subject to JSTOR Terms and Conditions

1584

The Journal

of Finance

characteristics.
To the extent
that the regression
model
is well
the
specified,
is twofold. First,
there should be less cross-sectional
in
variation
expectation
the formation
Second,
period as a result of sorting on the residual
leverage.
in the average
should quickly
levels across portfolios
any differences
leverage
as
the
random
shock
outcome
the
Neither
is the
of
disappear
impact
dissipates.
case. Figure
2 presents
the graphs
for the unexpected
and
leverage portfolios
to those presented
shows that the results are nearly
identical
in Figure
1.
In particular,
still varies over a large range (43% for book leverage,
leverage
in the portfolio
for market
formation
that
leverage)
period,
suggesting
in capital
structure
of the variation
is found in the residual
of existing
we see similar patterns
across
As time progresses,
of convergence
specifications.
across
while
in
the
the
the portfolios.
average
Finally,
leverage
portfolios
spread
in each event year has decreased,
there still remain
differences
for
significant
even 20 years after the portfolio
most periods. For example,
formation
period,
of Low levered firms
is significantly
the average
below
that of all
leverage
the
other portfolios,
both in terms of book and market
leverage. Additionally,
from that
levered firms is significantly
different
average
leverage of Very High
are economically
as well,
ofMedium
levered firms. These differences
significant
in event year 20 equal to 11%
with the range in leverage across the portfolios
even after removing
all observable
(12%) for book (market)
Thus,
leverage.

49%
most

with
traditional
determinants
of capital
associated
structure,
heterogeneity
remain
still
highly persistent.
leverage differences
of the data generating
In sum, the figures
features
reveal several
interesting
1
that
extreme
of
for
shows
values
very
process
leverage. Figure
leverage?both
over time, on average. Further,
to converge significantly
this
high and low?tend
run.
in
to
concentrated
the
short
this
be
convergence
appears
Finally,
despite
across
are
in
ratios
firms
differences
convergence,
leverage
highly persistent,
even after 20 years.
as indicated
convergence
by the incomplete
identified
2 shows that controlling
for previously
determi
leverage
Figure
is seen
nants has two effects. The first effect is a short-run
that
phenomenon
in the portfolio
formation
in the decreased
period. The second ef
dispersion
of the
that is seen in the increased
fect is a long-run phenomenon
convergence
over time. However,
to
both of these effects appear, at first glance,
portfolios
across the leverage portfolios
sorted by un
be relatively
small. The variation
across the leverage
than the variation
is not much
smaller
leverage
expected
the
sorted by actual leverage. Similarly,
among the unex
convergence
portfolios
and economically
still leaves statistically
significant
leverage portfolios
pected
across
some
20
later.
differences
years
portfolios
because
it
It is troubling
and troubling.
is both interesting
This last result
of
to
account
for
much
not
do
variables
traditional
control
that
appear
suggests
that an
it suggests
because
in leverage. The result is interesting
the variation
this
and
of
that
from
factor
is
leverage
specifications
existing
missing
important
as
or
well
a significant,
time-invariant
factor contains
component,
permanent,
as a slowly decaying
The next section
component.
transitory
an eye toward quantifying
of leverage with
these features
to existing
determinants.
relative
capital structure

further examines
their importance

This content downloaded from 210.56.13.7 on Fri, 26 Feb 2016 14:08:45 UTC
All use subject to JSTOR Terms and Conditions

III. The

Economic

Back

to the Beginning

Importance

of Persistence

1585
in Capital

Structures

1 and 2 is that leverage appears


to be persis
in Figures
A prominent
feature
this feature of the data further using three sets of analysis.
tent. We investigate
ratios play in determin
initial leverage
the role that firms'
First, we examine
we
a
of
variance
ratios.
future
Second,
perform
decomposition
leverage
ing
determinants
and test
to quantify
the explanatory
power of existing
leverage
for the presence
of unobserved
or, loosely speak
firm-specific
heterogeneity
traditional
specifications
empirical
ing, firm fixed effects. Finally, we expand
our findings
a distributed
to consider whether
of leverage
lag model
using
or
are due to managers
to shifts in long-run
levels of pre
reacting
expected
as opposed
to short-run
in their
fluctuations
identified
determinants,
viously
values.

A.

The Role

of Initial

Leverage

of leverage
of the permanent
revealed
One implication
component
by Fig
are
ures 1 and 2 is that firms' future
to their
ratios
leverage
closely related
initial leverage
ratios. However,
the figures
limited quantitative
evi
provide
measure
To
the
dence of initial leverage ratios' economic
importance.
impact of
the following
initial leverage on future leverage, we estimate
regression,
Leverageit

= a +
?Xit-i

+ yLeverage?0

+ vt + sit,

(1)

firms; t indexes years; X is a set of 1-year lagged control vari


is firm ?'s initial leverage, which we proxy for with
the first
ables; Leverage^
v
s
a
a
er
for
value
is
fixed
and
random
is
year
effect;
leverage;
nonmissing
ror term assumed
to be possibly
and correlated within
heteroskedastic
firms
(Petersen
(2007)). To avoid an identity at time zero, we drop the first observa
tion for each firm from the regression.
The coefficient
of interest
is y, which
measures
in determining
of firms'
initial leverage values
the importance
fu
ture values of leverage.
In light of the figures,
the average
y estimates
leverage
across firms over time. By incorporating
difference
the control variable X, we
can also compare the importance
of firms' initial conditions
to those of
relative
where

near

i indexes

contemporaneous

determinants.

results from estimating


(1) using book and market
equation
leverage are
in
II.
A
Table
Panel
the
results
the
full
presents
presented
using
sample; Panel
B presents
the results using the subsample
of survivors.
In order to facilitate
we scale each coefficient
variable's
standard
comparisons,
by the corresponding
measures
cor
deviation.
each
estimate
in
the
Thus,
reported
change
leverage
a
to
in
one-standard
deviation
X.
The
first
column
responding
change
presents
the results for a model
of solely initial leverage. Panel A reveals that
consisting
a one-standard
deviation
in a firm's initial book leverage
ratio (Initial
change
an
to
7%
in
of
future
values
of
book lever
average
Leverage)
corresponds
change
These
age. An even larger effect, 11%, is found for market
leverage.
findings
are consistent
with the behavior
in Figure
1.
illustrated
The

This content downloaded from 210.56.13.7 on Fri, 26 Feb 2016 14:08:45 UTC
All use subject to JSTOR Terms and Conditions

1586

The Journal

of Finance

two sets of determinants


into the specification.
The
Next, we incorporate
first set consists
of those variables
(1995) and
suggested
by Rajan and Zingales
used in many
studies
capital structure
subsequently
(e.g., Baker and Wurgler
and Zender
(2003), and Lemmon
(2002), Frank and Goyal
(2007)), augmented
are largely consistent
with calendar year fixed effects. The coefficient
estimates
in terms of sign and statistical
with previous
Neverthe
evidence,
significance.
and reveals a small (economic)
remains highly
less, Initial Leverage
significant
and 11% to 9% for mar
from 7% to 6% in the case of book leverage
change
ket leverage. Despite
the statistical
and economic magni
the change, however,
to the marginal
effects of the
tudes of these effects are dramatic when compared
is the single most
other determinants.
Initial Leverage
determinant
important
are consistent
in this specification.
structure
These
results
of future
capital
Table II
The
The

consists

sample

table
from

Effect

of Initial

of all nonfinancial

estimates,
presents
parameter
of book
OLS regressions

panel

on Future

Leverage

Leverage

in the Compustat
from 1965 to 2003. The
database
of the underlying
deviation
scaled by the standard
variable,
on several
The
different
and market
specifications.
leverage
firms

a one-standard
in leverage
associated
is the change
with
of each measure
interpretation
a one-standard
deviation
in the determinant.
For example,
in the first
column,
a 7% change
in book leverage.
Panel A presents
results
with
is associated
leverage
a subsample
of firms
results
Panel B presents
entire
(All Firms).
using
required
sample
change
initial

deviation
change

in

the
using
to survive

are trimmed
at the upper and lower 0.5-percentiles.
All variables
(Survivors).
are provided
denote whether
calendar
in the Appendix.
Year Fixed Effects
year
are computed
errors
are included
standard
The ?-statistics
in the specification.
fixed
effects
using
at the firm
level and heteroskedasticity.
to both clustering
robust
(i.e., dependence)
for at

least

20 years
definitions

Variable

Panel
Book
Variable
Initial

leverage

A: All

Firms
Market

Leverage

Leverage

0.07

0.06

0.04

0.11

0.09

0.07

(41.57)

(38.1)
0.02
(11.58)

(28.63)
0.03
(16.89)

(52.27)

(43.16)
0.02
(13.73)

(33.15)
0.03
(18.09)

Log(Sales)
Market-to-book

-0.02

-0.01

-0.06

-0.04

(-20.31)

(-12.11)

(-40.49)

(-35.68)

-0.03

Profitability

(-22.88)
0.04
(27.7)

Tangibility

(-30.89)
0.04
(24.55)

Industry median lev.


(42.63)

(46.27)

(-1.81)

(-3.35)

fixed

0.00

?0.03 ?0.05

payer
(-29.82)

(-24.16)
Year

(-30.03)
0.03
(15.92)
0.06 0.08
0.00

Cash flow vol.


Dividend

-0.04

-0.05

-0.03

(-23.78)
0.03
(17.94)

No

effects

0.13
Adj.i?2

Obs.

117,914

Yes
0.21

117,914

Yes
Yes
Yes No
0.30

117,914

0.20

117,300

0.42
0.34

117,300

117,300
(continued)

This content downloaded from 210.56.13.7 on Fri, 26 Feb 2016 14:08:45 UTC
All use subject to JSTOR Terms and Conditions

Table
Panel
Book

Variable
Initial

leverage

1587

to the Beginning

Back

II?Continued
B: Survivors
Market

Leverage

Leverage

0.07

0.05

0.04

0.10

0.07

0.05

(28.56)

(24.13)

(18.55)

(30.86)

(25.25)

(19.19)

0.02

Log(Sales)

(9.13)
-0.02

Market-to-book

(-10.75)

0.02

0.03

(11.45)

(11.78)
-0.06

-0.01
(-6.21)

Profitability

-0.03

-0.03

(-26.22)
-0.06

(-15.64)

(-19.8)

Tangibility

(-14.91)
0.03
(17.3)

Industry
flow

Cash

Dividend

median

lev.

vol.

payer

0.02

0.05

(10.59)
0.05

(18.86)

fixed

effects

Adj.i?2
Obs.

-0.04
(-22.83)
(-20.12)
0.02

-0.05

(10.65)
0.08

(26.52)
-0.01

(32.65)
-0.01

(-3.34)
-0.03

(-3.75)
-0.04

(-15.53)
Year

0.03

(12.47)

(-18.84)

No

Yes

Yes

No

Yes

0.15

0.23

0.32

0.17

0.37

0.45

68,736

68,736

68,736

68,224

68,224

68,224

Yes

in Figure
the regression
those presented
here
2, although
specification
a
more
of
in leverage differences
test
since the
presents
stringent
persistence
are allowed
determinants
to update
each period.
(other than initial leverage)
The final specification
additional
variables
in part,
motivated,
incorporates
an
Frank
and
who
of
exhaustive
(2004),
by
perform
Goyal
analysis
capital
structure
determinants.
the
effects,
significant
Despite
statistically
marginal
of these additional
inclusion
variables
does little to diminish
Initial Lever
on Initial Leverage
The estimated
coefficient
is still
age's relative
importance.
and larger in magnitude
than all other determinants
but for
highly
significant
The results emphasize
that historical
is an
Leverage.
Industry Median
leverage
even after controlling
determinant
of future leverage,
for traditional
important

with

sources

of variation.

The

in Panel B produce qualitatively


results for the Survivor
similar
sample
as one concern with the results of Panel
is reassuring
This similarity
findings.
A is that initial leverage
simply proxies for leverage
lagged only a few periods.
the
median
of
number
time-series
observations
for firms
in our Sur
However,
vivor sample is 30 years. This statistic
that
initial
implies
leverage corresponds
to a median
of y for the Survivor
lag time of 15 years. Thus, our estimate
sample
15 years ago is one of the most
of
determinants
implies that leverage
important
a
near
second
to
firm's
median
leverage
today,
only
contemporaneous
industry
leverage.

In concert with
future
termining

1 and 2, the importance


of firms' initial leverage
in de
Figures
an
values of leverage
the
First,
suggests
following.
important

This content downloaded from 210.56.13.7 on Fri, 26 Feb 2016 14:08:45 UTC
All use subject to JSTOR Terms and Conditions

1588

The Journal

of Finance

is a time-invariant
of leverage
from existing
component missing
specifications
seem to contain
determinants
lit
factor(s).
Second, most
relatively
existing
to this time-invariant
tle information
about corporate
relative
factor.
leverage
the possibility
exists
each variable
has an economically
that while
However,
as
a
determinants
whole have an economi
small impact of its own, the existing
on
it is unclear precisely
how
Further,
cally important
impact
capital structure.
this
important
other measurable

B. Variance

latent firm-specific
effect is, beyond
its importance
The next section addresses
determinants.
these

Decomposition

relative
issues.

to

of Leverage

of book and market


variance
decomposition
begin with a nonparametric
we
of
and
variation
the
withinbetween-firm
More
compute
leverage.
precisely,
are
12.9%
and
For
these
estimates
book
19.9%, respectively.
leverage,
leverage.
are 15.5% and 22.9%, respectively.
For market
these estimates
Thus,
leverage,
is approximately
50% larger than the within-firm
variation
the between-firm
this suggests
that leverage
of leverage.
variation
for both measures
Intuitively,
over time,
more
across
as
to
firms
varies
within
firms,
opposed
significantly
We

1 and 2.
in Figures
observed
the patterns
to a parametric
of covariance
framework,
analysis
in leverage attributable
the variation
which enables us to decompose
of leverage:
factors. We do so by estimating
the following model
with
consistent
We now turn

Leverageit

= a +
?Xit-i

+ m + vt + sit,

(ANCOVA),
to different

(2)

are as defined
in equa
where
rj is a firm fixed effect and the other variables
our
im
to
the relative
understand
tion (1). Because
goal at this stage is only
on
we
focus
in
determinants
of various
capturing
leverage variation,
portance
below.
for this analysis. We relax this restriction
static specifications
for several spec
the results of the variance
Table III presents
decompositions
to a different model
in the table corresponds
Each column
ifications.9
specifi
the last
in the body of the table, excluding
cation for leverage. The numbers
sum of squares for a
to the fraction of the total Type III partial
row, correspond
sum
we
for each effect
of
the
divide
That
model.10
squares
is,
partial
particular
across
in
This
sum
the
model.
all
effects
of
the
squares
aggregate
partial
by
sum
one.
to
to
a
columns
forces
the
that
normalization
Intuitively,
provides
sum
of the model
to are the fractions
in the table correspond
what
the values
9
Because
performing

of the large
the variance

number

and computer
in our panel
limitations,
memory
(19,700)
on the entire
As such, we randomly
is infeasible.
sample
on this subsample.
To minimize
the analysis
and perform

of firms

decomposition
in the panel

10% of the firms


100
the variance
and performing
of sampling
the process
decomposition
error, we repeat
sampling
the results.
times and average
10
to the
is sensitive
for two reasons.
III sum of squares
We use Type
First, Type I sum of squares
the dependent
involves
the computation
because
of the covariates
projecting
sequentially
ordering
of
our data are "unbalanced"
in the sense
that the number
onto each variable.
variable
Second,
than
observations
to each effect is not the same
(some firms have more
observations
corresponding
see Scheffe
of the methods
used here,
For a discussion
(1959).
others).
sample

This content downloaded from 210.56.13.7 on Fri, 26 Feb 2016 14:08:45 UTC
All use subject to JSTOR Terms and Conditions

Back

1589

to the Beginning
III

Table

Variance

Decompositions

in the Compustat
from 1965 to 2003. The table
database
for several different model
with adjusted
i?-squares
specifications,
decomposition
sum of squares
for each effect in the model
and then
at the bottom. We compute
the Type III partial
normalize
each estimate
by the sum across the effects, forcing each column to sum to one. For example,
sum of squares
4% of the explained
covariates
in model
(d) for book leverage,
by the included
captured

The

consists
sample
a variance
presents

can be attributed
Variable

definitions

Firm

FE
FE

firms

to LogiSales).
Firm FE are firm fixed
are provided
in the Appendix.

(a)

Variable

Year

of all nonfinancial

1.00

(b)

Book

Leverage

(c)

(d)

0.98

1.00

0.02

(e)

effects.

(f)

0.95

ig)
0.92

0.01

0.05

0.01

0.04

0.01

0.07

0.09

0.00

0.03

Profitability

0.11

0.01

Tangibility

0.27

0.01

Industry med
Cash flow vol
Dividend
Industry
Adj.i?2

lev

payer
FE
0.01

0.61

0.18

(a)

(b)

1.00

(c)
0.06

0.63

year fixed

effects.

Leverage

(d)

0.94
1.00

(e)

(f)

0.89

ig)
0.85

0.21

0.06

0.12

0.02

0.02

0.02

0.04

0.03

0.00

0.31

0.01

0.19

0.01

0.06

0.01

0.08

0.01

0.05

0.01

0.08

0.01

0.09

0.01

0.03

0.01

0.46

0.02

0.35

0.03

0.00

0.00

0.01

0.00

0.16

0.01

0.13

0.01

0.09

0.38
0.60

are calendar

Market

0.11

Log(Sales)
Market-to-book

Year FE

0.29

0.08

0.29
0.65

0.61

0.06

0.65

0.31

0.05

0.68

0.41

0.70

to particular
"effects" (e.g., firm, year, size, market-to
of squares attributable
one
in the model,
is
included
effect
the entire explained
book, etc.). When
only
sum of squares
to that effect. For example, when we examine
is attributable
takes on a value of
only the firm effect (r?i) in column (a), the reported estimate
the year effect
examines
1.00, or 100%. The same is true in column
(b), which
in isolation.
III presents
to
the adjusted
The last row of Table
i?-square
corresponding
in column
each specification.
effects alone, as displayed
(a), cap
Firm-specific
ture 60% of the variation
in book leverage, while
1%
the time effects capture
as shown
our previous
in column
of the variation,
consistent
with
(b). Thus,
in capital structure
of the total variation
the majority
is due to time
evidence,
invariant
is important
because
it suggests
factors.11 This finding
that theories
a
of capital
structure
based on volatile
in
time-series
sense, are un
factors,
structure
for
ratios
likely explanations
Rather,
capital
heterogeneity.
leverage
are relatively
stable over time.
Column
the results from the specification
(d) presents
inspired by Rajan and
and industry fixed
(1995) and shows that asset tangibility
Zingales
(Tangibility)
effects (Industry FE) account for most of the explanatory
power in this specifica
tion. However,
the adjusted
is 18% for book leverage
(31% for market
i?-square
and practically
than
lower, both statistically
leverage),
significantly
speaking,
11
An
leverage

paper
early working
in the U.S. has been

version

of Frank

remarkably

stable

and Goyal
(2005)
over time.

provides

evidence

that

This content downloaded from 210.56.13.7 on Fri, 26 Feb 2016 14:08:45 UTC
All use subject to JSTOR Terms and Conditions

aggregate

1590

The Journal

of Finance

the adjusted ?-square


from the initial firm fixed effect regression.
In column
we
the
in
firm
column
with
fixed
effects
and note
(d)
(e),
augment
specification
that the adjusted ?-square
for book leverage more than triples from 18% to 63%,
the significant
incremental
contribution
of the firm fixed effects.
highlighting
Columns
similar results using
the specification
(f) and (g) present
inspired
(2004). Including firm fixed effects into the book leverage
by Frank and Goyal
in the adjusted ?-square
increase
from
again leads to a substantial
specification
are
29% to 65%. The results using market
In
leverage
analogous.
unreported
the effect of further
to
the specification
results, we also examine
expanding
include measures
of asset volatility
and Petersen
(Faulkender
(2006)), proxies
for the marginal
tax rate (Graham
terms
(1996)), and higher-order
polynomial
of each determinant
to capture potential
nonlinear
All of these
associations.
modifications
have
little effect on our results.
In fact, a kitchen
sink model
that includes
and cubic terms for all of the determinants
linear, quadratic,
increases
the adjusted ?-square
mentioned
by less than 8% for book leverage.
At this point,
it is worth
the results
what
of the vari
clarifying
precisely
ance decomposition
an
the results
show that leverage
contains
imply. First,
ex
unobserved
that is not fully captured by
important
firm-specific
component
In other words,
for previously
identified
deter
isting determinants.
controlling
minants
does not alleviate
the concern over heterogeneous
result
intercepts?a
we
with potentially
for
studies
that
elaborate
important
implications
empirical
below. Second,
the variance
reinforces
the finding
that the ma
decomposition
as
in
of
to
the
total
is
due
variation
cross-sectional
differences,
jority
leverage
to
time-series
variation.
opposed
The variance
does not necessarily
that exist
decomposition
imply, however,
are
in
determinants
little
value
in
variation
of
ratios.
ing
explaining
leverage
comes from cross
If much
of the explanatory
determinants
power of existing
as opposed to time-series
sectional
then the importance
of
variation,
variation,
will necessarily
these determinants
fall as the firm fixed effects remove all such
across specifications
A comparison
of adjusted
variation.12
suggests
i?-squares
that this is indeed the case.
Column
that a specification
of only firm fixed effects
(a) of Table III reveals
an
of
60%.
that including
Column
(e) suggests
existing
yields
adjusted
i?-square
in this specification
boosts the adjusted
determinants
by only 3%, yet
i?-square
in leverage
18% of the variation
alone explain
determinants
(column
existing
comes
of
from
determinants
(d)). Thus, much of the explanatory
power
existing
as
to
the
variation.
However,
cross-sectional,
time-series,
explanatory
opposed
for the variation
falls well short of accounting
determinants
power of existing
raises the possi
(18% vs. 60%). This finding
captured
by the firm fixed effects
in
determinants
is simply "noise"
the
that
time-series
variation
standard
bility
are really con
which
about
around their long-run equilibrium
managers
levels,
cerned. We

12
See

the

examine

study

this possibility

by Griliches

and Mairesse

in the next

(1995)

subsection.

for a critical

discussion

of fixed

effects

tors.

This content downloaded from 210.56.13.7 on Fri, 26 Feb 2016 14:08:45 UTC
All use subject to JSTOR Terms and Conditions

estima

Back
C

Short-Run

versus Long-Run

1591

to the Beginning

Effects

are concerned
If managers
in the long-run
only about changes
equilibrium
it
level of leverage
then
is
that existing
determinants,
perhaps
unsurprising
in
such as equation
little of the variation
(2), explain
relatively
specifications,
on
assumes
X
that
the
effect
of
is
(2)
leverage. Equation
Leverage
delayed
by
one period and is "complete,"
in the sense that there are no persistent
effects of
are slow to respond to changes
inX on Leverage. However,
ifmanagers
changes
in X or if managers
in X, then equa
to changes
adjust Leverage
gradually
an incomplete
tion (2) provides
of
the
data generating
description
leverage
or
ifmanagers
short-term
fluctuations
process. Alternatively,
ignore
transitory
an
in the factors that determine
(2) provides
then, again, equation
leverage,
of capital structure.
Given
the relative
of
incomplete
description
infrequency
structure
these alternative
(2005)),
adjustments
(Leary and Roberts
capital
seem plausible.13
behaviors
we estimate
a distributed
To examine
these alternatives,
of lever
lag model
the traditional
in equation
(1) to incor
age, effectively
specification
expanding
variables:
porate deeper lags of the independent
n
Leverageit

= a +

]P ?sXit-s

+ yLeveragei0

+ vt + eit.

(3)

s=l

to the lag order of each independent


variable
contained
Here, n corresponds
in the vector X. Again, we assume
that s is potentially
heteroskedastic
and
correlated within firms, and we drop the first observation
for each firm to avoid
an identity at time zero.
To determine
the appropriate
two specification
lag length, we undertake
searches
the
Akaike
Criterion
In
Information
and
the Bayesian
(AIC)
using
a
formation
Criterion
of
(BIC). Both searches
suggest
lag length
eight periods,
4 and 10 peri
between
though there is little difference,
statistically
speaking,
ods. As such, we present
to n = 8; however,
in unreported
results corresponding
we also examine
the effect of alternative
analysis,
lag orders (e.g., 4 through
are not presented.
similar and therefore
10). The results are qualitatively
Rather
than presenting
all of the estimated
Table IV presents
coefficients,
two summary measures
for each independent
inX. The first measure
variable
as the product
is the scaled short-run
of the estimated
coef
impact, defined
ficient on the 1-year lag and the corresponding
variable's
standard
deviation.
This measure
the short-run
with a one
presents
change in leverage associated
in
a
deviation
standard
the
determinant.
For
one-standard
change
example,
in a firm's profitability
deviation
increase
has a near-immediate
(one-period
lag) impact of lowering book leverage by 2%.
The second measure
is the scaled long-run
impact, defined as the sum of the
estimated
coefficients
times
the
standard
of the corre
deviation
eight
slope
as the change
variable.
This measure
is interpreted
in the expecta
sponding
tion (or long-run, equilibrium
deviation
level) of Leverage
given a one-standard
13
We

thank

the associate

editor

for suggesting

this

possibility

and

inspiring

this

This content downloaded from 210.56.13.7 on Fri, 26 Feb 2016 14:08:45 UTC
All use subject to JSTOR Terms and Conditions

analysis.

1592

The Journal

of Finance

Table

The

consists

sample

firms

of all nonfinancial

measures
summary
presents
tributed
Each
lag specifications.
measures:
the short-run
summary

from panel

table

IV

Lag Model

A Distributed

of Leverage

in the Compustat
OLS regressions

from

database
of book

and market

1965

to 2003.

The

on dis
leverage
two scaled
present

variable
contains
eight
lags. We
and long-run multiplier,
each multiplied
multiplier
by the stan
on the
The short-run
determinant.
is the coefficient
multiplier

independent

of the corresponding
a one-standard
in leverage
and presents
the change
with
associated
deviation
lag variable
on all lags. The
in the determinant.
The long-run multiplier
is the sum of the coefficients
change
in the expectation
of each measure
is the change
(or long-run
level)
interpretation
equilibrium
a one-standard
of leverage
in the expectation
associated
with
deviation
(or long-run
change
equi

dard

deviation

1-year

a one-standard
For example,
in Profitability
deviation
change
in book
of 2% and a long-run
of 5%. All
leverage
change
change
are trimmed
at the upper
and lower 0.5-percentiles.
Year Fixed Effects
denote whether
are included
are computed
in the specification.
The ?-statistics
effects
year fixed
using
errors adjusted
are
at the firm level. Variable
for clustering
definitions
(i.e., dependence)

librium

of the determinant.

level)

is associated
variables
calendar
standard

with

a short-run

in the Appendix;
from a regression

provided
residuals

for this analysis,


however,
Log(Sales)
on a time trend.
of Log(Sales)
Book

Variable

Short

Initial

Run

Market

Long

Run

0.04
(4.74)

Market-to-book

Profitability

Tangibility

Cash

Dividend

Year

Run

0.00
(-3.81)
-0.02

(-5.44)
-0.05

(-10.98)
0.04

(-15.46)
0.02

0.04
(4.47)
-0.03

0.03
(9.3)
-0.06

(-17.92)
-0.03

(-20.05)
-0.08

(-14.84)
0.04

(-21.71)
0.03

(11.11)
0.07

(10.39)
0.08

vol.

(21.41)
-0.03

(22.01)
-0.01

(29.65)
-0.09

(23.48)
-0.02

payer

(-2.77)
-0.03

(-6.09)
-0.03

(-7.54)
-0.04

(-6.36)
-0.04

(-14.71)

(-11.12)

(-18.37)

(-14.21)

AdjJ?2
Obs.

Yes

Yes

0.29

0.43

55,355

54,963

k corresponds
in the expectation
of Xk, where
change
a one-standard
deviation
increase
nant.14 For example,
in expected Leverage
with a 5% decrease
is associated
14
For

Long

(8.59)
0.06

effects

fixed

Run

0.04

lev.

median

flow

Leverage

(14)
0.02
(9.3)
-0.01

(10.19)
Industry

Short

the

by using

0.04

(12.92)
Log(Sales)

de trended

Leverage

0.03

leverage

is first

to a particular
determi
in expected Profitability
in the full sample?an

and its
two conditions.
to be strictly
First, Leverage
true, we require
interpretation
this condition
is true
of LogiSales),
With
must
be jointly
the exception
stationary.
of
the stationarity
and Lin (1993). To ensure
to the testing
by Levin
strategy
according
proposed
on a time trend in our regressions.
of LogiSales)
from a regression
firm size, we use the residuals
is
than 8 years
effect of determinants
that the marginal
The second
condition
lagged more
requires
a
is
this
reasonable
zero. Casual
that
of
assumption.
suggests
regressions
inspection
higher-order
this

determinants

This content downloaded from 210.56.13.7 on Fri, 26 Feb 2016 14:08:45 UTC
All use subject to JSTOR Terms and Conditions

Back

to the Beginning

1593

twice as large as the short-run


impact of shocks to
the
of
the
effect
indicates
that
firms with persis
Moreover,
sign
profitability.
even
to
have
lower
tend
than those firms
tently high profitability
leverage
shocks to their profitability?consistent
with dynamic
transitory
experiencing
theories
(2001)).
(DeMarzo and Fishman
contracting
In many
is not necessarily
the long-run multiplier
instances,
than,
greater
or even different
This feature
is a result of an
from, the short-run multiplier.
That is, we place no constraints
unrestricted
lag distribution.
(e.g., polynomial,
on
Consider
the marginal
the
effect of cash flow
coefficients.
geometric)
slope
The short-run multiplier
in
that a one-standard
deviation
suggests
volatility.
an initial 3% decrease
crease
in cash flow volatility
in book
coincides with
the long-run multiplier
that the long-run
suggests
leverage, whereas
impact
an
of this change
is only 1%. This pattern
of
suggests
initially
large response
an
to
and
eventual
of
cash
flow
reduction
the
leverage
volatility
impact. Overall,
in the sense that some determinants
the results are somewhat mixed
exhibit
a stronger
to
in
short-run
whereas
changes
sensitivity
leverage determinants,
others exhibit a stronger
long-run sensitivity.
are as follows. First, de
the implications
of this analysis
For our purposes,
effect

that

is more

than

the dynamic
structure
of the model
spite expanding
along each determinant
initial leverage
is still economically
and statistically
in
dimension,
significant
every specification.
Thus, even when firms are allowed to respond to both short
run and long-run
shifts in leverage determinants,
initial leverage
ratios still
role in determining
future leverage
ratios.
play an important
the long-run
is relatively
small when
Second,
impact of each determinant
to the unconditional
standard
deviation
of leverage,
21% for book
compared
a one-standard
For example,
and 26% for market.
deviation
permanent
change
in median
the single most
determinant
of lever
industry
leverage,
important
or long-run, book leverage. This
to a 6% change
in expected,
age, corresponds
is less than 29% (6%/21%) of the typical unconditional
estimate
in
variation
In
in
other
the
levels
of
words,
leverage.
large changes
long-run
equilibrium
the determinants
lead to relatively
small changes
in the unconditional
expec
tation of leverage.15
In sum, previously
account for relatively
identified
determinants
little of the
in leverage, regardless
one takes a short-run
or a long-run
variation
of whether
of a significant
unobserved
perspective.
Coupled with the presence
firm-specific
effect, these findings
suggest that the identification
studies
is more difficult
than implied by previous
IV. Implications

for Empirical

Studies

in capital

problem
research.

of Capital

structure

Structure

of our analysis
in the previous
section is that simple pooled
implication
least
of leverage
of
(OLS) regressions
squares
ordinary
ratios, the workhorses
are likely to be misspecified
the empirical
because
capital structure
literature,
One

15
Unreported
similar
itatively

estimates

from

the sample

of firms

that

survive

for at

least

20 years

findings.

This content downloaded from 210.56.13.7 on Fri, 26 Feb 2016 14:08:45 UTC
All use subject to JSTOR Terms and Conditions

reveal

qual

1594

The Journal

of Finance

time-invariant
ratios that is
of leverage
component
they ignore a significant
with
correlated
traditional
side
variables.
The
of
presence
likely
right-hand
one
concern
this unobserved
of
that
component
suggests
leverage
potential
with existing
estimates
of capital structure
is that the parameter
regressions
estimates
and inferences
drawn from the data may be tainted by omitted vari
able bias (Arellano
in tech
differences
(2003) and Hsiao
(2003)). For example,
market
and/or
behavior
motivated
the
have
power,
nologies,
managerial
long
firm
in
investment
Kuh
of
fixed
effects
and
(1963))
incorporation
(e.g.,
produc
tion function
(1962) and Hoch
(1962)) as estimates
regressions
(e.g., Mundlak
to obtain. In so far as these unobserved
of such factors are difficult
factors are
if not strictly time-invariant,
their impact on capital structure
slowly changing,
is mostly

absorbed
effect.
by the firm-specific
1 and
the unidentified
revealed
component
Similarly,
transitory
by Figures
2 may also have consequences
for traditional
OLS leverage
The
regressions.
and can adversely
of this component will lead to inefficient
estimates
presence
inference
affect statistical
(1993)). In so far as financial
(e.g., Greene
adjust
are
or autocorrelated
ment
economic
shocks
is costly,
persistent,
independent
one might
are omitted from the specification,
of
variables
the presence
suspect
in the error structure.
serial correlation
To explore
the importance
of these considerations,
Table V presents
the re
a
structure
OLS
sults of estimating
regressions
using
pooled
capital
approach
in the error structure,
effects and serial correlation
that ignores firm-specific
and potentially
from using a fixed effect specification
with the results
serially
errors. More
the
OLS
estimates
correlated
equa
pooled
approach
precisely,
and assumes
that the errors are possibly
tion (1), excluding
initial leverage,
heteroskedastic
effect estimation

within

and equicorrelated
estimates

Leverageit

ot +

firms

?Xu-i

(Petersen

r}? -j- vt +

uit,

(2007)).

The

fixed

(4)

where
Uit

PUit-l

(Oit,

u is assumed
to be serially
and cross
to be stationary,
and co is assumed
heteroskedastic.
uncorrelated
but
possibly
sectionally
are highly
illustrate
The results
that most determinants
statistically
signifi
the estimated
of
model
the
However,
cant, regardless
magnitudes
specification.
are very sensitive
the estimated
serial corre
to the specification.
Additionally,
lation coefficient,
while bounded well below one, is statistically
large for both
with a gradual
consistent
book (0.66) and market
(0.65) leverage,
decay in the
on
the
book
shocks.
of
every coeffi
results,
leverage
Focusing
leverage
impact
from the
in magnitude
estimate
moving
fairly large declines
experiences
in the
The
coefficients
to
the
fixed
effect
OLS
specification.
pooled
specification
on
82%
decline
book (market)
(62%),
by approximately
leverage
specification
cient

average.

This content downloaded from 210.56.13.7 on Fri, 26 Feb 2016 14:08:45 UTC
All use subject to JSTOR Terms and Conditions

Back

Table

Parameter

1595

to the Beginning

Sensitivities

to Model

Specification

in the Compustat
database
from 1965 to 2003. The
in parentheses
for book and market
table presents
estimates
and ?-statistics
parameter
leverage
are presented
corre
two sets of parameter
estimates
For each measure
of leverage,
regressions.
errors for the
to a pooled OLS regression
and a firm fixed effects
The standard
regression.
sponding
are robust
to heteroskedasticity
firm equicorrelation.
and within
The stan
pooled OLS regression
are robust
to heteroskedasticity
effects
dard errors for the firm fixed
and within-firm
regression

The

consists

sample

of all nonfinancial

firms

is the percent
in the magnitude
of the coefficient
correlation.
Also presented
when
change
the model
from Pooled OLS to Firm FE. AR(1)
is the estimated
first-order
changing
specification
are provided
coefficient.
Variable
in the Appendix.
serial correlation
definitions
serial

Book

Pooled OLS

Variable
Log( Sales)
Market-to-book

Profitability

Tangibility
Industry
Cash

flow

Dividend

Year

fixed

Adj. R2

lev.

median
vol.

payer

effects

0.013

0.008
(8.37)
-0.003

(-14.01)
-0.159

(-7.14)

(-23.93)

(-10.41)

-0.032

0.160

0.076

(20.05)

(15.07)

0.569

0.046

(48.31)
-0.031

(8.05)

(-2.71)
-0.084

(1.11)

(-28.22)

(0.02)

0.020
0.000

Yes

0.262

AR(1)
Obs.

FirmFE

(16.96)
-0.010

Yes

Market

Leverage

% Change
-41%
-74%

Pooled OLS
0.016
(16.86)
-0.039
(-36.16)

-80%

-0.232
(-30.38)

0.024
(21.17)
-0.001

% Change
52%
-96%

(-3.23)
-0.053

-77%
-36%
-88%

-52%

0.156

(-14.13)
0.099

-92%

(16.7)
0.544

(15.76)
0.065

(52.53)

(12.59)

-163%

-0.099

0.010

-110%

-100%

(-8.14)
-0.111

(0.48)
0.002

-102%

(-31.42)

(0.98)

Yes

Yes

0.376
0.660

106,097

Leverage

Firm FE

106,097

0.653
105,532

105,532

are striking; however,


one must
of these differences
The magnitudes
take
care with their interpretation,
much
like the results of the variance
decomposi
tion (Griliches and Mairesse
removes
(1995)). The fixed effects transformation
some
the between
of
which
is
determinants.
variation,
captured
by existing
the variance
shows that the fixed effects capture
Nonetheless,
decomposition
more of the variation
in leverage than can be attributed
to existing
significantly
determinants.
the distributed
model
shows
that
neither
short
Additionally,
lag
run fluctuations
nor long-run equilibrium
are likely responsible
considerations
for this discrepancy.
determinants
do not adequately
Therefore,
proxy
existing
or the transitory
for either the permanent
omitted
of leverage.
components
The implication
of ignoring the permanent
of the data in empirical
component
is summarized
in Hsiao
models
(2003, p. 8): "Ignoring the individual
[i.e., firm]
units but are not captured
specific effects that exist among cross-sectional
by
can lead to... inconsistent
the included explanatory
or meaningless
variables
estimates
of interesting
While
somewhat
less severe,
the finite
parameters."

This content downloaded from 210.56.13.7 on Fri, 26 Feb 2016 14:08:45 UTC
All use subject to JSTOR Terms and Conditions

1596

The Journal

of Finance

errors can be significant


for statistical
of serially correlated
sample implications
inference
(Greene (1993)).
To be clear, this is not to say that all empirical
should employ
specifications
or least-squares
the firm fixed effects
esti
variable)
(a.k.a., within,
dummy
errors. These
error
mator with
to
serial correlated
the
assumptions
relegate
structure
of the leverage process
features
need to
that, ultimately,
prominent
the within
estimator
be understood.
sweeps out all of the cross-sectional
Worse,
so that the model
cannot identify what
variation
is responsible
for the majority
in leverage ratios. Rather,
of the variation
the model
decision must
specification
If the goal is to identify the marginal
effects
depend on the goal of the research.
then firm fixed effects offer one of several alterna
of a particular
determinant,
as
tives (e.g., differencing,
structural
estimation
estimation,
quasi-structural
concerns
over
to
in Olley and Pakes
natural
address
etc.)
(1996),
experiments,
errors are but one of several
omitted
variables.
correlated
Similarly,
serially
cor
alternatives
"new" serially
variable,
(e.g., lagged dependent
differencing,
autocorrelation
etc.) for addressing
variables,
explanatory
is
confidence
what
these
greater
Ultimately,
provide
approaches
fication
of marginal
effects.
related

V. What
A.

Is "Active" Financial

Lies

behind

Management

the Transitory
behind

in leverage.
in the identi

Component?

the Convergence?

con
2 illustrates
that leverage
ratios in the tails of the distribution
Figure
more
That
moderate
levels
of
there
is
toward
is,
verge significantly
leverage.
an important
A
of
ratios.
number
unobserved
component
transitory
leverage
their leverage
ratios to
that firms actively manage
of recent studies
suggest
an optimal or target level of leverage
Graham
and
maintain
(2001),
(e.g.,
Harvey
(2005), Hovakimian
(2001), Leary and Roberts
Hovakimian,
Opler, and Titman
and Titman
and Rangan
(2007)). Simulation
(2006), Kayhan
(2006), Flannery
in Shyam-Sunder
and Myers
evidence
(1999) and Chang and Dasgupta
(2006),
with active
in equating mean-reversion
restraint
leverage
however,
suggests
on whether
the
In this section, we provide
evidence
additional
management.
are a function
toward
of active management
in our figures
apparent
dynamics
behavior.
ratios or of more passive
desired
leverage
we examine
the net security
between
these two possibilities,
To distinguish
in
described
four
the
of
firms
issuance
leverage portfolios
unexpected
activity
ease the
are presented
A
To
B.
Panels
and
in Figure
earlier. The results
3,
on
we
net debt
first
intervals.
the
confidence
suppress
Focusing
presentation,
we
to
the
issue
debt no
find
issuance activity
(Panel A),
that, initially,
tendency
is that the propensity
is interesting
ticeably differs across the portfolios. What
ra
to
firms'
related
to issue (net) debt is monotonically
leverage
negatively
as
we
move
more
net
debt
tios. Firms undertake
issuing activity
progressively
remain for
to the Low portfolio. These differences
from the Very High portfolio
This
is consis
3 to 5 years before becoming
finding
largely indistinguishable.
that
who
tent with Leary and Roberts
(2005) and Hovakimian
(2006),
suggest

This content downloaded from 210.56.13.7 on Fri, 26 Feb 2016 14:08:45 UTC
All use subject to JSTOR Terms and Conditions

1597

to the Beginning

Back
Paiwl A;^t

Debt IsstungActivtty

A4

*fl?i

1^

'
.1 2 '.'a' 4- 5 .?'' *: 8' ?-^.i^?O^?t???^;^
^-.?^t?L
^^^

^ "?^^V1^"^

in event
time. The sample
of unexpected
leverage
portfolios
in the Compustat
database
from 1965 to 2003. Panel A (B) presents
net debt
issuance
of period
assets
for each
scaled by beginning
the average
(net equity)
activity
That
of four unexpected
year, we form four portfolios
is, for each calendar
by
portfolios.
leverage
on their unexpected
the portfolios
fixed for
firms
based
(defined
below). Holding
leverage
ranking

Figure
consists

the next

3.

behavior
Financing
firms
of all nonfinancial

net
the average
20 years, we compute
in 1975 we sort firms
For example,

portfolio.
ratios. For

1975
year from
We
four portfolios.
After
horizon.
performing
each

to 1994,

we

debt
into
compute

scaled by assets
for each
issuances
on the unexpected
based
groups
leverage
net debt
issued
for
the average
(net equity)

(net equity)
four

for every year in our


and averaging
1965 to 2003, we
for each year from
across
Net
to obtain
net issuances
the lines in the figure.
the average
as the change
t
1 to t divided
in total debt from
issued
is defined
debt
by total assets
period
r - 1. Net
as the split-adjusted
in shares
in period
is defined
issued
change
outstanding
equity
in period
t
t Hot
the year divided
times
share price during
the average
from
by total assets
each

of these

sample
then average

repeat
this

this

process
and
sorting

of sorting

averaging
"event time"

a cross-sectional
as the residuals
from
of leverage
is defined
regression
leverage
are
variables
where
all independent
and tangibility,
market-to-book,
profitability,
are
and
French
in
indicator
variables
1
Also
included
the
(Fama
year.
industry
regression
lagged
are provided
in the Appendix.
Variable
definitions
classifications).
38-industry
1. Unexpected
firm size,

on

This content downloaded from 210.56.13.7 on Fri, 26 Feb 2016 14:08:45 UTC
All use subject to JSTOR Terms and Conditions

1598

The Journal

of Finance

an important motivation
behind debt policy is capital structure
It
rebalancing.
also helps
the
mechanism
behind
the
initial
of
convergence
identify
leverage
ratios (i.e., the transitory
of leverage)
in Figures
1 and 2.
observed
component
Panel B presents
the average net equity issuing activity across the four port
folios and reveals a different
firms with Low leverage are
story.16 In particular,
more
to
issue
This
counterintuitive
result
appears
significantly
likely
equity.
for a rebalancing
because
these
firms
low
have
story; however,
leverage to begin
cases zero leverage?net
with?in
issuances
have
little or no effect
many
equity
on their capital structure
are largely irrelevant
in terms of their
and therefore
distribution
of leverage. We also note that firms
impact on the cross-sectional
with Very High
amount of equity, on aver
leverage appear to issue a significant
to the Medium
and Low portfolios.
This finding
is consistent
with
age, relative
levered firms using equity to reduce their leverage
(Lemmon and
(very) highly
in leverage
Zender
the initial decline
(2007)) and also helps to further explain
1 and 2. There
in Figures
is little systematic
for the Very High portfolio
differ
ence between
to the other two
the Medium
and High portfolios,
though relative
these firms appear to use less equity on average.
portfolios
for the traditional
This analysis
indicates
deter
that, even after controlling
an
in
current
state variable
minants
of capital structure,
is
leverage
important
net security
issuance
decisions.
That is, the transitory
component
correspond
1 and 2 appears
to be driven,
of leverage ratios in Figures
ing to the convergence
at least in part, by active rebalancing.
random
Therefore,
(Chang and
although
or
financ
order
and
(1999))
(2006))
(Shyam-Sunder
Myers
Dasgupta
pecking
can
mean
in
reversion
cash
flows
lever
schemes
with
generate
ing
coupled
are
in
1
to
similar
those
observed
and
age paths
2, they
Figures
ultimately
are
an important
feature
of the data?financing
inconsistent
with
decisions
to follow the pecking
and
not random nor do they appear
order (see Helwege
(2003), Fama and French
(2005), and Leary and
Liang (1996), Frank and Goyal
are
our results
illustrate
that firms' financial
Roberts
(2006)). Overall,
policies
ratios relatively
their leverage
at least partially
toward maintaining
geared
are concerned
that managers
close to their long-run mean
values,
suggesting
with

their

leverage

B. Are Firms

ratios.

Adjusting

Toward

a Moving

Target1?

lever
of time-varying
the importance
Several recent studies have emphasized
Ran
and
Titman
and
and
(2001)
age targets
Hovakimian,
Flannery
Opler,
(e.g.,
our findings, which
this result would appear to contradict
gan (2006)). However,
in general,
indeed leverage
level of leverage,
instead suggest
that any desired
As such, we take a closer look at this claim.
is largely time-invariant.
a direct way to quantify
of time-varying
effects
the importance
Specifically,
the impact that they have on the es
is to examine
for identifying
target leverage
In particular,
model.
timated speed of adjustment
(SOA) in a partial adjustment
are a crucial component
of firms' target leverage,
if time-varying
characteristics
16
See Dittmar
retire

and

issue

(2000)
equity,

and Dittmar

and Thakor

(2007)

for recent

investigations

into why

respectively.

This content downloaded from 210.56.13.7 on Fri, 26 Feb 2016 14:08:45 UTC
All use subject to JSTOR Terms and Conditions

firms

1599

to the Beginning

Back

as previous
studies
then omitting
them will substantially
reduce the
suggest,
SOA since firms will be adjusting
to a target that is different
estimated
from
This
the one modeled
the
econometrician.
exercise
is
to
by
equivalent
adding
error to the target, precisely
as Flannery
measurement
and Rangan
(2006) do
to illustrate
in adjustment
the reduction
in
decreases
speed that accompanies
ratio of the target estimate.
the signal-to-noise
Table VI presents
the raw (i.e., unsealed)
estimates
and ?-statistics
parameter
from estimating
several different partial adjustment
for leverage,
specifications

Table VI
of Adjustment

Speed
The

sample

table

consists

provides

of all nonfinancial

estimates

firms

for the following

^Leverageit

in the Compustat
model

autoregressive

= a +
Xi?Xit?i

database
of book
+

Leverageit_i)

r??+

from

1965

to 2003.

The

leverage:
vt +

ea,

is a set of pre-determined
v is a firm-invariant
effect

effect potentially
correlated
r\ is a time-invariant
covariates,
error term that
correlated
with X, and e is a random
potentially
is potentially
heteroskedastic
and autocorrelated
within
firms.
Columns
(a) through
(c) present
least squares
estimates
that restrict
OLS estimates
n? = 0. Columns
(d) and (e) present
ordinary
after transforming
means.
the original
data into deviations
from firm-specific
Columns
(f) and (g)
or SOA,
GMM
and Bond
estimates.
The
of adjustment,
(1998))
(Blundell
present
system
speed
X

where
with

X,

is given
errors
are computed
us
estimates
of the ?'s, where
standard
by ?. The table presents
as the time (in
half-life
is defined
(i.e., the delta method).
ing a Taylor
approximation
Leverage
that it takes a firm to adjust
back to the target
after a one-unit
shock to e, ln(0.5)/
years)
leverage
?
are provided
are in parentheses
definitions
in the Appendix.
The ?-statistics
and
?). Variable
ln(l
are robust
to heteroskedasticity
and within-firm
correlation.

Pooled OLS
Variable

(a)

SOA
Initial

0.13
(63.8)
leverage

Firm Fixed Effects

(b)

(c)

(d)

(e)

(f)

(g)

0.15
(63.31)

0.17
(62.92)

0.36
(80.05)

0.39
(79.19)

0.22
(27.44)

0.25
(29.36)

0.25

0.17

(23.97)

(18.34)
0.00
0.01

Log(Sales)
(-2.6)

(5.98)
-0.01 0.00

Market-to-book
(-4.89)

(-4.47)
-0.18 -0.10

Profitability
(-11.97)

lev.

Half-life

0.42 0.13

Year fixed effects

4.96
(59.45)
No

4.35
(58.4)
No

R2
Obs.

0.72
145,726

0.72
145,726

(24.11)
3.62
(57.09)

(-4.87)
-0.01
(-6.02)
0.01

(-9.19)

(10.05)
median

0.00

(0.63)

0.090.11 0.10
(8.81)

Tangibility
Industry

GMM

(12.36)

0.32

(9.01)

Yes

1.53
(63.32)
No

1.42
(61.42)
Yes

2.82
(24.2)
No

0.72
145,726

0.72
145,726

0.71
145,726

145,726

This content downloaded from 210.56.13.7 on Fri, 26 Feb 2016 14:08:45 UTC
All use subject to JSTOR Terms and Conditions

(18.79)
2.41
(25.34)
Yes
145,726

1600

The Journal
ALeverageit

= a0 +
\(p*t

of Finance
Leverage^)

+ e??, (5)

where
fi*t

?Xit-i

+ ^ + vt

are as defined earlier. The parameter


is "target" leverage and all other variables
measures
of
last period's
the
the
the
fraction
gap between
?,
SOA,
leverage and
measure
firms
each
this period's
that
close
The
target
? parameters
period.
in the determinants,
and the standard
the sensitivity
of the target to variation
are computed
errors for these parameters
(the "delta"
using a Taylor expansion
method).17
We present
results from three estimation
(pooled OLS, firm fixed
techniques
our results
of moments
method
GMM) to benchmark
effects, and generalized
our conclusions.
and reinforce
with previous
studies
Column
(a) shows that
term
to a population
constant
the speed of adjustment
(i.e., a single intercept
is approximately
13% per year. Columns
in the target specification)
(b) and
coefficient
estimates
the Pooled OLS heading,
(c), under
present
using firms'
show
of target leverage. These results
initial leverage
ratios as a determinant
even
in
firms'
initial
the
of
presence
that,
lagged leverage,
leverage
one-period
ratios or,
determinants
of future
ratios are statistically
leverage
significant
In fact, column
of firms'
(c) shows that initial
targets.
alternatively,
leverage
an
determinant
of future
is
also
leverage,
economically
significant
leverage
more
most
other
the
market-to-book
than
ratio,
factors,
including
important
asset tangibility,
and firm size. (This can be seen more clearly by multiplying
of each regressor
deviation
by its standard
see
levered
firms
that
(low)
Thus, again,
initially high
level of leverage,
for the current
such even after controlling
the

coefficient

we

in near

contemporaneous

found in Table
I.)
as
tend to remain
as well as changes

determinants.

adjust
Next, we note that the Pooled OLS and firm fixed effects estimated
ment
those found in Fama and French
(2002) and
(SOA) closely match
speeds
that firms close 13% to
and Rangan
(2006), respectively,
suggesting
Flannery
or
and this period's
last
36% to 39% of the gap between
17%
leverage
period's
re
to
firm
the
fixed
effects
the
OLS
from
Pooled
regression
target. Moving
22% (39-17%)
of approximately
in the speed of adjustment
sults in an increase
terms ((39-17%)/17%),
consistent
terms and over 129% in relative
in absolute
effects for identifying
of firm-specific
with the importance
leverage targets. This
is Flannery
and Rangan's
(2006) insight.
finding
we see that the difference
two
methods
these
estimation
within
However,
determinants
of
from
the
inclusion
in adjustment
time-varying
speeds arising
in the SOA mov
the change
Consider
is negligible.
in the target specification
that all firms target a fraction
(b) assumes
(b) to (c). Column
ing from column
does not change over time.
of their initial leverage
definition,
ratio, which,
by
17
we are not using maximum
Because
likelihood,
derive ? by dividing
of ? and ?, from which we must
we are forced to use a Taylor
to derive
expansion

the estimation
by the estimated
the appropriate

the product
estimates
procedure
k. Because
of this nonlinearity,
error and ?-statistic.
standard

This content downloaded from 210.56.13.7 on Fri, 26 Feb 2016 14:08:45 UTC
All use subject to JSTOR Terms and Conditions

Back

1601

to the Beginning

the speed of adjustment


determinants
increases
from 15%
Adding
time-varying
to 17%, a difference
of only 2%. In the firm fixed effects regressions,
including
the speed of adjustment
increases
from 36% to 39%,
determinants
time-varying
a difference
to the differences
in adjustment
of only 3%. Relative
speeds result
are
firm fixed effects
into the target, these differences
ing from incorporating
is most
for identifying
is sim
Thus, what
important
target leverage
negligible.
or
to firm characteristics
ratios, as opposed
ply firms' long-run mean
leverage
our
with
macroeconomic
consistent
earlier
results
the
rela
factors,
highlighting
in capital structures
variation
and the low explanatory
tively small time-series
determinants.
power of traditional
While
the Pooled OLS and firm fixed effects regressions
enable us to directly
our
with
both
from potentially
results
suffer
compare
studies,
they
previous
severe biases, particularly
to the speed of adjustment
with respect
(and, con
is derived from the estimated
SOA). Specif
sequently, with respect to ?, which
that
the
of fixed
OLS
presence
ically, pooled
regressions
(incorrectly)
ignore
in an estimated
effects will result
SOA that is biased downward,
while firm
fixed effects regressions
will result in an estimated
SOA that is biased upward
results using
(Hsiao (2003)). As such, columns
(f) and (g) in Table V present
and Bond (1998)), which
is specifically
de
(Blundell
system GMM estimation
concerns associated
with estimating
signed to address the econometric
dynamic
in the presence
of firm fixed effects.18 The results
confirm
panel data models
that including
in the target specification
characteristics
has little
time-varying
effect on the estimated
(22% vs. 25%) and, consequently,
speed of adjustment
contributes
little toward the identification
of firms' target leverage ratios. Thus,
our finding
these results
that any target leverage
further reinforce
is largely
are of limited use in explaining
time-invariant
and that existing
determinants
in leverage.
cross-sectional
variation
The GMM method
also reveals a less extreme
estimate
of the speed of adjust
to
the
of
the
Pooled
OLS and firm fixed
ment,
equal
approximately
midpoint
effects estimates.
This finding
is comforting
in that the biases
in the pooled
and fixed effects estimates
that the true
go in opposite
directions,
suggesting
lies somewhere
between
these two extremes.
This distinction
from
parameter
and Rangan's
is due to differences
in the instruments
(2006) estimate
Flannery
and highlights
the difficulty
in ob
used, as well as in the estimation
strategies,
in these models.19
estimates
taining accurate
parameter
In sum,
these
results
that
is mean-reverting
around
suggest
leverage
a largely
to the conclusions
of
time-invariant,
level, contrary
firm-specific
18

In particular,
the system GMM approach
includes
variable
levels, as well
set to address
instrument
the problem
of persistent
when
regressors,
which,
little information
for parameter
identification.
19
The system GMM utilizes
levels and differences
of the dependent
lagged
nous variables.
as an instrument,
and Rangan
Flannery
rely on book leverage
a weak

as differences,

variable
which

in the
contain

differenced,
and
we

exoge
is
suspect
the within

are very close to those found using


their results
surprisingly,
Fixed Effects). We also note that small sample
biases
(Huang and Ritter
and Engel
be relevant
for the specification
(2007)) and
(Caballero
and
(2004)) may
estimation
of these models,
these
issues
fall outside
the scope of this study.
though
instrument.

transformation

Not

(i.e., Firm
nonlinearities

This content downloaded from 210.56.13.7 on Fri, 26 Feb 2016 14:08:45 UTC
All use subject to JSTOR Terms and Conditions

1602

The Journal

of Finance

and Rangan
(2001) and Flannery
(2006). While
Hovakimian,
Opler, and Titman
we are hesitant
to draw any inferences
based solely on this mean-reversion?for
reasons
our results
in Section V.A (and the evidence
discussed
earlier?
from
recent studies
cited therein)
that
this
mean-reversion
is
suggest
accomplished,
at least in part, by active management
of leverage
ratios via net debt issuing
activity.
VI. What

Lies

behind

the Permanent

Component?

The key question


at this point is:What
is the economic mechanism
generating
the unobserved
revealed
permanent
component
by Figure 2? Or, less formally,
seem to have high
others always
why do some firms always
leverage, while
seem to have low leverage, despite being similar along many
dimensions
(e.g.,
a complete
answer
to this question
is
size, profitability,
industry, etc.)? While
beyond the scope of this paper, we take a step toward this goal by investigating
in leverage persist
whether
back in time.
differences
A. How

Far Back Does

the Firm-Specific

Effect

Go?

To answer
this question, we form unexpected
at the time
leverage portfolios
in the number
of the IPO. Because
of a significant
reduction
of observations
in Figure
the analysis
2.
5,000 IPO firms), we modify
slightly
(approximately
We begin by computing
initial leverage as the average
of the first three public
on leverage
in event years 0 (year of the IPO), 1, and 2. We com
observations
of
the
determinants
initial
values
(size, market-to-book,
pute
corresponding
manner.
a
in
similar
and
helps minimize
tangibility)
Averaging
profitability,
in this smaller
observations
noise and mitigate
the effect of extreme
sample.
on the initial values
a regression
for the
of initial leverage
We then estimate
as
as well as indicator variables
for calendar year and industry
determinants,
the differen
of the time of the IPO. The calendar year variables
help address
in the regression.
The residuals
cold markets
tial effect of IPOs in hot versus
are then used to sort firms
into four unexpected
from this regression
leverage
as described
earlier.
portfolios,
4 plots the average
interval
for
and 95% confidence
actual
leverage
Figure
in event time, where
event year 0 is the year of
each of the four portfolios
eliminates
the regression
the IPO. We note the following.
any mean
First,
the results
the Low and Medium
between
Second,
portfolios.
ingful difference
are somewhat
in
because
of the
to
those
noisier
2,
Figure
primarily
compared
and
of observations.
the general
reduced number
however,
Overall,
patterns
some
are
in
differences
initial
similar.
convergence,
quite
Despite
implications
some
across
for
book
time.
More
for
firms
importantly,
quite
persist
leverage
are established
prior to the IPO. Firms
leverage, we see that these differences
with high (low) leverage as private firms also have high (low) leverage as public
in the distribution
of
It therefore
that the significant
firms.20
appears
changes
20
Unreported

results

for the

sample

of IPO firms

surviving

for at

least

20 years

patterns.

This content downloaded from 210.56.13.7 on Fri, 26 Feb 2016 14:08:45 UTC
All use subject to JSTOR Terms and Conditions

reveal

similar

in event
time
(IPO sub
portfolios
in the Compustat
from 1965 to
database
of four portfolios
the average
presents
leverage
are formed based
on quartiles
IPO. Portfolios
of f?rms, un

of unexpected
leverage
Average
consists
The sample
of all nonfinancial
sample).
we have an IPO date. Each panel
2003 for which

Figure

1603

to the Beginning

Back

4.

leverage
firms

in event
year 0 is the firm's
time, where
as the residuals
of initial
initial
from a cross-sectional
ratios, defined
regression
expected
leverage
on initial values
and tangibility.
Also
for leverage
for firm size, profitability,
values
market-to-book,
are year and industry
in the regression
indicator
variables
included
(Fama and French
38-industry
are provided
in the Appendix.
The initial values
for continuous
definitions
Variable
classifications).
variables
ing the

as the average
over the first
are computed
in the regression
variables
IPO year). Binary
for each quartile
and event year, measured

leverage
A and B present

results

for book

and market

leverage,

3 years
of the firm's public
existence
are measured
at the IPO date. The
from

the

IPO,

is plotted

in the figure.

(includ
average
Panels

respectively.

and the access to capital markets


that ac
control, the information
environment,
costs and benefits
that determine
the IPO do little to alter the relative
company
ratios.
firms' preferred
leverage
While
that firms maintain
their pre-IPO
the evidence
rankings, we
suggests
concern.
in the year prior to the
the following
Leverage
acknowledge
potential
as a private firm, more generally.
of leverage
IPO may not be representative
structure
of the firm may
That
is, in the year prior to the IPO, the capital
we make
two remarks
in anticipation
of this event. However,
have been altered
concern.
to this
in book (as well as market)
in response
the differences
First,
are
across
in event year -1
the portfolios
very large: 18% between
leverage

the Very High and High portfolios and 12% between the High and Medium
that firms alter their capital structure
it is unlikely
by
portfolios. We believe
or
in
two
this belief is
the year
the IPO. Second,
these magnitudes
preceding
in Kaplan,
(2005), who show
Sensoy, and Stromberg
supported
by the evidence
and investment
the operations,
assets,
that, other than human
policy
capital,
of firms remain very stable from the literal birth of a firm to several years after
the IPO.
are limited,
in the United
States
the FAME
data on private firms
Although
van
on
a
number
of pri
data
database
from Bureau
significant
Dijk contains
vate firms in the United Kingdom.
(1995) argue that, except
Rajan and Zingales
in the rights
in bankruptcy
of creditors
for differences
(driven by differences
in bankruptcy
the
United
States
and
the
United
U.K.
laws between
Kingdom),
in an environment
firms operate
that is similar to that of U.S. firms. This fact
This content downloaded from 210.56.13.7 on Fri, 26 Feb 2016 14:08:45 UTC
All use subject to JSTOR Terms and Conditions

1604

The Journal

Panel A: Ujwxp?cted Bb?k


I^?w^

of Finance

Portfolk? (All Finns)

Panel B: Unexpected Book Leverage Portfolios (Surri^

"

5. Average
Figure
leverage
held U.K.
The sample
firms).

MtmvnmKwnii

of unexpected
leverage
portfolios
consists
of nonfinancial
held
privately

Firma)

?ft?itefn?*
in event

(unlisted)

time
firms

(privately
in the United

Kingdom in the FAME database during the period from 1993 to 2002. Each Panel presents the
book

average
period.

That

of four portfolios
leverage
year,
is, for each calendar

in event
we

0 is the portfolio
formation
time, where
year
on their
four portfolios
firms
based
by ranking
the portfolios
fixed
for the entire
sample
period, we

form

(defined
below). Holding
leverage
as the
the average
for each portfolio.
actual book leverage
is defined
Unexpected
leverage
on firm size, profitability,
from a cross-sectional
of leverage
market-to-book,
regression
are lagged
and tangibility,
1 year. Also
in the regression
where
all independent
variables
included
are industry
indicator
variables
Variable
definitions
(Fama and French
classifications).
38-industry

unexpected
compute
residuals

are provided
in the Appendix.
The average
for each quartile
leverage
the figure.
Panel A presents
the results
for the sample
of all private
results
for the sample
of all private
firms that have nonmissing
data

in
event year
is plotted
B presents
the
firms; Panel
for the entire
20-year
panel.
and

is echoed

and extended
(2004) and Allen,
John, and Sundaram
by Acharya,
note
and
who
the
similarities
between
the U.S.
many
Carletti,
(2006),
Marquez
in
and U.K. economic
environments.
We
how
and,
general,
legal
recognize,
across
two
that
other
differences
the
selection
ever,
concerns,
countries,
sample
and other potential
effects
limit the extrapolation
of any infer
confounding
ences to U.S. firms. Nonetheless,
on the cap
of information
given the paucity
can
ital structures
of private firms, we believe
the following
analysis
provide
new insight
into the behavior
of capital
structures
for private firms, as well
as provide

an out-of-sample

robustness

check

for our results

based

on U.S.

data.

in Figure 2 using data on privately


the analysis
held firms
Figure 5 replicates
in the FAME database
from the United Kingdom
contained
for the period 1993
to 2002.21 Panel A presents
at
results
for all firms, while
Panel B restricts
tention to those firms with nonmissing
data for the duration
of the panel. The
a pattern
results
illustrate
for U.K. firms, which
is virtually
identical
to that of
21
Specifically,
least

?700,000
is unobserved

variables

we

focus

or classified

for private
in the portfolio
of tangible
assets

a market
firms with
of at
(i.e., private)
only on unquoted
capitalization
as medium
or large by the Companies
House.
Since market-to-book
in sales as a proxy. The remaining
firms, we use the percentage
change
are
to
formation
similar
those
used for the Compustat
regressions
sample:

to total assets,
the ratio
the log of total assets
deflated
by the United
to total assets,
and Fama and French
CPI, the ratio of operating
year dummies,
profits
see Michaely
dummies.
For further
the FAME
details
database
and Roberts
regarding

This content downloaded from 210.56.13.7 on Fri, 26 Feb 2016 14:08:45 UTC
All use subject to JSTOR Terms and Conditions

Kingdom's
38 industry
(2006).

Back

to the Beginning

1605

the public U.S. firms in Figure 2. As can be seen in Figure 5, there are relatively
between
the average
leverage of the unexpected
large differences
leverage port
date
ratios for the
folios in the formation
(event
0), the average
period
leverage
over
in
and
differences
converge
average
time,
significant
portfolios
leverage
across the portfolios
the entire period.
persist
throughout
5 suggests
of leverage
that the stability
ratios documented
above
Figure
States
also characterizes
the financing
for public firms
in the United
behav
in the United
ior of private
firms
the findings
indicate
Overall,
Kingdom.
are partially
structures
in firms' capital
that differences
determined
prior
to the time that they go public and that these differences
for many
persist
years.

VII.

Conclusion

a wealth
last several decades
have produced
of information
about cor
our
financial
results
that
the
of this
value
porate
suggest
policies. However,
more
cor
limited
than
We
information
be
find
that
may
previously
thought.
are
over
structures
stable
of
time:
Firms
that
have
porate capital
long periods
of
tend to remain as such for over 20 years. This feature
high (low) leverage
the leverage data-generating
is
firm
after
for
process
present
entry
controlling
and exit, as well as for previously
identified
of capital structure.
determinants
show that the majority
of variation
in capital
is
structure
Thus, our findings
time-invariant
of this variation
is unaccounted
and that much
for by existing
The

empirical
specifications.
our findings
While
dim picture of existing
paint a somewhat
empirical models
of capital structure,
the scope for future
The
they also narrow
investigations.
over time suggests
cross
of capital structure
that the factors driving
stability
as well.
in leverage
sectional
variation
ratios are stable over long horizons
we
because
also
observe
this
among private firms and
Additionally,
persistence
after initial public offerings,
to be largely unaffected
these factors appear
by
in capital market
the changes
of control, and information
access, distribution
environment
that occur at the time of the IPO.
Of course,
the other implication
of our findings
is that the identification
more
is
difficult
than
of the empiri
problem
previously
portrayed
by much
cal literature.
Static pooled OLS regressions
of leverage
ratios appear
inade
in
quate for dealing with the unobserved
cap
present
corporate
heterogeneity
ital structures.
The presence
of a significant
unobserved
component
transitory
are necessary
that dynamic
and Whited
suggests
specifications
(e.g., Hennessy
of a significant
unobserved
(2005, 2007)). The presence
permanent
component
identification
such as fixed effects estimates
requires more creative
strategies,
and Schoar
ex
and Goyal
(2003) and Frank
(e.g., Bertrand
(2006)), natural
and
Lemmon
and Roberts
(2006)
periments
(e.g., Leary
(2006)), instrumental
variables
and Petersen
estimation
(e.g., Faulkender
(2006)), and structural
(e.g.,
and Whited
in this
(2005, 2007)). We look forward to future research
Hennessy
vein.

This content downloaded from 210.56.13.7 on Fri, 26 Feb 2016 14:08:45 UTC
All use subject to JSTOR Terms and Conditions

1606

The Journal

of Finance

Appendix
This appendix
stat sample. All

details
numbers

the variable
construction
for analysis
of the Compu
in parentheses
refer to the annual Compustat
item

=
=
=

the

number.

Total Debt
Book

Leverage
Firm Size

Profitability

short-term
debt (34) + long-term
debt (9).
assets
total debt/book
(6).
are deflated
where
assets
log(book assets),
by
GDP deflator.
=
income before depreciation
(13)/book
operating

as

sets.

Cash Flow

Volatility
Tax Rate

Market
Equity
Market
Leverage
Market-to-Book

=
=
=

Collateral
Z-Score

=
=

Tangibility
Debt Issuance

=
=

Issuance

Marginal

of historical
the standard deviation
income,
operating
at
least
3
years of data.
requiring
tax rates obtained
from John
simulated
marginal
Graham.

Net
Net

Equity

stock price (199) * shares outstanding


(54).
debt + market
total debt/(total
equity).
stock liqui
-f- total debt + preferred
(market equity
- deferred
tax
taxes
value
and
investment
(10)
dating
credits (35))/book assets.
(3) + net PPE (8)/book assets.
Inventory
income (170) + sales (12) -f-1.4 * retained
3.3 * pre-tax
- current
lia
(36) + 1.2 * (current assets
(4)
earnings
bilities
(5))/book assets.
assets.
net PPE/book
the change in total debt from year t? 1 to year t divided
by the end of year t?1 total assets.
in shares
The
change
outstanding
split-adjusted
*
times the
(data25t
(data27t_i/data27t))
data25t_i
stock
+
(datal99t
average
price
split-adjusted
*
end
dividend
the
(data27t/data27t-i))
datal99t_i
by
of year t?1 total assets.

REFERENCES
variations
Sundaram,
2004, Cross-country
John, and Suresh
Business
School.
of bankruptcy
paper, London
codes, Working
and Robert Marquez,
Elena
2006, Stakeholder
capitalism,
Carletti,
Franklin,
Allen,
of Pennsylvania.
ernance
and firm value, Working
paper, University

Viral,
Acharya,
tures: The

Arellano,

Kose

struc

in capital

role

2003,

Manuel,

Panel

Data

Econometrics,

Oxford

University

Press

(Oxford,

corporate
United

gov
King

dom).
Baker,

Malcolm,
of Finance

and Jeffrey
1-30.

Marianne,
Bertrand,
on firm policies,
Blundell,
panel

Wurgler,

2002,

The market

timing

theory

of capital

structure,

Journal

57,

and Antoinette
Quarterly
and Stephen

Richard,
data models,

Journal

Journal

Schoar,
2003,
of Economics

with
Managing
118, 1169-1208.

conditions
1998, Initial
Bond,
87, 115-143.
of Econometrics

style:

and moment

The

effect

restrictions

This content downloaded from 210.56.13.7 on Fri, 26 Feb 2016 14:08:45 UTC
All use subject to JSTOR Terms and Conditions

of managers
in dynamic

to the Beginning

Back

1607

M. R. A. Engel,
out: Reply
to
strikes
and you're
2004, Three
J., and Eduardo
and Willis,
paper, Yale University.
Working
is the
behavior
and financing:
How
conclusive
2006, Targeting
Chang, Xin, and Sudipto
Dasgupta,
of Science
evidence?
and Technology.
paper, Hong Kong University
Working
Ricardo

Caballero,

Cooper

DeMarzo,
vately
Dittmar,
Dittmar,
54.
Fama,

and Michael
financial
2001, Optimal
Peter,
Fishman,
long-term
observed
cash flows, Working
paper, Stanford
University.
do firms
stock? Journal
2000, Why
of Business
Amy,
repurchase
and Anjan

Amy,

2005,

Thakor,

do firms

Why

R. French,
2002,
F., and Kenneth
Eugene
about dividends
and debt, Review
of Financial
R. French,
F, and Kenneth
Economics
76, 549-582.

Fama, Eugene
Financial

2005,

Testing
Studies
Financing

and Mitchell
Petersen,
2006,
Faulkender,
Michael,
ture? Review
Studies
19, 45-79.
of Financial

Does

and Kasturi
2006, Partial
Mark,
Flannery,
Rangan,
Economics
Journal
79, 469-506.
of Financial
K. Goyal,
2003, Testing
Frank, Murray
Z., and Vidhan

issue

73, 331-355.
Journal
of Finance

equity?

trade-off

and

source

adjustment

issues

of capital

toward

the pecking

order

pecking

15, 1-33.
decisions:
Who
the

with

contracting

order

predictions

of capital

of

struc

capital

capital

theory

62,

stock? Journal

affect

target

pri

structures,
structure,

Economics
67, 217-248.
of Financial
K. Goyal,
Z., and Vidhan
2004,
Murray

Journal

are
structure
decisions:
Which
factors
Capital
of British
Columbia.
paper, University
Working
important?
K. Goyal,
and pecking
order theories
of debt, Working
2005, Tradeoff
Frank, Murray
Z., and Vidhan
Columbia.
of British
paper, University
Frank,

reliably

K. Goyal,
How much
do
2006,
Z., and Vidhan
Murray
leverage
adjustment:
Corporate
of British
Columbia.
managers
paper, University
really matter?
Working
for the corporate marginal
tax rate, Journal
John R., 1996, Proxies
Economics
Graham,
of Financial
42, 187-221.
Frank,

John R., and Campbell


Graham,
dence from the field, Journal

of corporate
finance:
Evi
2001, The theory and practice
Harvey,
Economics
60, 187-243.
of Financial
and James
and the
R., Michael
Lemmon,
Schallheim,
1998, Debt,
taxes,
leases,
tax status, Journal
of corporate
53, 131-161.
of Finance
2nd Ed. (Macmillan
New
H., 1993, Econometric
Analysis,
Publishing
Company,

John

Graham,

endogeneity
William
Greene,
York).
Grilliches,
NBER
Helwege,
firms,

and

Zvi,

Jacques

paper
Working
and Nellie
Jean,

Mairesse,
No. 5067.

1995,

Production

functions:

Is there a pecking
order?
1996,
Liang,
Economics
40, 429-458.
of Financial
A., and Toni M. Whited,
Christopher
2005, Debt dynamics,

The

search

Evidence

for

from

identification,
a panel

of IPO

Journal

Journal

of Finance

60,1129

is external
A., and Toni M. Whited,
2007, How
costly
Journal
estimation,
62, 1705-1745.
of Finance
of production
function
1962, Estimation
Hoch,
Irving,
parameters
combining
section
30, 34-53.
data, Econometrica

financing?

Evidence

Hennessy,
1165.
Hennessy,
from

Christopher
a structural

Hovakimian,
Journal
Hovakimian,
nancial
Hsiao,

time-series

and

Armen
structures
determined
2006, Are observed
capital
by equity market
and Quantitative
of Financial
41, 221-243.
Analysis
Tim Opler,
and Sheridan
Armen,
Titman,
2001, Debt-equity
choice, Journal
and Quantitative
36, 1-24.
Analysis
2003,

Cheng,

of Panel

Analysis

Data

(Cambridge

University

Press,

theories

of capital

structure

Cambridge,

cross

timing?
of Fi
United

Kingdom).
Huang,

and

Rongbing,

Jay

Ritter,

speed of adjustment,
Working
Steven N., Berk A. Sensoy,
Kaplan,
to public
companies,
Working
Kayhan,

Ayla,
of Financial

and

Sheridan

Economics

2007,

Testing

paper, University
and Per Stromberg,
paper,

Titman,
1-32.

University
2007, Firms

and

estimating

the

of Florida.

are firms? Evolution


2005, What
of Chicago.
and their capital
histories
structures,

83,

This content downloaded from 210.56.13.7 on Fri, 26 Feb 2016 14:08:45 UTC
All use subject to JSTOR Terms and Conditions

from birth
Journal

1608

The Journal

Stock Growth:
Edwin,
1963, Capital
Amsterdam).
ing Company,
loan supply,
T, 2006, Bank
Leary, Mark
paper, Cornell University.

Kuh,

Mark T, and Michael


60, 2575-2619.
of Finance
T, and Michael
Leary, Mark

R. Roberts,

Leary,

A Micro-Econometric
lender

2005,

choice

Do firms

and

corporate

rebalance

their

R. Roberts,

2006, The pecking


order,
of Pennsylvania.
University
R. Roberts,
The
2006,
response
of credit: A natural
supply
experiment,

icy to changes

in the

Publish

(North-Holland

Approach

paper,
Working
and Michael
Michael,

asymmetry,
Lemmon,

of Finance

capital

structure,

capital

structures?

debt

and

capacity,

of corporate

Journal
information

investment

paper,

Working

Working

pol
of

University

Pennsylvania.
and Jaime
2007, Debt
Zender,
of Utah.
University
root
and Chien-Fu
Lin, 1993, Unit
Levin, Andrew,
93-56.
UCSD

Lemmon,

Michael,

capacity

and

tests

of capital

structure,

Working

paper,

tests

in panel

data: New

results,

paper,

Working

ratios? Evidence
from historical
market-to-book
2005, Do firms have
Laura,
target
leverage
of Science
and Technology.
and past returns, Working
paper, Hong Kong University
affect firm financial
structure?
Re
and Gordon
2005, How does industry
Peter,
Phillips,
Mackay,
Liu,

view
Michaely,
firms,

Studies
of Financial
and Michael
Roni,
Working

paper,

18, 1433-1466.
R. Roberts,
2006, Corporate
of Pennsylvania.
University

dividend

policies:

Lessons

from

function
free of management
1962, Empirical
bias, Journal
production
43, 44-56.
structure
Journal
Stewart
39, 575-592.
1984, The capital
C,
of Finance
puzzle,
in the telecommunications
of productivity
and Ariel Pakes,
1996, The dynamics
Steve,

of Farm

Yair,
Mundlak,
Economics
Myers,

Olley,
ment

industry,
Mitchell

Econometrica

64,

equip

1263-1298.

errors
in finance
data
standard
A., 2007, Estimating
panel
Studies.
Review
of Financial
forthcoming,
do we know
about
1995, What
G., and Luigi Zingales,
capital
Rajan,
Raghuram
from international
evidence
50, 1421-1460.
data, Journal
of Finance
and Sons, New York).
(John Wiley
Henri,
1959, The Analysis
Scheffe,
of Variance
Petersen,

private

sets:

Comparing

approaches,

structure:

static tradeoff
C. Myers,
and Stewart
Lakshmi,
1999, Testing
pecking
against
Shyam-Sunder,
Economics
of capital
Journal
models
51, 219-244.
structure,
of Financial
of zero-leverage
firms, Working
2006, The mystery
Strebulaev,
Ilya, and Bauzhong
Yang,
Stanford
University.
and Roberto
Sheridan,
Titman,
Finance
43, 1-19.

Wessels,

1988,

The

determinants

of capital

structure,

This content downloaded from 210.56.13.7 on Fri, 26 Feb 2016 14:08:45 UTC
All use subject to JSTOR Terms and Conditions

Some

order

paper,

Journal

of

You might also like