How Stable Are Corporate Capital Structures?
by
Harry DeAngelo*
Richard Roll*
March 2011
Revised February 2012
Abstract
Capital structure stability is the exception, not the rule. Substantial instability of book leverage, market
leverage, and the net-debt ratio is the norm. Firm fixed effects differ significantly across decades,
rejecting stability of firm-mean leverage and of the cross section. Sequential comparisons of leverage
cross sections show that stability is short-lived, with similarities between cross sections evaporating as the
time between them lengthens. Cross-sectional migration is pervasive: 69.5% of firms listed 20-plus years
have book leverage in at least three quartiles, and 30.4% are in all four quartiles at different times over the
average 20-year period. Stability occurs infrequently and mainly at low leverage, and is virtually always
temporary, with firms abandoning low leverage en masse during the 1950s and 1960s.
Coordinates DeAngelo
Phone
1-213-740-6541
USC Marshall School
Address
ACC 308D
Los Angeles CA 90089
E-Mail
hdeangelo@marshall.usc.edu
Roll
1-310-825-6118
UCLA Anderson C408
110 Westwood Plaza
Los Angeles CA 90095
rroll@anderson.ucla.edu
*This research was supported by the Kenneth King Stonier Chair at the USC Marshall School of Business
and by the Joel Fried Chair at the UCLA Anderson School of Management. For helpful comments, we
thank Tony Bernardo, Fabio Braggon, Daniel Carvalho, Steve Cauley, Tom Chang, Tom Copeland, Linda
DeAngelo, Andrea Eisfeldt, Eugene Fama, Wayne Ferson, Murray Frank, Stuart Gabriel, Mark Grinblatt,
Gareth James, Lyndon Moore, Kevin J. Murphy, Oguzhan Ozbas, Chris Parsons, Jay Ritter, Lori
Santikian, Eduardo Schwartz, Berk Sensoy, Douglas Skinner, René Stulz, Avanidhar Subrahmanyam, Ivo
Welch, Mark Westerfield, and Toni Whited. We thank Ed Tinoco for help in accessing data from the preCRSP/Compustat era, and Amy Allen, Xiaolin Gong, Richard Graham, Mauri Gustafson, Michael
Neagoe, Jonathan Pack, and Matthew Wong for superb work on that data. We also thank Chao Zhuang
for outstanding research assistance.
How Stable Are Corporate Capital Structures?
1.
Introduction
Capital structure stability is a critical feature of corporate financial policy. If leverage ratios
exhibit only modest inter-temporal variation, theorists should largely focus on time-invariant determinants
of capital structure. However, if leverage instability is pervasive, time-varying determinants – such as
investment or market-timing opportunities – are plausibly major, and arguably essential, components of
any empirically credible theory of financial policy.
The literature contains conflicting views on leverage stability. Welch (2004) reports that stock
return-induced changes in market leverage ratios are often large and that firms do little to offset them. In
a similar vein, Baker and Wurgler (2002) find that market value fluctuations have “very long-run impacts
on capital structure.” Fama and French (2002) and Hovakimian and Li (2011) find glacial rates of mean
reversion in leverage, while Chang and Dasgupta (2009) show that mean reversion can reflect random
financing rather than target rebalancing. These studies collectively suggest that keeping leverage near a
target ratio is not of much import to managers and, more generally, that leverage is not particularly stable.
Yet, the view that leverage is quite stable now dominates the literature. In their literature review,
Frank and Goyal (2008, p.156) state that “a satisfactory theory must account for why firms keep leverage
stationary.” This mandate echoes Lemmon, Roberts, and Zender (2008), who find significant firm fixed
effects in panel leverage regressions and conclude that (not-yet-identified) time-invariant factors generate
remarkably stable capital structures for two decades and longer, and that time-varying factors are unlikely
sources of capital structure heterogeneity. The literature reviews of Parsons and Titman (2008, p.25) and
Graham and Leary (2011, section 5.2) also highlight significant firm fixed effects and the implied need to
identify time-invariant determinants of leverage. These findings lead Rauh and Sufi (2011) to state that
“the extant research strongly argues that cross-sectional variation in capital structure is where researchers
should focus.” Leverage stability has thus become a major “stylized fact” that concentrates researchers’
attention on identifying time-invariant determinants of leverage; see also Graham, Harvey, and Puri
(2009), Hennessy, Livdan, and Miranda (2010), and Malmendier, Tate, and Yan (2011).
This paper provides a comprehensive picture of the nature and extent of leverage stability. The
data show that leverage stability is the exception, not the rule, and that substantial instability is the norm
at both the firm and cross-sectional levels. We document a wide variety of regularities that narrow the
scope of empirically credible theories of capital structure. For example, leverage cross sections more than
a few years apart differ markedly, with differences growing each year – and not reverting or stabilizing –
until there is almost no similarity in cross-sectional snapshots taken at different times. With stability of
the cross section evaporating after some brief stickiness, time-varying determinants are essential to
explain the cross-firm distribution of leverage at any given time. The evolution of the cross section is
thus at odds with theories in which an important managerial objective is to keep leverage close to a stable
(or reasonably stable) target ratio. Our evidence is consistent with firms having target leverage zones that
set “soft” or flexible limits on acceptable leverage ratios, as suggested by Fama and French (2005, p. 580)
and the CFO survey responses in Graham and Harvey (2001).
To preview our findings, Figure 1 plots leverage ratios of GM, IBM, and Kodak from 1926 to
2008. Within-firm leverage variation is substantial for all three firms, with market leverage varying more
widely than book. Mean leverage ratios for each firm (taken individually) differ markedly across subperiods. IBM has had long periods of leveraging and deleveraging, and large variation also characterizes
GM’s leverage, although both had relatively stable leverage in the 1960s and 1970s. Kodak had stable
(near-zero) leverage for many years, but its leverage sky-rocketed in the 1980s, followed by deleveraging
and re-leveraging, with leverage remaining well above that of earlier years. The leverage plots in the
appendix for these and 21 other Dow Jones Industrial Average (DJIA) firms illustrate our large-sample
findings that (1) substantial within-firm variation is the norm, but episodes of stability arise occasionally,
(2) changes over time in a given firm’s mean leverage are commonplace, and (3) the relative positions of
firms in the leverage cross section are sticky in the short run, but far from stable over horizons of more
than a few years, with similarities evaporating as the time between cross sections lengthens.
We gauge the extent of leverage stability in three distinct senses, which correspond to intuition
about within-firm variation, stationarity in firm-mean leverage, and stability of cross-firm differences:
2
1.
Stability: a firm’s leverage ratio remains consistently within a narrow band, with broader
bands corresponding to greater degrees of instability.
2.
Firm-mean stability: the expected value of a firm’s leverage remains constant over time.
3.
Cross-sectional stability: the relative (high versus low) leverage positions of firms in the
current cross section closely resemble their positions in future cross sections.
Stability in the basic sense (1) captures the intuitive notion that a given firm’s leverage does not change
much over time, as indicated by time-series ranges and standard deviation-based bands. Stability in the
firm-mean sense (2) stipulates that deviations from a given firm’s mean leverage are temporary, and not
changes in expected leverage. Stability in the cross-sectional sense (3) stipulates that individual firms’
relative (high or low) positions in the leverage cross section remain similar over time. We analyze
leverage variation over intervals ranging from one year to multiple decades to assess whether, or to what
extent, leverage tends to be stable or unstable over short, medium-term, and long horizons.
We find substantial leverage instability in all three senses. Large within-firm variation (gauged
by time-series ranges and standard deviations) pervasively characterizes book leverage, market leverage,
and net-debt ratios. Firm fixed effects differ significantly across decades in panel leverage regression
tests, indicating both firm-mean and cross-sectional instability, and firm-specific sources of time-series
variation have substantial explanatory power. Stability of the leverage cross section is short-lived, with
the ability of a firm’s current leverage to predict its leverage in future cross sections declining sharply
over five to 10 year horizons, and thereafter continuing to erode to near-zero levels. Migration over the
leverage cross section is pervasive: 69.5% of firms listed for 20-plus years have book leverage ratios that
appear in at least three different sample quartiles, and 30.4% have leverage in all four quartiles at
different times over the average 20-year period.
In terms of altering researchers’ priors, perhaps our most important findings are those challenging
the literature’s view that cross-sectional distributions of leverage exhibit a high degree of stability over
long horizons. Stability in this sense implies that a firm’s current high or low leverage (relative to other
firms) reliably predicts a comparable relative position in leverage cross sections extending well into the
future. Long-run stability is widely accepted as descriptive of the cross section based on high R2s
3
associated with firm fixed effects in panel leverage regressions.
But these panel-R2s gauge the
explanatory power of average firm differences over all years’ observations, whereas stability concerns the
explanatory power of current leverage for leverage in each marginal year moving forward in time. Shortrun stickiness in the data implies that the average effect exceeds the marginal, which means that these
panel-R2s can – and empirically do – offer a substantially upward-biased estimate of the extent to which
different cross sections remain similar to one another over extended periods of time.
To assess the degree of stability of the cross section, one accordingly needs to gauge the
explanatory power of a given cross section for the sequence of cross-sectional “slices” going forward one
year at a time, and extending well into the future. When we do so, we find that similarity of leverage
cross sections erodes over time to near-zero levels.
Our tests indicate that vestiges of stability in the cross section remain at horizons of 15 or 20
years, and this fact reflects our finding that leverage stability does occur from time-to-time at individual
firms. However, stability arises only infrequently and then largely when firms have low leverage, and it
is virtually always temporary. Departures from stability are infrequently followed by rebalancing to the
prior stable leverage regime or by establishment of a new stable regime. Departures from stability and the
attainment of leverage peaks and troughs are associated with company expansion and external funding
thereof. Leverage peaks and troughs are not exogenous shocks that induce rebalancing to more moderate
leverage, as managers choose to (i) issue debt to fund expansion when attaining a peak, then immediately
begin repaying debt, and (ii) pay down debt when attaining a trough, then issue debt to fund expansion.
We also find wholesale abandonment of conservative leverage during the 1950s and 1960s by
Compustat-listed firms. Case studies of our 24 DJIA firms indicate that this trend toward higher leverage
is associated with the funding of expansion during the booming post-World War II economy. Taggart
(1985) and Graham, Leary, and Roberts (2011) also report increases in average leverage ratios after the
war, consistent with our inference that leverage is not stable over long horizons.
Lemmon, Roberts, and Zender (LRZ, 2008) conclude that time-varying determinants of leverage
are unlikely to be important sources of capital structure heterogeneity. They base this claim on their
4
inference that leverage is stable over long horizons, with firms that have relatively high (or low) leverage
tending to remain as such for over two decades. By stable leverage, LRZ mean that within-firm variation
is largely transitory noise around permanent (time-invariant), firm-specific target ratios. In our lexicon,
LRZ say that leverage is stable in the firm-mean and cross-sectional senses. As supporting evidence,
LRZ point to (i) differences that remain over long horizons in cross-sectional average leverage ratios of
large (quartile-sorted) sub-samples, and (ii) significant firm fixed effects in panel leverage (ANOVA)
models, with R2s far greater than the near-zero explanatory power they find for year fixed effects.
The LRZ results, which we confirm, are not sufficient to establish their claims about stability and
the unimportance of time-varying determinants, and a closer look at the data supports the opposite
conclusions. Regarding (i), long-lasting differences in cross-sectional averages of large sub-samples of
firms do not establish leverage stability at the firm or cross-sectional levels. Large-sample averaging can
– and empirically does – mask substantial (1) within-firm variation, (2) firm-specific changes in mean
leverage, and (3) changes in the relative leverage positions of firms within the cross section.
Regarding (ii), LRZ’s low R2s for year dummies – in absolute terms or relative to firm dummies –
do not establish that time-varying factors are systematically unimportant sources of capital structure
heterogeneity. The reason is that their ANOVA models have only “additive” firm and year fixed effects,
hence they exclude firm-specific sources of time-series variation. Moreover, significant firm fixed effects
do not establish leverage stability in the cross-sectional or firm-mean senses (as discussed above). Nor do
they even imply that firms have leverage targets of any type, let alone stationary targets. They only
indicate that average leverage differs across firms during the sample period. Such average differences can
arise in finite samples when firms have no targets, e.g., if they follow the pecking order theory.
Our tests, which include LRZ’s additive structure as a nested alternative, strongly reject stability
in the firm-mean and cross-sectional senses, as they reveal that firm dummies exhibit highly significant
differences across decades. Firm/decade interaction effects, which capture firm-specific time-series
variation and which are excluded from purely additive ANOVAs, substantially increase R2s and account
for a nontrivial portion of the explanatory power attributed to firm fixed effects by additive models.
5
The high R2s for firm fixed effects in purely additive ANOVAs not only reflect the suppression of
firm/time interactions, but also the fact that the Compustat population is dominated by firms with just a
few years of data. Over half the firms in our full sample have nine or fewer years of data, and more than
two-thirds have 14 years or fewer. Given the short-run stickiness in leverage, firm dummies capture a
large portion of the leverage variation for such firms, thus inflating the R2 averaged over the sample as a
whole and giving a misleading picture of the explanatory power of firm fixed effects for leverage over
long horizons. It seems of questionable validity to draw strong conclusions about long-horizon leverage
behavior – and especially about the “staying power” of firm dummies – from R2s for samples that have so
many firms with so few years of data. We accordingly examine firms with 50 or 75 years of leverage
data, and find much lower explanatory power for firm fixed effects than in our full sample, as well as
greater explanatory power for both firm/time interactions and time main effects.
All these findings support the view that time-varying determinants of leverage are systematically
important drivers of capital structure heterogeneity. Finally, and consistent with the latter view, we find
that industry-median leverage ratios exhibit substantial instability at the four-digit, three-digit, and twodigit SIC levels and for Fama and French’s 49-industry and 12-industry groupings. Frank and Goyal
(2009) report that the strongest determinant (among many) of a given firm’s current leverage is the
contemporaneous median leverage ratio across firms in its four-digit industry. Taken together, our
findings and those of Frank and Goyal imply that a materially time-varying factor has the strongest
known influence on leverage. Because industry-median leverage is a proxy for fundamental determinants
of leverage common to firms in a given industry, this evidence underscores the need to understand timevarying sources of capital structure heterogeneity.
We discuss other implications of our evidence in section 9. Sections 2, 3, and 4 analyze variation
in leverage at individual firms and in the cross section. Sections 5 and 6 study stable leverage regimes
and leverage instability at the firm level. Section 7 analyzes leverage variation in the DJIA sub-sample,
while section 8 gauges the stability of industry-median leverage ratios.
6
2.
Basic facts: Time-series variation in leverage
We analyze 15,096 industrial firms in the CRSP/Compustat file over 1950-2008. Industrial firms
are those with SIC codes outside the ranges 4900-4949 (utilities) or 6000-6999 (financials). The sample
excludes firms incorporated outside the U.S. and those not assigned a CRSP security code of 10 or 11. A
firm enters the sample the first year it has non-missing values for total assets and share price, and stays as
long as Compustat continues to report non-missing values of total assets and its shares remain listed.
To gauge leverage behavior over long horizons, we often focus on the subset of 2,751 firms with
20 or more years on Compustat, and on a “constant composition” sample of 157 firms listed from 1950 to
at least 2000. The former group accounts for 92.9% of total market capitalization and 91.7% of book
assets in the median sample year over 1950-2008, and the latter accounts for 44.1% and 41.4%. We also
analyze hand-collected leverage data back to before the Great Depression for 24 Dow Jones Industrial
Average (DJIA) firms in the constant composition sample; the appendix describes this sub-sample.
2.1
Range and standard deviations of book leverage, market leverage, and the net-debt ratio
Table 1 reports the time-series range of book leverage (panel A), market leverage (panel B), and
the net-debt ratio (panel C). Book leverage is the ratio of total debt to total assets (Debt/TA) where debt
does not include non-financial liabilities. Market leverage is the ratio of total debt to the sum of total debt
plus the market value of common stock. Because many tax/bankruptcy cost models make predictions
about leverage measured net of cash, we report the net-debt ratio, debt minus cash divided by total assets.
Strikingly, 29.2% of firms listed for 20-plus years have book leverage (Debt/TA) ratios that vary
by more than 0.500 between their highest and lowest years, while 55.4% and 65.9% have similar variation
in market leverage and net-debt ratios. The median range of Debt/TA is 0.391, while the median ranges
in market leverage and net-debt ratios are 0.536 and 0.599. Only 2.3% of these firms have Debt/TA ratios
that remain within a band of 0.100, and just 8.4% (2.3% + 6.1%) have Debt/TA ratios that remain within
a relatively wide band of 0.200. Substantial time-series variation in leverage also characterizes firms
listed less than 20 years. The median Debt/TA range is 0.357 for firms listed 15 to 19 years, while it is
0.314, 0.241, and 0.110 respectively for firms listed 10 to 14 years, 5 to 9 years, and 2 to 4 years. The
7
median range is greater for market than for book leverage for all year categories, but only slightly so for
firms listed 2 to 4 years. In sum, substantial leverage instability is pervasive over intermediate to long
horizons, and nontrivial instability is common over short horizons.
As a rough guide to put these findings in perspective, we calculate firm-specific ranges in fitted
values from a panel regression with book leverage as the dependent variable and with lagged values of the
Rajan and Zingales (1995) determinants on the right-hand side. These fitted values can be viewed as
proxies for time-varying “targets.” Among firms listed 20-plus years, the median range in fitted values is
0.101, or about one-fourth the median range of 0.391 in actual book leverage. For the other rows in panel
A, the median fitted value ranges are about one-third the actual ranges. These comparisons suggest that
changing targets can account for at best a modest portion of the time-series variation in Table 1.
Table 2 presents histograms of the time-series standard deviation (
of book leverage, market
leverage, and the net-debt ratio. Because firms listed just a few years are a large fraction of the firms on
Compustat, we conservatively report maximum likelihood estimates of , i.e., with divisor equal to the
number of annual observations for the firm, N, not N-1. Substantial s are the norm for all three leverage
measures, with the net-debt ratio generally having higher s than market leverage, and market leverage
having higher
than book. For example, for firms listed 20-plus years, average s for book leverage,
market leverage, and the net-debt-ratio are 0.115, 0.146, and 0.167. These figures translate to plus/minus
two-sigma bands of width 0.460, 0.584, and 0.668, thus implying that wide variation in leverage is the
norm, consistent with the ranges reported in Table 1. The range and
are highly correlated for all sample
partitions and leverage measures (column (4) of Table 2), which makes sense because the range reflects
the cumulative result of the year-by-year leverage variation that the s capture.
Table 2 also shows that, among firms listed 2 to 4 and 5 to 9 years, 50.6% and 29.9% respectively
have book leverage s below 0.050, while only 7.5% of firms listed 20-plus years and 2.6% of firms in
the constant composition sample have comparably low s (column (5)). Market leverage and the net-debt
ratio show a similar differential incidence. The higher incidence of relatively low s among firms listed
8
just a few years reflects a tendency toward short-run stickiness in leverage, a property that is a prominent
feature of the evolution of the cross-sectional distribution of leverage, as documented in section 4.
While market leverage shows greater variation than book in both Tables 1 and 2, the difference is
perhaps not as great as expected. The reason is clear from Table 3, which reports a surprisingly high
correlation between the two leverage measures. The correlation between book and market leverage is
0.878 for the median firm in the full sample, and similarly high correlations pervade the sample.
These correlations reflect Welch’s (2004) estimate that 60-70% of market leverage variation is
due to net issuance activity, i.e., stock-return volatility notwithstanding, managers choose the bulk of
observed variation in market leverage. Since managerial choices likely contribute more to book leverage
variation, Welch’s estimate implies material co-movement in book and market leverage. Also, since book
and market leverage both vary widely and the latter more so, these data support Welch’s inference that
managers do not systematically offset stock return-induced changes in market leverage. In what follows,
we focus on book leverage in part because Table 3’s high correlations suggest there is not all that much
incremental information in the market series and also because, as intuition suggests and Tables 1 and 2
confirm, book variation probably provides a lower bound on the instability in market leverage.
2.2
Evolution of leverage over 20-year horizons
Table 4 reveals that nontrivial variation in leverage is typical over short to intermediate time
spans for firms listed 20-plus years. [We find substantively identical results for firms listed 5, 10, 15, 25,
or 30-plus years (details not tabulated).] The Debt/TA ratio is tracked from its first year over the next 19
years and the table records the percentage of firms whose leverage has moved beyond +/-0.050 (and +/0.100, etc.) from the initial Debt/TA.
Table 4 shows that, by the time five years have passed, 81.5% of firms have had leverage ratios at
some time outside a band delineated by +/-0.050 of their initial leverage. Over the same relatively short
horizon, 61.1% of firms have moved outside a band of +/- 0.100 around the initial Debt/TA, and 29.7%
have moved outside a +/-0.200 band. By the time 10 years have elapsed, 79.7% have had leverage ratios
outside the former band, while 47.9% have had leverage outside the latter band.
9
Over the full 19-year horizon, few firms maintain leverage in a relatively compact band and many
firms have widely varying leverage ratios. By the final year in Table 4, the Debt/TA ratios of 96.8% of
firms have moved outside the +/-0.050 band surrounding initial leverage at some point. In other words,
over a 19-year period, only 3.2% of sample firms consistently keep Debt/TA within the band of width
0.100 that is delineated by +/-0.050 of initial leverage. Over the same horizon, 66.5% of firms have at
some point moved outside the +/-0.200 band, while 39.5% have moved outside the +/-0.300 band and
22.2% have had Debt/TA ratios outside the band defined by +/-0.400 of their initial ratios.
The evidence in Table 1, 2, and 4 of pervasive, substantial time-series variation in leverage
contrasts sharply with the virtually constant leverage ratios attained after 10-15 years in Figures 1 and 2
of Lemmon, Roberts, and Zender (LRZ, 2008). The two sets of findings are, in fact, both correct and
fully consistent. Our tables report inter-temporal variation in leverage for individual firms. LRZ track the
cross-sectional average leverage ratios over 20 years for groups sorted by current leverage (Figure 1) and
unexpected leverage (Figure 2). With hundreds of firms in each LRZ group, one would expect little
fluctuation in the group average, even with huge time-series variation in the leverage ratios of its
constituent individual firms, unless leverage changes are materially correlated across firms.
LRZ’s stability inference is based on differences that remain after 20 years in the cross-sectional
average leverage ratios of four (quartile-sorted) groups. We confirm this finding, but disagree that it
establishes firm-level leverage stability. Differences in cross-sectional averages are completely consistent
with gross instability in the leverage of individual firms. Nor do such differences establish stability in the
mean leverage ratios of individual firms (see immediately below and section 3). They do offer a hint of
stability in a cross-sectional sense. Section 4 documents that stability of the leverage cross section erodes
significantly over time, with only vestiges of stability remaining after 20 years.
2.3
A first look at whether firm-mean leverage is stable over time
Leverage stability in the firm-mean sense is rejected by the data in Table 5. For each firm in the
constant composition sample, we divide the time series of Debt/TA ratios in half, and record the p-value
from a t-test that the mean leverage is the same in the earlier and later periods. If mean leverage is stable
10
for all firms, roughly 5% of them should have p-values less than 0.05, and thus reject the hypothesis of
equality over time. A higher frequency of rejection implies inequality. Since the t-tests for these 157
firms are not independent, we also conduct a Hotelling T2 test of the hypothesis of equality of earlier and
later (firm-specific) means for all firms viewed collectively.
We find much higher rejection rates than the 5% expected if mean leverage were equal in the
earlier and later sub-periods. The null is rejected against a two-sided alternative for 72.6% of firms when
we assume a constant variance over the full period, and for 72.0% of firms when variances are allowed to
differ across sub-periods (row 1 of Table 5). The Hotelling T2 test also indicates rejection of leverage
stability in the firm-mean sense, as do the panel regression tests in section 3. The null of no difference in
mean leverage is rejected at 62.4% of firms against the alternative that mean leverage increases (row 2),
versus 14.0% against the alternative that it decreases (row 3). In other words, over 1950-2008, leverage
increased more frequently than it decreased among firms in the constant composition sample, consistent
with the wholesale abandonment of conservative leverage documented in section 3.2.
3.
Leverage stability: Panel regression analysis
Lemmon, Roberts, and Zender (2008, p. 1589) conclude that time-varying determinants of
leverage are unlikely sources of capital structure heterogeneity. They base this conclusion on panel
leverage (ANOVA) models in which firm fixed effects (firm-specific dummy variables) generate R2s
above 0.500, while year fixed effects (year dummies) generate near-zero R2s. These findings – especially
the significant firm fixed effects – are why LRZ’s conclusion that leverage tends to be stable over long
horizons has become a widely accepted “stylized fact.”
We confirm LRZ’s findings, but disagree that these conclusions follow from the empirical results,
and we report evidence to the contrary about both. In terms of the underlying economics, year dummies
consider only a narrow type of time variation – all firms have identical simultaneous shifts in expected
leverage. Year dummies miss firm-specific sources of time variation in leverage as can arise, e.g., with
the evolution of investment opportunities. Because firm-specific variation tends to wash out in large
11
sample averages, it is not generally detectable in year dummy variation. Thus, low R 2s for year dummy
regressions – in absolute terms or relative to firm dummies – do not establish that leverage is generally
stable, or that time-varying determinants are unlikely sources of capital structure heterogeneity.
Moreover, significant firm fixed effects can, and in the data do, co-exist with substantial leverage
instability at the firm and cross-sectional levels. A finding of significant firm dummies indicates that
average leverage differs across firms over the sample period. It does not rule out instability in the
leverage of a given firm in the basic sense shown in Tables 1, 2, and 4. Nor does it establish stability in
the sense of constant mean leverage for a given firm. Individual firms’ mean leverage ratios can change
radically during the sample period, but as long as they differ materially from those of other firms over the
full period, firm dummies will be statistically significant in panel estimations. More generally, significant
firm fixed effects can arise given a short-run “stickiness” in leverage that fades strongly over time, so that
any tendency toward stability of the cross section is in fact quite weak. Such is the case for the data.
3.1
Additive models miss firm-specific sources of time-series variation
The fundamental inference problem is that a regression model that includes only additive firm
and year fixed effects has no ability to capture firm-specific sources of time-series variation. One can,
however, test for stability in a more general model structure that is able to capture possible firm-specific
variation. For example, Kim, Morse, and Zingales (2009) test whether firm (university in their study)
fixed effects vary over decades, and conclude that the research productivity advantages of being affiliated
with a top school declined from the 1970s to the 1990s.
In our Table 6, panel leverage regression tests indicate that firm fixed effects differ significantly
across decades. These stability tests compare regression (1) in which firm dummies can differ for each
decade with regression (2) in which each firm has a time-invariant dummy. We also report F-tests for
comparing regressions (4) and (5), which add year dummies to (1) and (2). Panel A of the table reports
results for the 24 DJIA firms, while panels B and C analyze the constant composition sample, firms listed
20-plus years, and the full Compustat sample. F-tests reject the equivalence of models (1) and (2) (and
12
models (4) and (5)) at high significance levels, with p-values less than 0.0001 in all cases.1
In panel A, the (adjusted) R2 of the firm/decade regression (1) for the DJIA sample is 0.841, or
0.570 above the 0.271 R2 of the firm fixed effects regression (2). The F-statistic of 38.68 strongly rejects
the hypothesis that firm fixed effects are constant across decades. Viewed individually, the firm dummy
regression (2) and year dummy regression (3) have R2s of 0.271 and 0.218. With firm and year dummies
together in (4), the R2 is 0.503 or 0.338 below the R2 of model (1). The F-statistic of 25.50 in the far-right
column (for comparing (4) versus (5)) also indicates that firm dummies differ across decades. Thus
taking firm-specific time-series variation into account substantially improves explanatory power.
In panel A, the R2 of the firm fixed effect model (2) shrinks markedly as the sample period
lengthens. The R2 is 0.271 for the full 75 years, or about half the 0.543 R2 when the model is estimated
for the last 20 years. As with the F-tests, the fact that the R2 fades as the analysis period lengthens raises
doubts about the literature’s consensus that the existence of significant firm fixed effects indicates that
important permanent leverage determinants are missing from existing models.
In panels B and C, the constant composition results arguably are the most informative because
year dummies can be distorted by sample composition changes, and because R2s can be inflated by
adding firms with a few years of data to models with firm dummies. In any case, the other samples yield
R2s and F-statistics that are a bit lower, but all strongly support our conclusions.
In Panel B, we again compare models (1) and (2) (and models (4) and (5)) to test for the stability
of firm fixed effects. The R2 of the firm/decade model (1) is 0.767 for the constant composition sample,
or more than double the 0.365 R2 of the firm dummy model (2), and the F-statistic of 22.11 indicates that
firm dummies differ significantly across decades. The F-statistics of 9.92 and 5.99 for the 20-plus year
1
For brevity, we do not report details of additional sensitivity checks we performed, all of which yield findings
identical in all substantive respects to those in Table 6. These checks include (i) truncation of leverage at 0.99, (ii)
extension of the constant composition sample to include observations after the 1990s, (iii) dropping observations for
the 1950s (and then for the 1950s and 1960s) from the panel B and C regressions, and (iv) dropping 208 sample
firms (from the panel B and C regressions) that Compustat footnotes indicate had finance subsidiaries whose
parents’ accounting statements were affected by mandated consolidated reporting under SFAS 94.
13
and full samples also reject stability of firm dummies, as do the F-statistics for comparing (4) and (5).2
In panel C, the regressions add control variables that have been advanced in the literature (e.g.,
Rajan and Zingales (1995)) as plausible leverage determinants and, while the R2 gap narrows relative to
those in panel B, we still find greater explanatory power for the models that allow time-variation in firm
dummies, with the F-tests still rejecting the hypothesis that they are constant over time.
Mackay and Phillips (2005, p. 1424) and Lemmon, Roberts, and Zender (2008) find that firm
fixed effects explain a majority of the total variation in leverage, and we confirm this result with an R2 of
0.561 for the full sample in column (2) of Table 6’s panel B. This is not a general result, however, and a
plausible reason is that the full sample contains many firms with just a few years of data, and inclusion of
firm dummies enables a close regression fit for such firms. [Over half the firms in the full sample have
nine or fewer years of data and more than two-thirds have 14 years or less, per Table 1.] Consistent with
this explanation, in panel A, the R2 for model (2) is 0.271 over the full 75-year period. Tellingly, the R2
climbs monotonically, and eventually doubles to 0.543 as the analysis period shortens from 75 to 20
years. Similar results hold for (2) in panels B and C, with lower R2s for the constant composition sample
(which has 50 years of data for all firms) and for firms listed 20-plus years than for the full sample.
3.2
The systematic importance of time-varying determinants of leverage
Our full sample results confirm LRZ’s (2008, p. 1589) finding that firm dummies alone generate
a much higher R2 than do year dummies alone (see the full sample R2s in (2) and (3) of panel B). This
large R2 differential is the basis for LRZ’s conclusion that time-varying determinants are unlikely sources
of capital structure heterogeneity. However, because year dummies do not isolate firm-specific sources of
time variation in leverage, their low R2 indicates only that a narrow form of time variation –
contemporaneous identical changes in expected leverage for all firms – is of limited importance. It does
not establish that all time-series sources of leverage variation are empirically inconsequential.
2
The difference in R2s between the firm/decade (1) and firm dummy (2) specifications is smaller (but still exceeds
0.100) for the full sample than in the other regressions. Roughly half the firms in the full sample are listed nine
years or less. Inclusion of so many “short-horizon” cases in the firm fixed effects model (2) will tend to increase its
explanatory power for mechanical reasons, consistent with the R2s in Table 6. Still, the full sample has markedly
higher R2s for (1), which supports our inference that leverage exhibits significant firm-specific time-series variation.
14
The R2s support the opposite inference. Firm and year dummy models ((2), (3), and (4)) leave
much variation unexplained, while firm/decade dummy models ((1) and (5)) have markedly higher R 2s.
These comparisons indicate a need to explain firm-specific sources of time variation in leverage.
Moreover, LRZ’s finding that year dummies have trivial explanatory power in part reflects their
sample period, which begins in 1965. The DJIA and constant composition samples cover earlier periods
and they show much higher R2s for year dummies (see model (3) in Table 6). These differential yeardummy R2s are explained by the fact that a sample that begins in 1965 misses the bulk of a broad-based
uptrend in leverage following World War II (see immediately below and section 7 for details).
Table 7’s evidence on relative explanatory power reinforces the view that time-series sources of
leverage variation are important. The table reports variance decompositions for models with firm fixed
effects, decade fixed effects, and firm/decade interaction effects. Firm/decade interactions account for
between 37.8% and 41.4% of the total explained variation in the DJIA sample, the constant composition
sample, and among firms listed 20-plus years. Interaction effects in the full sample account for 22.4% of
the explained variation, which is smaller than in the other samples because roughly half the firms have
nine years or less of data. With so many firms having little or no ability to register cross-decade effects, it
is all the more notable that interactions account for over one-fifth of the total explained variation.
Table 7 also shows that a nontrivial portion of the explanatory power attributed to firm fixed
effects in purely additive models is due to the suppression of interaction effects.3 When interactions are
suppressed, firm fixed effects account for 54.8% of the total explained variation in the DJIA sample, with
the percent due to firm fixed effects declining in absolute terms by 23.9% to 30.9% when interactions are
allowed. For the other three samples, we find absolute declines of 31.8%, 36.3%, and 22.0% in the
percent of explained variation attributed to firm fixed effects when interaction effects are included.
Time-series effects that are common-to-all-firms have substantial explanatory power in rows 1-4
3
Specification of an additive model is a decision to ignore interaction effects. It is not a mandate of the data, even
with one observation per cell, e.g., one leverage observation per firm per year. Scheffé (1959, section 4.8) describes
how to test for interactions with one observation per cell – impose some restrictions on admissible interactions so
that degrees of freedom are not exhausted in the estimation. In Tables 6 and 7, we apply this approach and analyze
models in which interaction effects for a given firm are assumed constant within each decade.
15
of Table 7. In additive specifications, such effects account for 45.2% of the explained variation in the
DJIA sample (row 2) and 20.8% in the constant composition sample (row 4). These figures are much
larger than the 2% of the explained variation attributed to year fixed effects in LRZ’s analysis. The large
difference exists because LRZ’s sample begins with 1965, whereas the constant composition sample goes
back to 1950 and the DJIA sample begins before the Great Depression.
In our first draft, we documented pervasive leverage increases by Compustat-listed firms during
the 1950s and 1960s, and this trend helps explain why common-to-all-firms effects are so strong in our
Table 7. For brevity here, we exclude most of the details from the first draft and simply include Figure 2,
which shows that Compustat firms engaged in wholesale abandonment of conservative leverage over the
1950s and 1960s. The increased incidence of low leverage firms in recent years (panel A) is due to a
surge in listings by young growth firms that have little or no debt. The constant composition trend (panel
B) thus offers a clearer picture of the wholesale abandonment of conservative leverage that played out
after World War II. Taggart (1985, Table 1.1) and Graham, Leary, and Roberts (2011) also report a
general post-war trend toward higher leverage, which further supports our conclusions that leverage is
generally not stable, and that leverage instability in part reflects factors common to many firms.
3.3
Caveat and summary: Limitations and lessons of panel leverage (ANOVA) analysis
An important general caveat is that significant firm fixed effects can arise without the existence
of (or permanent cross-firm differences in) target leverage ratios, and firm/decade effects can arise
without the existence of (or time-series changes in) target leverage for a given firm. For example,
significant fixed effects of both types are consistent with Myers and Majluf’s (1984) pecking-order
theory, which has no targets. With pecking-order behavior, heterogeneous external funding needs imply
that average leverage in finite samples will generally differ across firms and over time for each firm.
Such differences are what firm and firm/decade dummies capture. Significant dummies can also arise
when (i) leverage ratios yield roughly equal firm value over wide ranges, or when (ii) there are no crossfirm differences in target ratios and rebalancing is not immediate. Leverage will change in different ways
at different firms due to choices driven by factors unrelated to a target (scenario (i)), or due to exogenous
16
shocks ((i) and (ii)). With little or no pressure to reverse these changes quickly, average leverage
differences in finite samples will arise across firms and over time for a given firm, and so significant fixed
effects can arise when firms do not have (or differ in) leverage targets.
In sum, Tables 6 and 7 show that (1) firm dummies differ significantly across decades, (2) firmspecific sources of time-series variation substantially improve explanatory power, and (3) firm/decade
interactions account for a large fraction of the total explained variation in leverage. Finding (1) indicates
leverage instability in the firm-mean and cross-sectional senses, while (2) and (3) indicate a need for
theories to explain firm-specific time-series variation in leverage, as do the findings in section 4.
4.
How stable is the leverage cross section?
The degree of long-run stability of the leverage cross section is not accurately measured by the
R2s associated with firm fixed effects in panel regression (ANOVA) models of the type studied in section
3. Such R2s gauge the explanatory power of average firm differences over all years’ observations in the
sample, whereas stability over long horizons concerns the explanatory power of current leverage for
leverage in each marginal year moving forward in time. Short-run stickiness in the data implies that the
average effect exceeds the marginal over horizons of more than a few years, which means that these R2s
will overstate the long-run stability the cross section – and substantially so, as we next document.
To assess the degree of stability over time of the cross section, we gauge the explanatory power
of a given cross section for the sequence of cross-sectional “slices” going forward one year at a time, and
extending well into the future. Figure 3 reports average R2s that measure the extent to which firms with
high (or low) leverage in a given cross section tend to have high (or low) leverage in each future cross
section taken one year at a time. For the constant composition sample (panel A) and the full sample
(panel B), the vertical axis plots the average squared cross-sectional correlation coefficient over all pairs
of sample years that differ by the amount on the horizontal axis. Let (t,T) denote the cross-sectional
correlation between leverage in years t and t+T. With 59 years in the sample (1950-2008), the number of
correlations for a given T is N(T) = 59-T. Thus, the average squared correlation plotted on Figure 3’s
17
vertical axis, with T on the horizontal axis, is R2 = ∑
Figure 3 shows that the average R2 for adjacent-year leverage cross sections is around 0.8 in both
samples, but declines to about 0.4 for cross sections five years apart and to almost 0.2 for cross-sections
10 years apart. Leverage cross sections that differ by 20 years have an average R 2 a bit below 0.1, while
those for longer horizons are lower but still (barely) positive. Thus, short-run stability in the crosssectional distribution of leverage fades strongly and almost disappears over long horizons. The small but
still-positive long-term R2s are consistent with section 5’s finding that stable leverage regimes do occur
from time-to-time at individual firms. High R2s for firm fixed effects in panel regression models (e.g., as
in section 3) reflect the short-run stickiness in leverage apparent in Table 3, coupled with the fact that
Compustat contains many firms with just a few years of data.
The striking regularity in Figure 3 is the large decline in the ability of a firm’s placement in the
current leverage cross section to predict its placement in future cross sections. Simply put, firm-specific
time-series variation in leverage is so great that cross sections more than a few years apart differ
markedly, with no tendency for those differences to stabilize or reverse. Instead, the similarities between
cross sections erode as the time between them lengthens, and approach near-zero levels in the long run.
The substantial instability of the leverage cross section stands out in bold relief in Table 8, which
presents quartile decompositions of cross sections for firms listed 20-plus years. For this analysis, we
first sort firms into four groups based on Debt/TA ratios in 1950. We track forward from this year of
group formation (event year t = 0) and record the fraction of firms still in the same quartile in t = 1, 2, …,
19. We repeat the process for 1951 through 1989, treating each calendar year in turn as the initial event
year and recording the fraction of firms that are in their formation-year quartile in each future year.
[Quartile cut-off points are determined separately for each calendar year based on the cross-sectional
distribution of leverage ratios in that year.] Columns (1) to (5) report the fraction of firms always in their
initial quartile as of event year t, while columns (6) to (10) report the fraction currently in their initial
group. The table reports averages over the 40 sample runs that correspond to initial years 1950 to 1989.
18
Migration of firms across quartiles of the leverage cross section occurs pervasively. For the full
sample, only 0.072 of firms always remain in their initial quartile group through year t = 19 (column (1)
of Table 8). A remarkable 0.695 of the full sample of firms have leverage ratios that place them in three
quartiles at different times over the 20-year period, while 0.304 spend some time in all four quartiles. For
the Low/Medium and Medium/High quartiles, a trivial 0.004 and 0.003 of firms fail to move to a new
quartile ((3) and (4)). The Lowest and Highest quartiles show some persistence in group membership,
with 0.163 and 0.117 of each initial group, or about 4.1% and 2.9% of the full sample, staying in the same
quartile ((2) and (5)). Persistent presence in a given quartile does not mean that firms necessarily have
stable leverage. It does reflect leverage stability for firms always in the Lowest quartile, with the median
such firm having a range in Debt/TA of 0.054 over the 20 years (datum not tabulated). However, because
the Highest quartile is wider than the others, firms can (and do) show large variation in leverage while
staying in that quartile. Among firms always in the Highest quartile, the median range in Debt/TA is
0.246 (not tabulated), which indicates nontrivial leverage instability.
Table 8 shows a modest tendency for firms to revert back to their earlier quartile placements. If
firms were allocated randomly to groups, then 0.250 would be the expected fraction of firms currently in
their initial group. Thus, in columns (6) to (10), a decline from 1.000 to a fraction near 0.250 indicates
essentially no persistence in the sense of a greater than expected (under the null of random assignment)
incidence of future quartile placements that match firms’ initial placements. For firms initially in the
Low/Medium and Medium/High groups, the fractions in those groups in year t = 19 are 0.294 and 0.300,
or just 0.044 and 0.050 above the fractions expected under random assignment (columns (8) and (9)).
The comparable fractions for the Lowest and Highest groups are 0.422 and 0.406, or 0.172 and 0.156
above the 0.250 expected under random assignment ((7) and (10)). Among firms initially in the Lowest
quartile, 0.632 are in the top two quartiles at some point (bottom panel of (2)), and 0.329 are there at t =
19, on average (not tabulated). Among those initially in the Highest leverage quartile, 0.646 spend time
in the lowest two quartiles (bottom of (5)), and 0.333 are there at t = 19 (not tabulated). Hence, even for
the extreme quartiles, there is a large migration of firms to the opposite side of the leverage cross section.
19
Group average leverage in year t = 19 is 0.175, 0.222, 0.255, and 0.304 for firms initially in the
Lowest, Low/Medium, Medium/High, and Highest quartiles (not tabulated). These differences in crosssectional average leverage ratios confirm findings in Lemmon, Roberts, and Zender (2008). Again,
because large-sample averaging masks variation at individual firms, such differences are consistent with
substantial leverage instability at the firm and cross sectional levels. As Table 8 shows, they are also
consistent with a weak tendency for leverage to remain in roughly the same zone over long horizons.
[See also the analysis of the low frequency of stable leverage regimes in section 5.] That such a tendency
is weak is also evident in Figure 3, which documents a positive, but low correlation between cross
sections 20 years apart. Our reading of Figure 3 and Table 8 is that, while there are vestiges of long-run
stability in the data, the leverage cross section is far from stable aside from short-run stickiness.
5.
Stable leverage regimes
We use the term “stable leverage regime” to refer to situations in which a firm’s Debt/TA ratio
continuously remains within a narrow band of values. Operationally, we define a stable regime to mean
that Debt/TA for a given firm remains in a band of width 0.050, as would be the case, for example, when
it remains between 0.324 and 0.374. We also consider three weaker characterizations of a stable regime,
which are defined as situations in which Debt/TA consistently remains within bands of width 0.100,
0.150, and 0.200. Arguably, when Debt/TA differs over time by 0.150 or 0.200, the label “stable leverage
regime” is not warranted. We nonetheless analyze stability under these weaker definitions to gauge the
extent to which leverage remains in roughly the same zone for extended periods of time. Table 9 reports
the length of the longest stable leverage regime for firms listed 20-plus years (panel A) and 40-plus years
(panel B), and for the constant composition sample (panel C).
The two main findings in Table 9 are that (i) a nontrivial minority of firms has a sub-period of
moderate length in which leverage remains reasonably stable, and (ii) virtually no firms have permanently
stable regimes. On the first point, 21.3% of firms listed 20-plus years maintain leverage within a narrow
range of values that differ by no more than 0.050 for at least 10 years (row A1), i.e., about one in five
20
such firms have at least one decade-long sub-period of leverage stability. The incidence of firms with a
decade-long period of comparable leverage stability increases to 32.0% and 51.6% when we consider subsamples of firms that have been listed for longer periods (rows B1 and C1).
Stable leverage regimes over longer periods are much less common. For example, only 7.6% and
2.5% of firms in the constant composition sample have stable leverage regimes that last 20 and 30 years,
and none has a stable regime that lasts 40 years (row C1).
These low percentages indicate that long periods of stability occur only infrequently, and that
permanently stable leverage is at best a rare phenomenon. Consistent with the latter finding, Minton and
Wruck (2001) report that roughly half of firms with low leverage abandon that policy within five years.
Using weaker definitions of a stable regime that allow reasonably wide variation in Debt/TA, we
find a much higher percentage of firms with stable leverage for extended periods. For example, 100.0%
of the 157 firms in the constant composition sample have a period of at least 10 years in length in which
Debt/TA remains in a band of values that differ by no more than 0.200, while an also quite large 87.9%
consistently maintain Debt/TA in a band of similar width for at least 20 years (row C4). On the other
hand, only 14.6% of these firms maintain leverage within a band of width 0.200 for 40 years or more (row
C4). Thus, even when a stable leverage regime is defined to allow Debt/TA to vary over a reasonably
broad range, leverage stability that persists for 40 years is clearly the exception, not the rule.
Stable leverage regimes largely occur when firms have low leverage, as shown in Table 10. The
table examines samples of firms with stable regimes that last at least 20 years (panel A) and that last at
least 10 years (panel B). For this analysis, we first identify the longest stable leverage regime for each
firm (as in Table 9) and calculate median Debt/TA during each such regime. The columns of Table 10
report the frequencies with which the median leverage ratio during the stable regime falls into one of five
categories ranging from conservative (Debt/TA < 0.100) to substantial leverage (Debt/TA ≥ 0.400).
We identify 115 firms whose Debt/TA ratios remain in a band of width 0.050 for at least 20
years, and 994 firms that have similarly stable leverage for at least 10 years (rows A1 and B1 of Table
10). A remarkable 100.0% of the former firms and 88.8% of the latter have median Debt/TA ratios of
21
0.100 or less during their extended periods of stability. When we define a stable regime as a situation in
which a firm’s Debt/TA ratio consistently remains within a band of width 0.100, we also find a strong
tendency toward low leverage during the stable period. In this case, 78.8% of firms with a stable regime
of 20 years or more have a median Debt/TA ratio below 0.100 during the stable period (row A2). And
62.2% of firms with stable regimes lasting at least 10 years have a median Debt/TA ratio below 0.100.
Few firms have high leverage during periods of leverage stability. This regularity is evident from
rows A4 and B4 of Table 10, which consider a weak notion of a stable leverage regime – one in which
Debt/TA consistently remains in a band of width 0.200. Only 3.8% of the 1,015 firms with stable
regimes that last 20 years have a median Debt/TA ratio of at least 0.400 during their period of stability
(row A4). A larger but still modest 10.5% of the 4,143 firms with stable regimes that last 10 years have a
median Debt/TA ratio of 0.400 or higher during that regime. Table 10 thus identifies an interesting new
stylized fact, namely that leverage stability and financial conservatism are strongly related empirically.
6.
Departures from stability and leverage peaks and troughs
Departures from stability refer to instances in which a firm’s Debt/TA ratio moves outside the
narrow band of values it has consistently maintained for many years. Since few firms maintain Debt/TA
within a narrow band of 0.050 for extended periods (per Table 9), we focus here on firms listed 20-plus
years that consistently keep Debt/TA within a more moderate-width band of 0.100 for 10 or more years,
and then have leverage move outside that band. We align firms’ data in event time and designate event
year t = -1 as the last year of a firm’s stable regime and t = 0 as its departure year. Table 11 reports
average values for years t = -3 to t = 3 of Debt/TA, asset growth rates, capital expenditures, financing
deficits, changes in debt, and four traditionally posited leverage determinants: EBITDA, the log of sales,
the market-to-book ratio, and the fraction of assets that are tangible.4
4
Table 11 examines the 945 firms that are still listed on Compustat in year t = 3. We obtain similar results when we
analyze all firms in the 20-plus year sample regardless of how long after t = 0 they remain on Compustat. In Table
12, we focus on the sub-sample of 575 firms with complete data through t = 10, and we again obtain similar results
when we impose no restriction on how long after t = 0 a firm is on Compustat.
22
Table 11 shows that departures from stable leverage regimes are associated with increased asset
growth and contemporaneous external financing, but on average have no material relation with changes in
traditional leverage determinants. The asset growth rate doubles from 0.100 in t = -1 to 0.204 in t = 0
(row 2). By itself, this asset increase would reduce leverage, but firms also typically increase borrowing
(row 5) so that Debt/TA increases from 0.165 to 0.232 (row 1). Capital outlays increase in t = 0 (row 3),
but less dramatically than asset growth. When we examine the constant composition sample (details not
tabulated), we also find a near doubling of asset growth and a modest increase in capital expenditures, but
now the latter change is not statistically significant. Thus, while capital outlays bear some relation to
departures from stability, the stronger operating policy link is to asset growth in general, which includes
acquisitions and other outlays that fall outside the accounting classification of capital investment.
The average financing deficit (measured as a fraction of total assets) is near zero for the last three
years of the stable leverage regime, and then leaps to 0.114 in year 0 (row 4 of Table 11). This increase is
fully due to the average increase in debt (compare rows 4 and 5). Tables 11 and 14 measure the net debt
issues component of the deficit by the change in outstanding debt, but we obtain qualitatively identical
results (not tabulated) when we examine firms with non-missing Compustat data on net debt issues.
Table 12 analyzes Debt/TA ratios over the 10 years following departures from stable leverage
regimes (defined as in Table 10), and reveals that:
(i) Few firms move from one stable leverage regime to another (row 1, year t = 10).
(ii) Only about one-fifth of firms show signs of temporarily reverting to a Debt/TA ratio within
their formerly stable region (row 3).
(iii) Many firms move well away from their former stable regime, with leverage increases
outnumbering decreases by about four-to-one (compare rows 5 and 6).
(iv) In the short run, leverage reductions mainly involve debt paydown (rows 9 and 10, years t = 1
and 2).
(v) Over longer horizons, leverage reductions reflect passive deleveraging, i.e., a tendency for
asset growth to outstrip new borrowing (rows 9 and 10, years t = 5 and t = 10).
Findings (i), (ii), and (iii) indicate that departures from leverage stability are generally not followed by
reversion to the prior stable regime, and that firms rarely move from one stable regime to another.
23
Finding (iv) documents that managers systematically choose to deleverage and move below the prior
stable regime, while (v) reveals a general downward drift in leverage due to managers’ decisions not to
increase debt in proportion to asset growth.
Table 13 reports the combinations of leverage maxima (peaks) and minima (troughs) for firms
listed 20-plus years. A remarkable 77.5% of these 2,751 firms have had Debt/TA ratios below 0.100 at
some point (rows 1 and 2), while 92.8% have had Debt/TA below 0.200 (rows 1 to 3) and 42.2% have
had no debt outstanding (row 1). An also substantial 62.1% of firms have had Debt/TA ratios larger than
0.400 at other points in time. Thus, conservative leverage is observed at some point at a large majority of
firms. Aggressive leverage is much less common, with only 15.5% of firms ever having Debt/TA above
0.700 (row 10) and only 0.2% keeping Debt/TA above 0.500 consistently (rows 7 and 8).
Table 14 reports average values of leverage and other financial variables from three years before
until three years after leverage peaks (panel A) and troughs (panel B) at firms listed 20-plus years. At
peak leverage in year t = 0, the average firm’s Debt/TA ratio is well above the leverage ratios in each of
the three prior years (row A1). The leverage increase from t = -3 to t = 0 reflects debt issuances (row
A5), especially in t = 0, that proportionately outweigh large asset growth (row A2). Capital expenditures
in year t = 0 are lower than in t = -1 but not significantly so (row A3), while EBITDA is now significantly
lower (row A6). This pattern is consistent with a cash flow squeeze around the time of leverage peaks, as
is the significant increase in financing deficits at that time (row A4). Traditional target behavior would
call for a debt reduction in the face of earnings erosion, but in this case, firms borrow more (row A5).
Leverage, asset growth, capital expenditures, and the financing deficit are lower in the three years
after peak leverage (rows A1 to A4 of Table 14). By t = 3, average Debt/TA is almost back down to its
level in t = -3. With firms paying down debt in t = 1, 2, and 3 (row A5), this decline in Debt/TA reflects
deliberate managerial action rather than simply passive deleveraging due to asset growth.
The pattern around leverage peaks – a surge in debt issuance to help fund expansion followed by
debt repayment – is hard to reconcile with the view that rebalancing to a target ratio drives the time path
of leverage. Managers choose to issue debt and almost immediately start to repay it, and so these are not
24
exogenous shocks that increase leverage followed by rebalancing to a more moderate target ratio. These
choices are consistent with transitory debt financing (DeAngelo, DeAngelo, and Whited (2011)) and with
the existence of target leverage zones with “soft” boundaries (Fama and French (2005, p. 580)).
Managerial decisions around leverage troughs are also consistent with the transitory debt and
target zone scenarios. Leverage troughs in year t = 0 typically reflect chosen debt repayments in t = 0 and
in t = -1 and -2 (row B5) – i.e., arrival at a trough is something managers make happen, not an exogenous
decline in leverage. Troughs also reflect leverage decreases over t = -2, -1, and 0 (row B1) due to asset
growth (row B2) coupled with little or no external funding (row B4). Thus, managers pay down debt,
reducing Debt/TA to very low levels (row B1) instead of increasing debt in proportion to asset growth, or
keeping the existing debt and paying the cash to stockholders. A trough is typically followed by asset
growth in t = 1 (row B2) funded by a debt issuance (row B5) that proportionately outweighs the asset
increase so that Debt/TA increases (row B1). Changes in sales, tangible assets, and the market-to-book
ratio (rows B7, B8, and B9) are consistent with traditional leverage determinants encouraging the increase
in Debt/TA in t = 1. However, traditional target ratio behavior cannot explain why managers pay debt
down while assets grow substantially, and then issue debt to fund more expansion.
7.
Leverage variation at major industrial firms
The leverage plots in the appendix for the 24 DJIA firms strongly support our inference that
leverage stability is the exception, not the rule. Some firms have kept leverage within a narrow band for
extended periods, but none has permanently maintained even approximately stable leverage. Virtually all
exhibit massive time-series variation in leverage, as reported in Table 15 (column (1)). Nineteen of the
24 have had zero debt at some point, and 23 have had Debt/TA ratios well below 0.100 (column (2)).
Dramatic leverage spikes abound (columns (3) and (4)), and the leverage plots show that long and
substantial drifts – both levering up and deleveraging – are commonplace. The case studies in the
supplemental appendix identify sources of firm-specific variation in leverage, which we catalog in Table
15 (column (5)). In these case studies, we largely summarize management’s explanations for financing
25
decisions from annual reports and the financial press, but we also rely on some executive memoirs and
company histories (see the supplemental appendix for source details).
Table 15 indicates that funding expansion pervasively underlies leverage decisions, and that these
decisions sometimes also reflect (i) financial flexibility concerns, (ii) rebalancing to lower leverage, (iii)
imitation of rivals, (iv) stock market timing, and (v) the personal views of top executives.
Funding of company expansion after World War II motivated large leverage increases according
to managers of GM, GE, P&G, Allied Chemical, Sears, Caterpillar, and International Harvester. A
connection to expansion plans also exists at firms that sought to raise funds for investment while restoring
or preserving low leverage (AT&T and Exxon Mobil), and at others that issued debt to make acquisitions
(Coca-Cola, Kodak, American Tobacco, DuPont, Texaco, and ChevronTexaco). International Paper
trumpeted its post-distress deleveraging to a zero-debt capital structure, which it maintained for many
years, and then sharply levered up to fund investment.5
Deleveraging also played a role at U.S. Steel and Bethlehem Steel in stock sales presciently timed
before the 1929 Crash. U.S. Steel’s deleveraging was the brainchild of board member Myron Taylor who
was soon to become chairman. Bethlehem’s president, Charles Schwab, had waffled on his view of
dangerous stock market speculation and initially indicated he had no inclination to follow U.S. Steel, but
he changed his mind, thereby joining Myron Taylor as amazing examples of executives timing the stock
market. Consistent with Bertrand and Schoar’s (2003) “managing with style” hypothesis, these cases also
show how managers’ personal views can affect financial policy, as does the Coca-Cola case.
8.
How stable is industry-median leverage?
In this section, we gauge the stability of industry-median leverage ratios, an inquiry motivated by
Frank and Goyal’s (2009) conclusion that the strongest determinant (among many) of a firm’s current
5
See Harford, Klasa, and Walcott (2008) and Uysal (2011) on leverage and acquisitions, Mayer and Sussman
(2004) and DeAngelo, DeAngelo, and Whited (2011) on leverage and investment outlays, and Denis and McKeon
(2012) on proactive leverage changes. Shyam-Sunder and Myers (1999), Frank and Goyal (2003), Flannery and
Rangan (2006), Kayhan and Titman (2007), and Huang and Ritter (2009) find that leverage changes are associated
with external financing, with the latter three studies agreeing that such changes reflect rebalancing toward a target.
26
leverage is the contemporaneous median leverage across firms in the same four-digit SIC industry. We
find that industry-median leverage ratios exhibit substantial instability for four-digit, three-digit, and twodigit SIC industries, and for Fama and French’s 49-industry and 12-industry groups. Taken together,
these findings indicate that a time-varying factor has the strongest known influence on the leverage of
individual firms. Because industry leverage is a proxy for fundamental determinants of leverage common
to firms in a given industry, we interpret the marked instability in industry-median leverage as reinforcing
the need to understand time-varying sources of capital structure heterogeneity.
Table 16 reports leverage ranges and standard deviations of industry-median Debt/TA ratios over
1950 to 2008. A comparison of Table 16 with the firm-specific statistics in Tables 1 and 2 shows that, for
four-digit and three-digit SIC industries, the ranges and standard deviations of industry-median Debt/TA
have distributions that are quite similar to those for individual firms. Most four- and three-digit industries
have relatively few firms and so it makes sense that the time-series variation in industry-median leverage
is comparable to that for individual firms.
Because they include many firms, the two-digit SIC and Fama and French industry groupings
show a “portfolio effect” dampening of industry-median leverage, but these industries nonetheless exhibit
substantial time-series variation in median leverge. For example, the typical two-digit SIC industry has a
range in median Debt/TA of 0.319 (panel A of Table 16), and its time-series standard deviation of 0.075
(panel B) translates to a plus/minus two-sigma band of width 0.300. Even at the high level of aggregation
of Fama and French’s 12-industry grouping, the typical industry has a range in median leverage of 0.230
(panel A) and a time-series standard deviation of 0.058 (panel B), which implies a plus/minus two-sigma
band of 0.232 in industry-median Debt/TA.
For all industry definitions, Table 17 shows strong signs of systematically important influences of
time-varying factors on industry-median leverage. This table reports panel leverage regressions and
variance decompositions for industry-median Debt/TA ratios analogous to the firm-specific results in
Tables 6 and 7. In all cases, introduction of industry/decade dummy variables into the panel regressions
substantially increases R2s and these interaction effects are highly significant. For example, at the four27
digit SIC level, industry-decade dummies have an R2 of 0.619, which is much higher than the R2 of 0.352
for the model that includes only industry fixed effects.
The variance decompositions highlight the systematic influence on industry-median leverage of
time-varying factors, both industry-specific and common-to-all industries. For example, industry/decade
interaction effects account for 45.3% of the total explained variation at the four-digit level and for 19.5%
in the 12-industry grouping. Decade fixed effects account for 21.4% of the explained variation for the 12industry grouping, implying that 40.9% of the total explained variation reflects time-varying factors,
despite the dampening effect on median leverage of having many firms in an industry. The 49-industry
grouping shows almost as much common-to-all industries variation (15.8%) and greater industry-specific
variation (33.0%). The explanatory power of decade fixed effects for the 12- and 49-industry groups
supports section 3’s inference of a general post-war trend away from conservative leverage.
9.
Summary and implications of the evidence
This paper presents a comprehensive picture of the nature and extent of stability in corporate
capital structures. The evidence contradicts the view that leverage is quite stable over long horizons. The
existence of significant firm fixed effects in panel leverage regressions, which is the empirical foundation
of prevailing wisdom, does not establish stability of firm-mean leverage, or stability of the leverage cross
section over long horizons. Nor does it imply that firms have target leverage ratios. We find that firm
fixed effects differ significantly across decades, rejecting stability of firm-mean leverage and the cross
section. Leverage cross sections more than a few years apart differ markedly, with differences growing
over time – and not reverting or stabilizing – until there is almost no similarity with earlier cross sections.
Migration of firms over the cross section is substantial and pervasive, with many firms appearing in three
or four different leverage quartiles over the typical 20-year period.
Substantial within-firm variation in book leverage, market leverage, and the net-debt ratio is the
norm, but episodes of stability do arise occasionally. Leverage stability at individual firms occurs mainly
at low leverage, and is virtually always temporary. Departures from stability are infrequently followed by
28
rebalancing to the old stable leverage regime or by establishment of a new stable regime. Departures
from stability and the attainment of leverage peaks and troughs are associated with company expansion
and external funding of that expansion. Leverage peaks and troughs are not exogenous shocks that induce
rebalancing to more moderate leverage; instead, managers typically choose to (i) issue debt and fund
expansion when attaining a peak, then immediately start repaying debt, and (ii) pay down debt when
attaining a trough, then issue debt and fund expansion. Compustat-listed firms abandoned conservative
capital structures en masse during the 1950s and 1960s, and case-based evidence indicates this is
associated with funding of expansion during the booming post-war economy. Industry-median leverage,
which prior research indicates is the strongest known determinant of a given firm’s leverage, also exhibits
substantial instability over time.
These findings imply that credible theories of capital structure must be able to explain significant
leverage instability, and they point to (firm-specific and common-to-all-firm) time-varying factors as
important determinants of leverage. As we next discuss, the findings also provide important guidance
about existing and potential theories.
Cross-firm and time-series variation in leverage. If leverage were stable over long horizons,
then explaining cross-firm variation would be a major research puzzle, and time-series variation would be
noise of little or no interest. In fact, both types of leverage variation are systematically important, and the
two are not appropriately treated as separable issues. Although cross-firm variation is substantial at any
given point in time, the leverage cross section is also far from stable over time. Therefore, development
of theories that can adequately explain the substantial time-series variation in leverage at individual firms
is not only important in its own right, but it is also essential to explain the (markedly different) crosssectional distributions of leverage that prevail at different points in time.
Instability of the leverage cross section. A significant puzzle for theorists is to explain why the
relative positions of firms in the leverage cross section are sticky in the short run, but far from stable over
horizons of more than a few years, with similarities between cross sections evaporating as the time
between them lengthens. This strong empirical regularity suggests that the evolution of leverage mainly
29
reflects transitory (not necessarily random) factors that generally out-weigh any tendency for leverage to
converge to, or hover near, stable permanent components. The evaporation of systematic commonalities
between disparate cross sections contradicts theories that predict that firms adhere closely to stationary (or
near-stationary) target leverage ratios.
Target leverage ratios. At the most basic level, the fact that leverage varies so much at so many
firms is at odds with theories in which keeping leverage close to a constant target ratio is an important
objective. To be clear, wide time-series variation in leverage is consistent with theories in which firms
have constant target ratios, but face only small costs when leverage differs markedly from target.
However, theories of this type are target ratio-driven in name – but not in substance – because they imply
that a desire to keep leverage near a particular (ideal) ratio has little effect on behavior. Such theories are
most appropriately viewed as variants of target zone theories (see below) in which there are only secondorder value differences across a reasonably broad subset of leverage ratios. They are conceptually distinct
from the (empirically problematic) theories that predict that firms have economically meaningful targets
in the sense of strong incentives to keep leverage close to a stationary level.
Theories with time-varying target ratios could explain wide leverage variation, but they struggle
to plausibly explain other aspects of the time path of leverage, e.g., why managers choose to move to a
peak or trough, then immediately reverse course, and why firms generally do not move from one stable
regime to another. These findings suggest that important ingredients are missing from theories in which
the benefit of moving toward (time-varying or stationary) target ratios drives leverage behavior. One
plausible such ingredient is the value of the option to move temporarily away from a target leverage ratio
to fund investment, e.g., as in DeAngelo, DeAngelo, and Whited (2011). In any case, our evidence does
not rule out distress costs and taxes as material influences on leverage. On the contrary, something like
distress costs must encourage rebalancing downward from very high leverage, since we find that many
firms have Debt/TA ratios above 0.500 at some point, but almost no firms keep Debt/TA consistently
above 0.500 for long periods of time.
Target leverage zones or ceilings. Our evidence is fully consistent with theories in which firms
30
have target leverage zones, provided that the bounds on allowed leverage are wide, or that managers treat
nominally narrow boundaries as “soft” or flexible limits. Target “ceiling” is arguably more descriptive
than target “zone,” given that many firms have Debt/TA ratios below 0.100 at some point, while ratios
above 0.700 are much less common, and it is rare to find firms with Debt/TA permanently above 0.500.
The notion that firms put ceilings on acceptable leverage is consistent with the importance that CFOs
attach to maintaining a given credit rating (Graham and Harvey (2001)), and with firms’ lower propensity
to issue debt when borrowing is more likely to trigger a rating downgrade, or soon after a downgrade
occurs (Kisgen (2006, 2009)).
The plausibility of target zone theories draws further support from Graham and Harvey (2001),
who find that 37% of CFOs say their firms have “flexible” leverage targets and 19% say their firms have
no target. Another 34% have a “somewhat tight” target and only 10% say they have a “tight” target debt
ratio. Thus, at least 56% (37% + 19%) of firms apparently allow non-trivial variation in leverage. It is
tough to say how well the other 44% of firms fit with target zone theories, absent data on (i) whether
managers treat nominally tight targets as rigid rules or as non-binding financial-planning guides, and (ii)
how much managers actually change (or violate) their tight or somewhat tight targets.
The key feature of target zone theories is that, over a reasonably wide range, the choice of
leverage does not have first-order value consequences that would provide strong incentives to keep
leverage consistently close to a target ratio. This view hearkens back, of course, to Modigliani and Miller
(1958). However, the point here is not that the debt/equity mix is literally irrelevant or that leverage
evolves as a neutral mutation, as Miller (1977) conjectured. The basic point is: Leverage varies so widely
at so many firms that it becomes hard to believe in large benefits from a particular level. It seems more
plausible that, over a fairly wide range, leverage per se is of second-order import for firm valuation, so
the main determinants of leverage are factors other than the benefits of adhering closely to a target ratio.
Is leverage determined as a residual? The capital structure literature treats determination of the
debt/equity mix as the central problem of financial policy, which fosters a natural reluctance to think
leverage might be of second-order import. The latter conjecture gains plausibility, however, when the
31
corporate finance literature is viewed holistically. How can the following four statements all be true? (1)
Firms adhere closely to target leverage ratios. (2) Lintner-style target payout ratios govern dividend
distributions. (3) Managers are reluctant to cut dividends and to sell equity. (4) Firms require capital to
fund investment, and they often obtain outside funds. Simply put, this system is over-determined, and all
four statements cannot be descriptive. Something has to give. This inference is closely related to
Lambrecht and Myers’ (2012) conclusion that, as a matter of logic, target-adjustment models for payout
and capital structure cannot co-exist. Their reasoning is that a firm’s budget constraint implies that a
dynamic theory of payout and investment effectively dictates a dynamic theory of capital structure.
Empirically, if “stylized facts” (2), (3), and (4) are descriptive, then the wide leverage variation
we observe plausibly reflects the determination of leverage as a residual by-product of other time-varying
components of financial policy. We are not claiming that the debt/equity mix is a “pure residual” that is
forced to adapt because investment, payout, and equity-issuance considerations are always more
important. But (2), (3), and (4) do have strong empirical support, and so it seems important to take
seriously the possibility that leverage often tends to evolve along residual lines. A related and arguably
more plausible possibility is that the shortcomings of target ratio theories might be repaired by attention
to trade-offs between investment/payout/equity-issuance objectives and desired adaptation to leverage
targets. These possibilities merit careful consideration by theorists and in future empirical studies, given
that they could plausibly explain the substantial instability that characterizes real-world capital structures.
Funding investment versus rebalancing the debt/equity mix. Our reading of the data is that
credible theories of capital structure will likely emphasize the funding of investment, e.g., as in Myers and
Majluf (1984), but without the strict pecking order, and as in the Hennessy and Whited (2005) class of
dynamic models. This judgment reflects buy-in to the premise that investment policy is the main driver
of corporate value. This premise suggests that funding investment is more important in determining the
time path of leverage than benefits of rebalancing the debt/equity mix of payouts. Whatever the ultimate
judgment on that conjecture, it seems unwarranted to ignore connections between the leverage time path
and the funding of investment, given the empirical association between company expansion and (i)
32
departures from stable leverage regimes, (ii) leverage peaks and troughs, and (iii) the post-war
abandonment of conservative leverage policies. Credible theories will almost surely include other timevarying factors such as credit-market conditions, stock market timing, valuation disagreements between
managers and investors, and managerial attitudes and social norms about debt.
Bottom line. Capital structure stability is the exception, not the rule. As a factual description of
real-world capital structures, the prevailing view that leverage is stable over long horizons is far wide of
the mark. For development of credible theories, the data thus call for greater attention to time-varying
determinants of leverage and less emphasis on leverage adhering to a particular target ratio. Solving the
capital structure puzzle will require a thorough understanding of the determinants of time-series variation
in leverage, and this is true not only to explain firm-specific leverage behavior, but also to explain the
cross-sectional distribution of leverage that prevails at any given point in time.
33
Figure 1
Leverage Ratios of General Motors, IBM, and Eastman Kodak: 1926 to 2008
Book leverage is the ratio of total book debt to total assets. Market leverage is total book debt divided by the sum of
total book debt and the market value of common stock. Leverage data are from company annual reports, Moodys
manuals, and Compustat. Market values are from CRSP.
1.000
General Motors
0.800
Market leverage
0.600
0.400
Book leverage
0.200
1926
1929
1932
1935
1938
1941
1944
1947
1950
1953
1956
1959
1962
1965
1968
1971
1974
1977
1980
1983
1986
1989
1992
1995
1998
2001
2004
2007
0.000
0.600
IBM
0.500
Book leverage
0.400
Market leverage
0.300
0.200
0.100
1926
1929
1932
1935
1938
1941
1944
1947
1950
1953
1956
1959
1962
1965
1968
1971
1974
1977
1980
1983
1986
1989
1992
1995
1998
2001
2004
2007
0.000
0.500
Eastman Kodak
0.400
0.300
Market leverage
0.200
Book leverage
0.100
1926
1929
1932
1935
1938
1941
1944
1947
1950
1953
1956
1959
1962
1965
1968
1971
1974
1977
1980
1983
1986
1989
1992
1995
1998
2001
2004
2007
0.000
Figure 2
Conservatively Levered versus Highly Levered Publicly Held Industrial Firms: 1950 to 2008
Leverage is measured as the ratio of the book value of total debt to the book value of total assets (Debt/TA). The
constituent firms in the full sample vary from year to year (per our sampling criteria). The constant composition
sample contains the sub-sample of 157 firms with non-missing total assets on Compustat in 1950 that remained
listed through at least 2000. The constant composition sample is unchanged over 1950 to 2000, but contracts over
2001 to 2008 due to the delisting of some firms. Conservatively levered firms are defined as those with no debt
outstanding, while highly levered firms are defined as those with Debt/TA > 0.400.
A.
Full sample incidence of conservatively levered and highly levered firms
0.400
Fraction of full sample that has no debt
0.350
Fraction of full sample that has Debt/TA > 0.400
0.300
0.250
0.200
0.150
0.100
0.050
1950
1952
1954
1956
1958
1960
1962
1964
1966
1968
1970
1972
1974
1976
1978
1980
1982
1984
1986
1988
1990
1992
1994
1996
1998
2000
2002
2004
2006
2008
0.000
Constant composition sample incidence of conservatively levered and highly levered firms
0.400
Fraction of constant composition sample that has no debt
0.350
0.300
Fraction of constant composition sample that has Debt/TA > 0.400
0.250
0.200
0.150
0.100
0.050
0.000
1950
1952
1954
1956
1958
1960
1962
1964
1966
1968
1970
1972
1974
1976
1978
1980
1982
1984
1986
1988
1990
1992
1994
1996
1998
2000
2002
2004
2006
2008
B.
Figure 3
Extent of Stability in the Cross-section of Leverage
These figures present average R2s that measure the extent to which high (or low) leverage in a given year’s leverage
cross-section corresponds to high (or low) leverage in future years’ cross-sections. Leverage is measured as the
ratio of total debt to total assets in book value terms. Figure 3A is based on the constant composition sample and
Figure 3B is based on the full sample, with both using 58 years of data (1950 to 2008). The horizontal axis denotes
the number of years between leverage cross-sections. The vertical axis plots the average squared correlation
coefficient over all pairings of sample years that differ by the amount specified on the horizontal axis. For example,
to generate the average R2 for the one-year difference in cross-sections, we first identify all firms with leverage data
in 1950 and 1951, and obtain the correlation between leverage in the two years. We repeat this process for 1951 and
1952 treated as a pair, then 1952 and 1953, and so on, and report in the figure the average R 2 across all pairings that
differ by exactly one year. In general, to obtain the average R 2 for a T-year difference in cross-sections, we repeat
this process using the following pairs of years: 1950 and (1950 + T), 1951 and (1951+T), 1952 and (1952+T), and
so on. Confidence intervals, (two standard error bands in dashes), are obtained with a bootstrap procedure, resampling with replacement the individual squared correlations for each value of T, using 1,000 sample replications.
A. Constant composition sample: Average R2 versus number of years between leverage cross-sections
0.9
0.8
0.7
R-Square
0.6
0.5
0.4
0.3
0.2
0.1
0
1
3
5
7
9
11
13
15 17 19 21 23 25 27
T, Years between cross-sections
29
31
33
35
37
39
33
35
37
39
B. Full sample: Average R2 versus number of years between leverage cross-sections
0.9
0.8
0.7
R-Square
0.6
0.5
0.4
0.3
0.2
0.1
0
1
3
5
7
9
11
13
15 17 19 21 23 25 27
T, Years between cross-sections
29
31
Table 1
Time-Series Range of Book Leverage, Market Leverage, and Net-Debt Ratios of Publicly Held Industrial Firms
Book leverage is the ratio of total debt to total assets. Market leverage is the ratio of debt to the sum of debt plus the market value of common stock. The netdebt ratio equals debt minus cash, divided by total assets. The sample contains 15,096 industrial firms in the CRSP/Compustat file over 1950-2008. The sample
excludes firms (i) with SIC codes in the ranges 4900-4949 (utilities) or 6000-6999 (financials), (ii) incorporated outside the U.S., or (iii) without CRSP security
codes of 10 or 11. A firm enters our sample in the first year that it has a nonmissing value for total assets on Compustat and a nonmissing share price on CRSP
(or Compustat). It remains in the sample as long as Compustat continues to report nonmissing values of total assets and the firm’s shares have not been delisted.
The constant composition sample contains 157 firms that are included in the sample in 1950 and remain until at least 2000. Panels A and C exclude the 0.22% of
firm-year observations with book leverage over 1.000, while panel B excludes the 1.67% of firms (almost all from the 2 to 4 year group) with insufficient equity
value data to measure the range of market leverage. The far right column gives the firm counts before these sample exclusions.
Percent of firms with a range of leverage ratios in the interval:
0.000
0.100
0.200
0.300
0.400
Above
to 0.100
to 0.200
to 0.300
to 0.400
to 0.500
0.500
A. Book Leverage
2.3%
6.1%
19.9%
23.4%
19.2%
29.2%
5.2%
11.8%
20.7%
20.9%
16.1%
25.3%
11.8%
15.1%
20.6%
17.8%
13.0%
21.8%
20.7%
20.7%
18.9%
14.9%
9.9%
15.0%
47.7%
19.9%
12.7%
9.1%
4.8%
5.8%
------------0.0%
1.3%
15.9%
32.5%
24.8%
25.5%
Years on
Compustat
Median
range
20-plus
15 to 19
10 to 14
5 to 9
2 to 4
1
Constant comp sample
0.391
0.357
0.314
0.241
0.110
--0.400
20-plus
15 to 19
10 to 14
5 to 9
2 to 4
1
Constant comp sample
0.536
0.462
0.393
0.294
0.117
--0.507
3.5%
9.1%
14.3%
23.5%
46.9%
--0.6%
B. Market Leverage
5.3%
8.0%
8.1%
10.5%
10.9%
12.3%
14.3%
13.1%
16.7%
12.7%
----4.5%
10.2%
12.4%
13.6%
13.7%
13.3%
8.3%
--14.7%
15.4%
14.9%
12.7%
11.5%
6.0%
--19.1%
20-plus
15 to 19
10 to 14
5 to 9
2 to 4
1
Constant comp sample
0.599
0.574
0.527
0.424
0.250
--0.624
0.0%
0.3%
0.4%
2.8%
21.9%
--0.0%
C. Net-Debt Ratio
1.0%
5.4%
2.5%
9.1%
5.2%
12.6%
11.6%
15.8%
20.4%
15.0%
----0.0%
1.3%
12.3%
12.5%
14.6%
16.0%
12.1%
--8.9%
15.5%
15.1%
13.9%
13.5%
8.7%
--16.6%
Median
ratio
Number
of firms
0.211
0.195
0.189
0.179
0.173
0.158
0.208
2751
1514
2408
3740
3779
904
157
55.4%
43.8%
36.2%
24.4%
9.4%
--51.0%
0.221
0.167
0.159
0.128
0.098
0.076
0.219
2751
1514
2408
3740
3779
904
157
65.9%
60.4%
53.4%
40.3%
21.9%
--73.3%
0.135
0.098
0.086
0.071
0.038
-0.008
0.140
2751
1514
2408
3740
3779
904
157
Table 2
Time-Series Standard Deviation ( ) of Book Leverage, Market Leverage, and Net-Debt Ratios of Publicly Held Industrial Firms
The time-series standard deviation of leverage, , is based on the maximum likelihood estimator, which uses a divisor equal to the number of observations, N, in
each firm’s time series, not N-1. Column (3) reports the cross-sectional standard deviation of for all firms in the sample for the row in question. Column (4)
reports the correlation between the time-series standard deviation of leverage and the range in leverage. Book leverage is the ratio of total debt to total assets.
Market leverage is the ratio of debt to the sum of debt plus the market value of common stock. The net-debt ratio equals debt minus cash, divided by total assets.
The sample contains 15,096 industrial firms in the CRSP/Compustat file over 1950-2008, with other sampling conditions as described in Table 1.
Average
Median
Years on
Compustat
(1)
(2)
Cross-sectional
standard
deviation of
(3)
20-plus
15 to 19
10 to 14
5 to 9
2 to 4
Full sample
Constant comp sample
0.115
0.116
0.108
0.098
0.073
0.098
0.109
0.106
0.106
0.098
0.084
0.049
0.088
0.106
0.054
0.064
0.069
0.075
0.077
0.072
0.040
20-plus
15 to 19
10 to 14
5 to 9
2 to 4
Full sample
Constant comp sample
0.146
0.137
0.127
0.114
0.081
0.117
0.135
0.144
0.136
0.124
0.104
0.053
0.110
0.128
0.067
0.077
0.081
0.087
0.086
0.085
0.061
B. Market Leverage
0.944
7.8%
0.966
14.6%
0.973
20.6%
0.985
29.3%
0.991
49.0%
0.952
27.0%
0.943
6.4%
18.1%
19.3%
19.7%
19.1%
18.6%
18.9%
26.1%
28.0%
24.0%
21.7%
19.3%
13.2%
20.4%
33.1%
20-plus
15 to 19
10 to 14
5 to 9
2 to 4
Full sample
Constant comp sample
0.167
0.180
0.175
0.170
0.144
0.164
0.159
0.153
0.161
0.156
0.145
0.109
0.143
0.153
0.076
0.093
0.095
0.107
0.125
0.104
0.059
C. Net-Debt Ratio
0.925
1.0%
0.957
2.1%
0.967
3.1%
0.978
7.3%
0.987
24.5%
0.919
9.3%
0.882
0.0%
17.3%
16.1%
19.3%
22.2%
22.6%
20.2%
16.6%
30.1%
26.4%
25.1%
22.6%
15.9%
23.1%
33.1%
Correlation
Percent of firms with standard deviation
between
0.000
0.050
0.100
0.150
and range
to 0.050
to 0.100
to 0.150
to 0.200
(4)
(5)
(6)
(7)
(8)
A. Book Leverage
0.926
7.5%
37.6%
32.6%
14.5%
0.957
13.3%
32.9%
28.8%
14.3%
0.968
20.4%
30.5%
24.5%
14.3%
0.982
29.9%
28.9%
19.3%
11.8%
0.990
50.6%
20.9%
13.5%
7.4%
0.941
27.6%
29.2%
22.3%
11.9%
0.859
2.6%
43.3%
40.1%
9.6%
in the interval:
0.200
Above
to 0.250
0.250
(9)
(10)
5.9%
6.9%
6.4%
5.3%
4.0%
5.4%
3.8%
1.8%
3.9%
3.9%
4.9%
3.6%
3.7%
0.6%
23.9%
21.8%
18.1%
14.5%
7.6%
16.0%
21.0%
15.7%
11.7%
12.3%
9.1%
5.9%
10.4%
8.9%
6.6%
8.7%
7.6%
8.7%
5.8%
7.4%
4.5%
23.9%
21.4%
19.8%
17.0%
12.0%
18.0%
28.0%
13.8%
14.5%
13.7%
11.4%
7.5%
11.6%
15.3%
14.0%
19.6%
19.1%
19.5%
17.5%
17.8%
7.0%
Table 3
Median Time-Series Correlation Between Book Leverage and Market Leverage:
Publicly Held Industrial Firms Partitioned by Range of Book Leverage
Book leverage is the ratio of total debt to total assets (Debt/TA). Market leverage is the ratio of debt to the sum of debt plus the market value of common stock.
For a given firm, the correlation between book and market leverage is calculated based on all sample years with nonmissing values of both leverage ratios. The
column partitions and row definitions are identical to those for book leverage in panel A of Table 1. The sample inputs are identical to those for Table 1, with
attention restricted to firms with at least two years of data in order to obtain meaningful correlation estimates.
Median correlation between book and market leverage
among firms with book leverage in the interval:
0.000 to 0.100 0.100 to 0.200 0.200 to 0.300 0.300 to 0.400
0.958
0.824
0.794
0.808
0.948
0.828
0.831
0.844
0.947
0.795
0.818
0.861
0.898
0.840
0.868
0.893
0.994
0.972
0.983
0.983
Number of years listed
20-plus
15 to 19
10 to 14
5 to 9
2 to 4
All firms in row
0.820
0.842
0.857
0.885
0.986
Constant composition sample
0.787
---
0.741
0.743
Full sample
0.878
0.959
0.876
0.856
0.400 to 0.500
0.822
0.827
0.871
0.916
0.984
Above 0.500
0.832
0.845
0.858
0.905
0.988
0.767
0.807
0.790
0.864
0.871
0.871
Table 4
Inter-temporal Variation in Book Leverage:
Magnitude and Speed of Departure from Original Leverage
Leverage is measured as the book value of total debt divided by the book value of total assets. The sample consists
of 2,751 firms with 20 or more years of data in the CRSP/Compustat file over 1950-2008. The first column indexes
event years relative to the date of the first leverage observation (at event year 0) for each firm in the sample. The
remaining four columns give the fraction of sample firms that, at some point up to the event date in question, has
had a Debt/TA ratio outside the specified interval. For example, the first entry in the third row indicates that, as of
three years after each firm’s initial leverage observation, 71.1% of sample firms have had a leverage ratio that is
more than 0.050 above (or more than 0.050 below) the leverage ratio that the firm had in year 0.
Year
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
Percent of firms for which book leverage has differed from its original value by at least:
+/- 0.050
+/- 0.100
+/- 0.200
+/- 0.300
+/- 0.400
41.1%
21.7%
7.0%
2.6%
1.0%
61.3%
37.0%
14.7%
6.0%
2.7%
71.1%
47.4%
20.8%
9.2%
3.9%
77.2%
54.8%
25.3%
11.5%
5.2%
81.5%
61.1%
29.7%
13.8%
6.6%
85.1%
66.5%
33.8%
16.3%
7.9%
87.6%
70.9%
38.2%
18.6%
9.5%
89.2%
74.3%
41.7%
21.0%
10.6%
90.6%
77.0%
44.4%
22.2%
11.5%
91.7%
79.7%
47.9%
24.6%
13.0%
92.6%
81.1%
50.5%
26.4%
14.1%
93.2%
82.8%
52.3%
27.5%
15.0%
93.8%
84.0%
54.4%
29.1%
16.0%
94.3%
85.6%
56.4%
30.6%
16.8%
95.0%
86.8%
58.1%
32.8%
17.9%
95.4%
88.3%
60.4%
34.3%
18.9%
96.0%
89.7%
62.2%
36.4%
20.3%
96.4%
90.4%
64.4%
38.1%
21.4%
96.8%
91.4%
66.5%
39.5%
22.2%
Table 5
Incidence of Mean Leverage Changes at Firms in the Constant Composition Sample
For each of the 157 firms firm in the constant composition sample, we conduct t-tests to assess the null hypothesis
that the mean Debt/TA ratio is equal over the earlier and later halves of the sample period. All firms in the constant
composition sample have data from 1950 through at least 2000, with observations potentially continuing through
2008. We tally the frequency across all firms of null hypothesis rejections for two-sided tests conducted at the 0.05
level, and report those rejection rates in row (1). We also report rejection frequencies when the (one-sided)
alternative hypothesis is that the later-period mean exceeds the earlier-period mean (row (2)), and that the earlierperiod mean exceeds the later-period mean (row (3)). The tests in the first column assume that the variances are
equal across the two periods, while those in the second column allow the variances to differ. Since these t-tests are
not independent across firms, the third column reports the significance level at which a Hotelling T 2 test rejects the
null hypothesis of equality of means for all firms. For the Hotelling test, we employ a quasi-bootstrap approach that
takes 100 random samples (with replacement) of 10 firms from the constant composition sample. We select samples
of size 10 because of the need for the test to be conducted with a non-singular estimated covariance matrix. In all
100 cases, the p-level for rejection of the null is below 0.0001. These calculations exclude Debt/TA ratios in excess
of 1.00.
Rejection frequency:
null hypothesis of equality of means
(constant variance)
(unconstrained variance)
Hotelling T2
significance level of
null rejection
1. Two-sided alternative
72.6%
72.0%
< 0.0001
2. Later mean > Earlier mean
62.4%
62.4%
---
3. Earlier mean > Later mean
14.0%
14.0%
---
Table 6
Incremental Explanatory Power of Firm Fixed Effects that Vary by Decade
The dependent variable is the ratio of debt-to-total assets (Debt/TA)jt where firms are indexed by j and years are indexed by t. All F-statistics indicate significant
differences at p-levels less than 0.0001. The first F-statistic gives the appropriate statistic to test the hypothesis that model (1), which allows firm fixed effects to
vary across decades, is indistinguishable from model (2) in which each firm has a dummy variable that is assumed constant over the analysis period. The second
F-statistic tests the same hypothesis for regressions (4) and (5), which differ from (1) and (2) by the inclusion of year dummies. Panel A reports regression
results for the 24 firms in the DJIA sub-sample, with data covering 1926 to 2000. The regressions in panel C differ from those in panels A and B by inclusion of
other control variables often hypothesized to affect leverage decisions: log (sales), market-to-book ratio, profitability, and asset tangibility. For the constant
composition analysis, we work with the five decades from the 1950s through the 1990s. For the analysis in the 20-plus year and full sample regressions, the
initial year for a given firm can be later than 1950 and the last year can be as late as 2008. In Regressions (2) and (4), the “firm dummy” variable for firm j takes
the value 1 for all observations corresponding to that firm, and the value 0 otherwise. In Regressions (3), (4), and (5), the “year dummy” variable for year t takes
the value 1 if the observation is for year t, and 0 otherwise. In Regressions (1) and (5), the “firm/decade” variables are decade-specific dummy variables for each
firm. The first firm/decade dummy for firm j takes the value 1 if the year falls in the first calendar decade in the estimation, and the value 0 if it falls outside that
decade or if it corresponds to any other firm. The second firm/decade dummy for firm j takes the value 1 if the year falls in the second decade of the estimation,
and the value 0 if it falls outside that decade or if it corresponds to any other firm. And so on for firm/decade dummies corresponding to each subsequent decade
for firm j. We exclude the 0.22% of observations with leverage above 1.00. The results are statistically indistinguishable when leverage is truncated at 0.99.
A. Basic regressions: DJIA sample
1926 to 2000
1931 to 2000
1941 to 2000
1951 to 2000
1961 to 2000
1971 to 2000
1981 to 2000
B. Basic regressions
Constant composition sample
Firms listed 20-plus years
Full sample
C. Regressions with ancillary controls
Constant composition sample
Firms listed 20-plus years
Full sample
Adjusted-R2 for model with:
Firm dummies
Firm
Year
and
dummies
dummies
year dummies
(2)
(3)
(4)
Firm/decade
dummies and
year dummies
(5)
F-statistic
to compare
(4) versus (5)
0.503
0.518
0.555
0.588
0.612
0.560
0.557
0.856
0.851
0.840
0.832
0.810
0.774
0.676
25.50
25.48
20.93
17.79
13.70
13.83
7.37
0.108
0.030
0.028
0.477
0.496
0.574
0.784
0.717
0.704
18.19
9.53
5.87
0.173
0.115
0.121
0.518
0.532
0.616
0.747
0.727
0.735
11.31
8.40
5.72
F-statistic
to compare
(1) versus (2)
Firm/decade
dummies
(1)
38.68
38.88
31.32
23.15
16.12
13.87
6.95
0.841
0.836
0.821
0.810
0.780
0.761
0.657
0.271
0.291
0.358
0.461
0.520
0.544
0.543
0.218
0.212
0.179
0.104
0.067
-0.007
-0.010
22.11
9.92
5.99
0.767
0.709
0.697
0.365
0.471
0.561
11.26
8.26
5.70
0.728
0.719
0.730
0.485
0.523
0.610
Table 7
Relative Explanatory Power of
Firm Fixed Effects, Decade Fixed Effects, and Firm/Decade Interaction Effects
The table presents variance decompositions for two-way ANOVA models that include firm fixed effects, decade fixed effects, and firm/decade interaction
effects. We analyze balanced panels for both the 157 firms in the constant composition sample with data on Compustat from 1950 through at least 2000, and for
the 24 firms in the DJIA sub-sample with data back to at least 1926. For this balanced panel analysis, the DJIA sample runs from the 1930s to the 1990s, while
the constant composition sample runs from the 1950s to the 1990s. We analyze an unbalanced panel for both the sample of 2,157 firms listed at least 20 years
and for the full sample. The percentages in the table are the type III sum of squares explained by each given effect relative to the total explained by all effects
included in the model. Because of computational limits with the full sample, we take 100 random samples of 1,510 firms (10% of the total of 15,096 firms) and
report the average over the 100 sample runs.
Percent of explained variation accounted for by:
Firm/decade
Firm
interaction effects
fixed effects
DJIA sample
1. Interaction-inclusive model
2. Purely additive model
Constant composition sample
3. Interaction-inclusive model
4. Purely additive model
Firms listed 20-plus years
5. Interaction-inclusive model
6. Purely additive model
Full sample
7. Interaction-inclusive model
8. Purely additive model
Decade
fixed effects
41.4%
----
30.9%
54.8%
27.7%
45.2%
40.2%
----
47.4%
79.2%
12.4%
20.8%
37.8%
----
60.5%
96.8%
1.7%
3.2%
22.4%
----
76.8%
98.8%
0.8%
1.2%
Table 8
Fraction of Firms Always in and Currently in Their Initial Leverage Quartile
We start with calendar year 1950 and sort firms into four equal-sized groups based on their Debt/TA ratios in that year. We track forward from this year of group
formation (event year t = 0) and record the fraction of firms that remain in the same group for event years t = 1, 2,…, 19. We repeat the process for 1951,
1952,…, 1989, treating each of these calendar years in turn as the initial event year and then noting the quartile location of each firm in each of the subsequent 19
years. In columns (1) to (5), we report the average over all 40 calculations of the fraction of firms that have remained in a given formation-year leverage group in
every year up to the event year in question. For example, in column (1), the year t = 19 entry of 0.072 indicates that an average of 7.2% of firms remain in the
same quartile for 20 years. In columns (6) to (10), we report the average over all 40 calculations of the fraction of firms that are currently in their formation-year
leverage group in the event year (even though they may have left that group sometime after t = 0 but before the current year). The rows at the bottom of the table
give the fractions of firms in 4 different quartiles, at least 3 different quartiles, and at least 2 quartiles at different times over the 20 years. The table analyzes the
sample of CRSP/Compustat industrial firms with at least 20 years of data.
Fraction of firms always in initial (t = 0) leverage quartile:
Full
Lowest
Low/Medium
Medium/High
Highest
Years
sample
leverage
leverage
leverage
leverage
elapsed
(1)
(2)
(3)
(4)
(5)
0
1.000
1.000
1.000
1.000
1.000
1
0.720
0.829
0.638
0.617
0.796
2
0.556
0.717
0.432
0.409
0.667
3
0.450
0.637
0.304
0.284
0.573
4
0.373
0.574
0.216
0.201
0.500
5
0.315
0.521
0.153
0.142
0.443
6
0.270
0.476
0.110
0.101
0.393
7
0.235
0.436
0.078
0.071
0.353
8
0.207
0.400
0.056
0.052
0.318
9
0.185
0.369
0.041
0.040
0.290
10
0.166
0.341
0.032
0.029
0.263
11
0.150
0.316
0.024
0.022
0.239
12
0.137
0.292
0.020
0.017
0.218
13
0.125
0.270
0.016
0.013
0.199
14
0.114
0.250
0.013
0.010
0.183
15
0.104
0.230
0.010
0.008
0.168
16
0.094
0.211
0.007
0.006
0.154
17
0.086
0.193
0.006
0.005
0.140
18
0.079
0.177
0.005
0.004
0.128
19
0.072
0.163
0.004
0.003
0.117
Fraction of firms with leverage in 4, 3, or 2 different quartiles in different years:
4 quartiles
0.304
0.366
0.267
0.254
0.331
at least 3
0.695
0.632
0.741
0.764
0.646
at least 2
0.928
0.837
0.996
0.997
0.883
Fraction of firms currently in initial (t = 0) leverage quartile:
Full
Lowest
Low/Medium
Medium/High
Highest
sample
leverage
leverage
leverage
leverage
(6)
(7)
(8)
(9)
(10)
1.000
1.000
1.000
1.000
1.000
0.720
0.829
0.638
0.617
0.796
0.622
0.752
0.516
0.507
0.714
0.570
0.705
0.461
0.451
0.663
0.534
0.666
0.422
0.419
0.628
0.505
0.631
0.391
0.395
0.603
0.481
0.603
0.370
0.376
0.574
0.464
0.582
0.353
0.365
0.557
0.452
0.562
0.346
0.359
0.540
0.439
0.545
0.336
0.350
0.525
0.428
0.529
0.332
0.344
0.507
0.417
0.513
0.327
0.338
0.491
0.407
0.499
0.321
0.329
0.478
0.397
0.482
0.317
0.324
0.465
0.391
0.470
0.312
0.325
0.458
0.380
0.457
0.308
0.312
0.442
0.371
0.446
0.300
0.308
0.430
0.364
0.436
0.295
0.304
0.423
0.361
0.431
0.299
0.301
0.415
0.355
0.422
0.294
0.300
0.406
-------
-------
-------
-------
-------
Table 9
Length of Stable Leverage Regimes
The first row in each panel defines a stable leverage regime as one in which the firm’s Debt/Total Assets ratio continuously remains in a range of values that
differ by no more than 0.050. Each subsequent row in the same panel considers a successively broader (more lax) definition of a stable regime. The second row
in each panel defines a stable leverage regime as one in which the firm’s Debt/TA range continuously differs by no more than 0.100, while the third and fourth
rows in each panel define stable leverage regimes as instances in which Debt/TA continuously remains within ranges that differ by no more than 0.150 and 0.200
respectively. To generate the data in the table, we first take a given firm and identify its longest stable leverage regime (based on each given Debt/TA range
definition of a stable regime). For example, to generate the data in row A1, we take a firm that has been listed at least 20 years and calculate the longest number
of consecutive years that its Debt/TA ratio remained within a range of values that differ by no more than 0.050. We repeat this process for all firms in the
sample, and report the resulting histogram in row A1, with the sample median given in the far-right column. To generate the numbers in row A2, we follow the
same procedure but now use 0.100 in place of 0.050 to identify stable leverage regimes. We repeat this process for each remaining row. Since some firms in the
panel A sample are listed less than the number of years specified in the column headers, some table entries are specified “n.m.” (not meaningful).
A. Firms listed at least 20 years
A1. Debt/TA range ≤ 0.050
A2. Debt/TA range ≤ 0.100
A3. Debt/TA range ≤ 0.150
A4. Debt/TA range ≤ 0.200
B. Firms listed at least 40 years
B1. Debt/TA range ≤ 0.050
B2. Debt/TA range ≤ 0.100
B3. Debt/TA range ≤ 0.150
B4. Debt/TA range ≤ 0.200
C. Constant composition sample:
C1. Debt/TA range ≤ 0.050
C2. Debt/TA range ≤ 0.100
C3. Debt/TA range ≤ 0.150
C4. Debt/TA range ≤ 0.200
Percent of firms with Debt/TA continuously in specified range for at least:
10 years
20 years
30 years
40 years
21.3%
4.2%
n.m.
n.m.
50.3%
9.9%
n.m.
n.m.
73.6%
22.4%
n.m.
n.m.
85.7%
36.9%
n.m.
n.m.
Median # of years of
longest stable regime
6.0
10.0
13.0
17.0
32.0%
75.4%
93.3%
97.8%
6.6%
20.2%
45.8%
69.0%
2.6%
5.9%
14.5%
32.7%
0.7%
1.6%
3.8%
9.9%
8.0
13.0
18.5
24.0
51.6%
94.9%
100.0%
100.0%
7.6%
28.0%
68.2%
87.9%
2.5%
7.6%
24.2%
51.0%
0.0%
1.3%
6.4%
14.6%
10.0
16.0
22.0
30.0
Table 10
Stable Leverage Regimes and the Level of Leverage
For each firm, we identify the longest stable leverage regime (as defined below), with panel A analyzing stable regimes that last at least 20 years and panel B
analyzing those that last at least 10 years. The columns of the table sort firms according to the median value of the Debt/TA ratio during its longest stable
regime, and report the percentage of firms (in the sample for the row in question) that falls in each specified leverage interval. The first row in each panel defines
a stable leverage regime as one in which the firm’s Debt/Total Assets ratio continuously remains in a range of values that differ by no more than 0.050. Each
subsequent row in the same panel considers a successively broader (more lax) definition of a stable regime. The second row defines a stable leverage regime as
one in which the firm’s Debt/TA range continuously remains in a range of values that do not differ by more than 0.100, while the third and fourth rows define
stable regimes as situations in which Debt/TA continuously remains within a range of values that do not differ by more than 0.150 and 0.200 respectively.
Percent of firms with median Debt/TA during stable regime that falls in interval:
0.100 or less
0.100 to 0.200
0.200 to 0.300
0.300 to 0.400
0.400 or higher
A. Stable leverage regimes of 20 years or more
A1. Debt/TA range ≤ 0.050
A2. Debt/TA range ≤ 0.100
A3. Debt/TA range ≤ 0.150
A4. Debt/TA range ≤ 0.200
B. Stable leverage regimes of 10 years or more
B1. Debt/TA range ≤ 0.050
B2. Debt/TA range ≤ 0.100
B3. Debt/TA range ≤ 0.150
B4. Debt/TA range ≤ 0.200
Number
of firms
100.0%
78.8%
53.8%
42.9%
0.0%
7.3%
16.7%
21.5%
0.0%
11.0%
20.1%
22.6%
0.0%
1.8%
5.3%
9.3%
0.0%
1.1%
4.1%
3.8%
115
273
617
1,015
88.8%
62.2%
48.7%
42.7%
3.6%
11.5%
14.7%
16.5%
3.3%
12.9%
16.9%
17.9%
2.1%
7.2%
10.8%
12.4%
2.1%
6.2%
8.8%
10.5%
994
2,158
3,267
4,143
Table 11
Departures from Stable Leverage Regimes:
Asset Growth, Financing Deficits, and Traditional Leverage Determinants
The table presents the mean values of leverage, asset growth, and various other financial variables surrounding the longest
stable leverage regime for 945 firms listed 20 or more years on Compustat. For this analysis, a leverage regime is
considered stable if the firm’s Debt/Total Assets ratio takes values that differ by no more than 0.100 for 10 or more
consecutive years. For each firm, the last year of its stable regime is designated event year -1 so that event year 0 is the
year of its departure from stability, and all other event years over t = -3 to t = 3 are defined analogously. Asset growth
equals assets in event year t minus assets in year t-1, all divided by assets in t-1. The same divisor is applied to the year t
capital expenditures, financing deficit, change in debt, and EBITDA. For tangible assets in year t, we divide by total
assets in year t. The financing deficit measures the amount of external financing net of distributions in a given year and
equals the sum of net equity issues and net debt issues. [A negative financing deficit (i.e., a financing surplus) indicates
that, on net, the firm does not raise outside funds in the period under consideration.] We employ the change in total debt
outstanding as the measure of net debt issues to avoid sample size shrinkage because of missing values on Compustat of
the latter variable. For inclusion in this table, firms must be listed on Compustat through year t = 3 relative to its
departure from a stable leverage regime in year t = 0. All variables are Winsorized at the 1% level. We use ***to
identify a significant difference at the 0.001 level or better between the t = 0 mean value of a variable and its t = -1 value.
The variables in rows 6 to 9 show no significant differences at the 0.10 level.
Mean value of
1. Debt/Total Assets
2. Asset growth
3. Capital expenditures
4. Financing deficit
5. Change in debt
-3
0.162
0.091
0.076
0.017
0.016
6. EBITDA
7. Log (Sales)
8. Market-to-book
9. Tangible assets
0.172
5.576
1.533
0.349
Event year relative to departure in year 0 from stable leverage regime:
-2
-1
0
1
2
0.162
0.165
0.232***
0.231
0.236
0.091
0.100
0.204***
0.106
0.102
0.076
0.080
0.094***
0.078
0.075
0.015
0.022
0.114***
0.038
0.032
0.014
0.016
0.116***
0.024
0.032
0.171
5.661
1.550
0.349
0.167
5.719
1.503
0.353
0.166
5.846
1.436
0.356
0.157
5.936
1.424
0.356
0.156
6.011
1.417
0.353
3
0.236
0.097
0.074
0.024
0.023
0.154
6.083
1.402
0.351
Table 12
Departures from and Reversions to Stable Leverage Regimes
A leverage regime is considered stable if the firm’s Debt/Total Assets ratio takes values that differ by no more than 0.100 for 10 or more consecutive years. For
each firm that has a stable leverage regime in this sense, the table considers only its longest such regime. The table also restricts attention to firms that have been
listed on Compustat for at least 20 years and that have 10 years of non-missing data after the end of its stable leverage regime. There are 575 firms that meet
these sampling conditions. We find qualitatively identical results when we examine (i) the sample of firms with complete data through three years after the end
of their stable leverage regimes and (ii) our constant composition sample. For a given firm, the last year of its stable regime is designated event year t = -1 so that
t = 0 is the year of its departure from stability, and all other event years over are defined analogously relative to t = 0. Row 1 documents the percent of firms
whose Debt/TA ratios beginning at t = 0 remain within a range of values that does not exceed 0.100, i.e., that enter a new stable leverage regime (per the same
stability criterion described above). Row 3 reports the percent of firms with Debt/TA ratios that fall within the bounds of the earlier stable regime, i.e., this row
documents the extent to which leverage reverts back to the zone it consistently inhabited for at least 10 years prior to t = 0. The frequencies with which leverage
remains outside that earlier stable zone are reported in rows 2, 4, 5, and 6. Row 7 (row 9) reports the frequency that a firm’s dollar value of debt outstanding in
the event year in question exceeds (falls below) its value in year t = -1. Row 8 (row 10) reports the frequency that this debt amount does not exceed (does not
fall below) its t = -1 value.
Percent of firms in specified year
(relative to year 0 departure from stable leverage regime):
1
2
3
4
5
-1
0
---
100.0%
87.3%
73.6%
58.7%
46.9%
37.3%
5.6%
0.0%
100.0%
0.0%
70.9%
0.0%
29.1%
61.3%
15.9%
22.8%
60.7%
17.4%
21.9%
60.1%
20.7%
19.2%
60.1%
20.0%
19.9%
61.3%
20.6%
18.1%
60.0%
23.1%
17.0%
5. Debt/TA more than 0.05 above high end of earlier regime
6. Debt/TA more than 0.05 below low end of earlier regime
0.0%
0.0%
41.8%
5.6%
45.1%
9.4%
46.1%
10.6%
46.3%
11.8%
44.6%
11.5%
43.6%
12.0%
47.0%
11.9%
Debt/TA above high end of stable regime that ended at t = -1:
7. Increased borrowing
8. No increase in borrowing
-----
99.0%
1.0%
99.1%
0.9%
99.1%
0.9%
99.1%
0.9%
98.6%
1.4%
99.1%
0.9%
98.0%
2.0%
Debt/TA below low end of stable regime that ended at t = -1:
9. Debt paydown
10. No debt paydown
-----
86.2%
13.8%
80.9%
19.1%
65.9%
34.1%
70.0%
30.0%
54.4%
45.6%
47.1%
52.9%
43.3%
56.7%
Establishment of a new stable leverage regime:
1. Firms with range of Debt/TA ≤ 0.100 beginning at t = 0
Debt/TA relative to leverage during regime that ended at t = -1:
2. Debt/TA above high end of earlier stable leverage regime
3. Debt/TA within earlier stable leverage regime
4 Debt/TA below low end of earlier stable leverage regime
10
Table 13
Percent of Firms with Specified Combination of Maximum and Minimum Leverage Ratios
Leverage is measured as the ratio of total debt-to-total assets (Debt/TA) in book value terms. The table examines the sub-sample of 2,751 industrial firms that
were listed on Compustat for at least 20 years over 1950-2008. Rounding error explains the cases in which the percentages in the body of the table do not sum to
the category total.
Minimum Debt/TA:
1. 0.000
2. 0.000 to 0.100
3. 0.100 to 0.200
4. 0.200 to 0.300
5. 0.300 to 0.400
6. 0.400 to 0.500
7. 0.500 to 0.600
8. 0.600 to 0.700
9. 0.700 or higher
10. Column total
0.000
0.2%
0.000
to 0.100
1.9%
0.1%
0.100
to 0.200
2.9%
0.7%
0.0%
0.200
to 0.300
6.8%
4.4%
0.7%
0.0%
0.2%
2.0%
3.6%
11.8%
Maximum Debt/TA:
0.300
0.400
to 0.400
to 0.500
8.4%
8.1%
9.1%
7.6%
2.5%
4.5%
0.2%
1.2%
0.0%
0.0%
0.0%
20.3%
21.6%
0.500
to 0.600
5.3%
5.6%
2.9%
1.0%
0.5%
0.0%
0.0%
0.600
to 0.700
2.8%
3.2%
1.9%
1.2%
0.7%
0.0%
0.0%
0.0%
15.3%
9.7%
0.700 or
higher
5.7%
4.6%
2.8%
1.5%
0.5%
0.3%
0.1%
0.1%
0.0%
15.5%
Row total
42.2%
35.3%
15.3%
5.1%
1.7%
0.3%
0.1%
0.1%
0.0%
100.0%
Table 14
Leverage Peaks and Troughs:
Asset Growth, Financing Deficits, and Traditional Leverage Determinants
Panels A and B respectively present the mean values of leverage, asset growth, and various other financial variables
surrounding leverage peaks (the firm’s highest-ever Debt/Total Assets ratio) and troughs (its lowest Debt/TA ratio) for
the sub-sample of industrial firms listed at least 20 years on Compustat. In panel A, event year t = 0 is the calendar year
of peak leverage while t = 1 is the year immediately after the peak. In panel B, t = 0 and t = 1 are respectively the years
of and immediately after trough leverage. When a firm has multiple periods with the same peak or trough leverage, we
use the first such period here. All other event years over t = -3 to t = 3 are defined analogously. Asset growth equals
assets in event year t minus assets in year t-1, all divided by assets in t-1. The same divisor is applied to the year t capital
expenditures, financing deficit, change in debt, and EBITDA. For tangible assets in year t, we divide by total assets in
year t. The financing deficit measures the amount of external financing net of distributions in a given year and equals the
sum of net equity issues and net debt issues. [A negative financing deficit (i.e., a financing surplus) indicates that, on net,
the firm does not raise outside funds in the period under consideration.] We employ the change in total debt outstanding
as the measure of net debt issues to avoid sample size shrinkage because of missing values on Compustat of the latter
variable. For inclusion in this table, firms must be on Compustat from t = -3 through t = 3. There are 1,795 such firms in
the panel A analysis and 1,601 firms in the panel B analysis. All variables are Winsorized at the 1% level. In panel A,
***, **, and * respectively indicate a significant difference at the 0.001, 0.01, and 0.05 levels between the mean value of
a variable in the peak year (t = 0) and its value in the immediately prior year (t = -1). In panel B, ***, **, and *
respectively indicate a significant difference at the 0.001, 0.01, and 0.05 levels between the mean value of a variable in
the year after a trough (t = 1) and the year of the trough (t = 0).
A. Leverage peaks
Mean value of
A1. Debt/Total Assets
A2. Asset growth
A3. Capital expenditures
A4. Financing deficit
A5. Change in debt
Event year relative to leverage peaks (panel A) and leverage troughs (panel B) in year 0:
-3
-2
-1
0
1
2
3
0.269
0.296
0.348
0.475***
0.380
0.330
0.297
0.188
0.185
0.169
0.212**
0.045
0.071
0.100
0.089
0.095
0.099
0.090*
0.057
0.062
0.069
0.101
0.106
0.124
0.214***
-0.031
-0.001
0.014
0.073
0.082
0.105
0.194***
-0.068
-0.027
-0.005
A6. EBITDA
A7. Log (Sales)
A8. Market-to-book
A9. Tangible assets
0.149
4.708
1.671
0.332
0.141
4.794
1.616
0.337
0.124
4.911
1.568
0.344
0.106***
4.976
1.523
0.341
0.120
5.056
1.555
0.333
0.137
5.130
1.555
0.326
0.147
5.213
1.565
0.323
B1. Debt/Total Assets
B2. Asset growth
B3. Capital expenditures
B4. Financing deficit
B5. Change in debt
0.155
0.131
0.071
0.051
0.002
0.132
0.109
0.072
0.019
-0.007
0.101
0.117
0.071
0.021
-0.015
0.057
0.132
0.079
0.009
-0.033
0.162***
0.307***
0.114***
0.203***
0.169***
0.190
0.169
0.090
0.088
0.066
0.208
0.132
0.080
0.071
0.048
B6. EBITDA
B7. Log (Sales)
B8. Market-to-book
B9. Tangible assets
0.176
4.319
1.863
0.309
0.182
4.424
1.894
0.308
0.183
4.495
1.965
0.304
0.187
4.600
2.012
0.303
0.181
4.765*
1.825**
0.324**
0.153
4.886
1.711
0.328
0.148
4.978
1.617
0.328
B. Leverage troughs
Table 15
Time-series Variation in Leverage of 24 Major Industrial Firms
The last column gives a capsule summary of case details that are reported in the supplemental appendix. The table lists each firm by its most familiar name, with
alternative names provided in parentheses for clarity in some cases. Leverage is measured as the Debt/Total Assets ratio.
General Electric
General Motors
IBM
Procter & Gamble
Allied Chemical (Honeywell)
Union Carbide
Sears Roebuck
International Harvester (Navistar)
Caterpillar
B.F. Goodrich
Goodyear Tire & Rubber
Altria (Philip Morris)
American Tobacco (Fortune Brands)
Eastman Kodak
DuPont
ChevronTexaco (Standard Oil of CA)
Texaco
Exxon Mobil (Standard Oil of NJ)
AT&T
U.S. Steel
Bethlehem Steel
International Paper
Woolworth (Foot Locker)
Coca-Cola
Median across 24 firms
Debt/TA
range
0.670
0.634
0.396
0.395
0.356
0.464
0.608
0.751
0.547
0.442
0.431
0.502
0.480
0.433
0.317
0.375
0.320
0.197
0.375
0.469
0.400
0.532
0.318
0.323
0.432
Min
Debt/TA
0.000
0.000
0.022
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.003
0.000
0.000
0.000
0.000
0.000
0.113
0.018
0.055
0.000
0.002
0.000
0.000
Maximum annual:
Debt/TA Debt/TA
increase decrease
0.255
-0.159
0.213
-0.345
0.080
-0.093
0.115
-0.117
0.251
-0.081
0.231
-0.092
0.347
-0.312
0.215
-0.183
0.196
-0.154
0.292
-0.108
0.203
-0.203
0.271
-0.212
0.199
-0.157
0.205
-0.301
0.138
-0.094
0.292
-0.134
0.207
-0.094
0.078
-0.045
0.126
-0.131
0.196
-0.138
0.172
-0.137
0.203
-0.186
0.091
-0.081
0.116
-0.094
0.203
-0.136
Capsule summary of notable case features
Fund post-WW II expansion
Fund post-WW II expansion while paying substantial dividends
Fund post-WWII expansion; mostly passive deleveraging
Fund post-WW II expansion; passive deleveraging
Fund post-WW II expansion; then passive deleveraging
Fund post-WW II expansion; then passive deleveraging
Fund post-WW II expansion of installment-sales business
Fund post-WW II extension of credit to customers and dealers
Fund post-WW II plant expansion
Fund expansion in 1960s; then mostly passive deleveraging
Transitory borrowing to buy back stock and deter a hostile takeover
Fund expansion during WWII
Fund diversifying acquisitions in 1960s
Fund diversifying acquisitions in 1980s
Fund 1980s acquisition of Conoco
Fund 1980s acquisition of Gulf Oil
Fund 1980s acquisition of Getty Oil
Keep debt conservative while funding growth opportunities
Fund post-WW II expansion with equity to build flexibility
Presciently timed deleveraging prior to Great Depression
“Follow the leader” deleveraging prior to Great Depression
Distress-induced deleveraging, then levering up to fund investment
Proactive deleveraging after levering up amid financial trouble
CEO with aggressive approach to debt (including for mergers)
Table 16
Industry-median Leverage: Time-series Ranges and Standard Deviations
Industry-median leverage in a given year is the cross-firm median of the Debt/Total Assets ratios among all firms in that industry. The first three columns report
time-series ranges and standard deviations of the industry-median leverage ratios within each four-digit, three-digit, and two-digit Standard Industrial
Classification (SIC) code industry. The Fama and French industries in the other two columns represent groupings based on four-digit SIC codes:
http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html. We exclude financial firms and utilities. Our analyses of the Fama and French 49 and
12 industry groups are accordingly based on 44 and 10 industry groups. For Panels A and C, we first calculate for each industry the time-series range (over
1950-2008) in its cross-firm median Debt/TA ratio. Panel A reports the median across all industries of those time-series ranges. For Panels B and D, we first
calculate for each industry the time-series standard deviation (over 1950-2008) of its cross-firm median Debt/TA ratio. Panel B reports the median across all
industries of those standard deviations.
Four-digit SIC
Three-digit SIC
Two-digit SIC
Fama French 49
A. Cross-sectional median of the time-series ranges in industry-median leverage (Debt/TA):
0.414
0.394
0.319
0.291
B. Cross-sectional median of the time-series standard deviations ( of industry-median leverage (Debt/TA):
0.110
0.104
0.075
0.067
C. Range histogram: Percent of industries with a time-series range of industry-median Debt/TA in the specified interval:
0.000 to 0.100
5.7%
5.1%
2.7%
0.0%
0.100 to 0.200
7.4%
5.1%
8.1%
9.1%
0.200 to 0.300
14.1%
16.3%
35.1%
43.2%
0.300 to 0.400
20.5%
24.1%
24.3%
34.1%
0.400 to 0.500
18.8%
17.4%
10.8%
9.1%
Above 0.500
33.4%
32.1%
18.9%
4.6%
D. Standard deviation histogram: Percent of industries with time-series of industry-median Debt/TA in the specified interval:
0.000 to 0.050
9.5%
7.3%
14.9%
13.6%
0.050 to 0.100
33.0%
41.5%
54.1%
72.7%
0.100 to 0.150
33.5%
30.3%
21.6%
13.6%
0.150 to 0.200
15.1%
11.9%
5.4%
0.0%
0.200 to 0.250
6.1%
5.5%
4.1%
0.0%
Above 0.250
2.9%
3.4%
0.0%
0.0%
Fama French 12
0.230
0.058
0.0%
20.0%
80.0%
0.0%
0.0%
0.0%
40.0%
60.0%
0.0%
0.0%
0.0%
0.0%
Table 17
Industry-median Leverage: Panel Regressions and Variance Decompositions
Panels A and B respectively follow the templates of tables 6 and 7 and follow the statistical methods described therein. The dependent variables are the crosssectional median book leverage (Debt/TA) ratios within each industry for each year in the period 1950-2008. Industries are based on four-digit, three-digit, and
two-digit SIC codes and on the Fama and French partitions of the four-digit SIC universe into 49 industry and 12 industry groups (see Table 16).
A. Panel (ANOVA) regressions: Incremental explanatory power of industry fixed effects that vary by decade
Industry definition
Four-digit SIC
Three-digit SIC
Two-digit SIC
Fama French 49
Fama French 12
F-statistic
to compare
(1) versus (2)
7.95
10.11
15.26
20.20
34.79
Industry/decade
dummies
(1)
0.619
0.641
0.747
0.771
0.878
Adjusted-R2 for model with:
Industry
Industry
Year
dummies and
dummies
dummies
year dummies
(2)
(3)
(4)
0.352
0.040
0.385
0.330
0.064
0.384
0.428
0.072
0.506
0.389
0.139
0.544
0.517
0.160
0.737
Industry/decade
dummies and
year dummies
(5)
0.629
0.652
0.766
0.802
0.928
F-statistic
to compare
(4) versus (5)
7.50
9.07
13.31
15.60
27.83
B. Variance decompositions: Industry fixed effects, decade fixed effects, and industry/decade interaction effects
Four-digit SIC
1. Interaction-inclusive model
2. Purely additive model
Three-digit SIC
3. Interaction-inclusive model
4. Purely additive model
Two-digit SIC
5. Interaction-inclusive model
6. Purely additive model
Fama French 49
7. Interaction-inclusive model
8. Purely additive model
Fama French 12
9. Interaction-inclusive model
10. Purely additive model
Industry/decade
interaction effects
Percent of explained variation accounted for by:
Industry
fixed effects
Decade
fixed effects
45.3%
----
52.1%
94.4%
2.6%
5.6%
46.3%
----
48.1%
88.7%
5.7%
11.3%
36.3%
----
57.1%
87.5%
6.6%
12.5%
33.0%
----
51.2%
75.8%
15.8%
24.2%
19.5%
----
59.1%
73.2%
21.4%
26.8%
Appendix: Debt/Total Assets Ratios for 24 Dow Jones Industrial Average (DJIA) Firms
Firms are listed here in the same order as in Table 15, which is grouped by case type. Case details are in
the supplemental appendix. All 24 firms are members of our constant composition sample, which means
they are included on Compustat from 1950 to 2000. All 24 also (i) were publicly held prior to the Great
Depression, (ii) issued annual reports back to at least 1926 with clearly delineated financial debt amounts,
and (iii) were included in the Dow Jones Industrial Average (DJIA) at some point. For each firm, we
track leverage back to 1900 if possible, but more generally as far back as annual report disclosures clearly
separate financial debt from other liabilities (e.g., notes payable versus accounts payable). In cases in
which firms had major financial subsidiaries whose debt obligations in some years were not consolidated
with the parent, we obtain whatever financial data for the subsidiaries are provided in company
disclosures and report estimated leverage ratios based on our construction of the relevant consolidated
balance sheets. The latter firms are AT&T, Caterpillar, General Electric, General Motors, Goodrich,
Goodyear, IBM, Kodak, International Harvester (Navistar), Altria (Philip Morris), Sears Roebuck,
Texaco, and Union Carbide. Two firms have financial subsidiaries whose operations are too small to
merit disclosure (Coca-Cola) or the information that is disclosed is insufficient to estimate the leverage of
the consolidated entity (U.S. Steel).
The date of the first (and sometimes the last) observation differs across companies, and so one must be
careful in scanning across firms to be sure that one is comparing leverage in the same year. Since
leverage ranges vary substantially, the scale of the vertical axis also differs across firms.
0.700
0.600
0.700
General Electric
0.500
0.500
0.400
0.400
0.300
0.300
0.200
0.200
1900
1907
1914
1921
1928
1935
1942
1949
1956
1963
1970
1977
1984
1991
1998
2005
0.000
0.100
0.000
0.400
IBM
0.350
0.400
Debt/Total Assets
Debt/Total Assets
1911
1917
1923
1929
1935
1941
1947
1953
1959
1965
1971
1977
1983
1989
1995
2001
2007
Debt/Total Assets
0.100
0.500
General Motors
0.600
Procter & Gamble
0.300
0.250
0.300
0.200
0.200
Debt/Total Assets
0.150
0.100
0.100
0.050
0.000
1919
1924
1929
1934
1939
1944
1949
1954
1959
1964
1969
1974
1979
1984
1989
1994
1999
2004
1911
1917
1923
1929
1935
1941
1947
1953
1959
1965
1971
1977
1983
1989
1995
2001
2007
0.000
0.400
Allied Chemical (Honeywell)
0.500
0.350
Union Carbide
0.400
0.300
0.250
0.300
0.200
0.150
0.050
Debt/Total Assets
Debt/Total Assets
1920
1925
1930
1935
1940
1945
1950
1955
1960
1965
1970
1975
1980
1985
1990
1995
2000
2005
0.000
0.700
0.100
0.000
1925
1930
1935
1940
1945
1950
1955
1960
1965
1970
1975
1980
1985
1990
1995
2000
2005
0.100
0.200
0.800
Sears Roebuck
0.600
0.700
0.500
0.600
0.400
Intl Harvester (Navistar)
0.500
Debt/Total Assets
0.300
0.400
0.300
0.200
0.200
0.000
0.000
Debt/Total Assets
1913
1919
1925
1931
1937
1943
1949
1955
1961
1967
1973
1979
1985
1991
1997
2003
2009
0.100
1907
1912
1918
1924
1930
1936
1942
1948
1954
1960
1966
1972
1978
1984
1990
1996
2002
0.100
0.500
0.600
B.F. Goodrich
Caterpillar
0.500
0.400
0.400
0.300
0.300
0.200
0.200
Debt/Total Assets
0.100
0.100
Debt/Total Assets
1925
1930
1935
1940
1945
1950
1955
1960
1965
1970
1975
1980
1985
1990
1995
2000
2005
1912
1918
1924
1930
1936
1942
1948
1954
1960
1966
1972
1978
1984
1990
1996
2002
2008
0.000
0.000
0.500
Goodyear Tire & Rubber
0.600
Altria (Philip Morris)
0.500
0.400
0.400
0.300
0.300
0.200
Debt/Total Assets
Debt/Total Assets
0.100
0.000
1911
1917
1923
1929
1935
1941
1947
1953
1959
1965
1971
1977
1983
1989
1995
2001
2007
0.000
1920
1925
1930
1935
1940
1945
1950
1955
1960
1965
1970
1975
1980
1985
1990
1995
2000
2005
0.100
0.200
0.600
0.500
American Tobacco (Fortune Brands)
0.500
Eastman Kodak
0.400
0.400
0.300
0.300
0.200
Debt/Total Assets
0.200
0.100
Debt/Total Assets
0.000
1926
1931
1936
1941
1946
1951
1956
1961
1966
1971
1976
1981
1986
1991
1996
2001
2006
0.000
1902
1908
1914
1920
1926
1932
1938
1944
1950
1956
1962
1968
1974
1980
1986
1992
1998
2004
0.100
0.350
0.400
DuPont
0.300
0.350
0.250
0.300
0.200
0.250
0.200
0.150
Debt/Total Assets
0.050
0.100
0.050
1921
1927
1933
1939
1945
1951
1957
1963
1969
1975
1981
1987
1993
1999
2005
0.000
0.350
Texaco
0.000
0.250
Std Oil of NJ (Exxon Mobil)
0.300
0.250
Debt/Total Assets
1911
1917
1923
1929
1935
1941
1947
1953
1959
1965
1971
1977
1983
1989
1995
2001
2007
0.100
0.150
Std Oil of CA
(ChevronTexaco)
0.200
Debt/Total Assets
0.200
0.150
0.150
0.100
0.100
0.050
Debt/Total Assets
0.050
0.600
US Steel
AT&T
0.500
0.500
0.400
0.400
0.300
0.300
0.200
0.200
Debt/Total Assets
0.100
0.100
0.000
1908
1914
1920
1926
1932
1938
1944
1950
1956
1962
1968
1974
1980
1986
1992
1998
2004
0.000
-0.100
Debt/Total Assets
1901
1907
1913
1919
1925
1931
1937
1943
1949
1955
1961
1967
1973
1979
1985
1991
1997
2003
0.600
0.000
1918
1924
1930
1936
1942
1948
1954
1960
1966
1972
1978
1984
1990
1996
2002
2008
1912
1917
1921
1926
1931
1936
1941
1946
1951
1956
1961
1966
1971
1976
1981
1986
1991
1996
0.000
0.500
Bethlehem Steel
0.600
International Paper
0.500
0.400
0.400
0.300
0.300
0.200
0.200
1905
1911
1917
1923
1929
1935
1941
1947
1953
1959
1965
1971
1977
1983
1989
1995
2001
0.350
0.300
0.100
0.000
0.350
Woolworth
(Foot Locker)
0.300
0.250
0.250
0.200
0.200
0.150
0.150
0.100
0.100
0.050
Debt/Total Assets
Debt/Total Assets
1912
1918
1924
1930
1936
1942
1948
1954
1960
1966
1972
1978
1984
1990
1996
2002
2008
0.000
Coca-Cola
Debt/Total Assets
0.050
0.000
1919
1925
1931
1937
1943
1949
1955
1961
1967
1973
1979
1985
1991
1997
2003
0.000
Debt/Total Assets
1909
1914
1920
1926
1932
1938
1944
1950
1956
1962
1968
1974
1980
1986
1992
1998
2004
0.100
References
Baker, M., Wurgler, J., 2002. Market timing and capital structure. Journal of Finance 57, 1-32.
Bertrand, M., Schoar, A. 2003. Managing with style: The effect of managers on firm policies. Quarterly
Journal of Economics 118, 1169-1208.
Chang, X., Dasgupta, S., 2009. Target behavior and financing: How conclusive is the evidence? Journal
of Finance 64, 1767-1796.
DeAngelo, H., DeAngelo, L., Whited, T., 2011. Capital structure dynamics and transitory debt. Journal of
Financial Economics 99, 235-261.
Denis, D., McKeon, S., 2012. Debt financing and financial flexibility: Evidence from proactive leverage
increases. Forthcoming Review of Financial Studies.
Fama, E., French, K., 2002. Testing tradeoff and pecking order predictions about dividends and debt.
Review of Financial Studies 15, 1-33.
Fama, E., French. K., 2005. Financing decisions: Who issues stock? Journal of Financial Economics 76,
549-582.
Flannery, M., Rangan, K., 2006. Partial adjustment toward target capital structures. Journal of Financial
Economics 79, 469-506.
Frank, M., Goyal, V., 2003. Testing the pecking order theory of capital structure. Journal of Financial
Economics 67, 217-248.
Frank, M., Goyal., V., 2008. Trade-off and pecking order theories of debt. In Handbook of Empirical
Corporate Finance, Vol. 2, Eckbo, E., ed., Elsevier B.V.
Graham, J., Harvey, C., 2001. The theory and practice of corporate finance: Evidence from the field.
Journal of Financial Economics 60, 187-243.
Graham, J., Harvey, C., Puri, M., 2009. Managerial attitudes and corporate actions. Working paper, Duke
University.
Graham, J., Leary, M., 2011. A review of empirical capital structure research and directions for the future.
Forthcoming Annual Review of Financial Economics.
Graham, J., Leary, M., Roberts, M., 2011, A century of capital structure: The leveraging of corporate
America. Slides for July 28 Stanford Institute for Theoretical Economics (SITE) conference
presentation.http://www.stanford.edu/group/SITE/SITE_2011/2011_segment_3/2011_segment_3
_papers/GrahamLearyRoberts%20SITE%202011.pdf
Harford, J., Klasa, S., Walcott, N., 2008. Do firms have leverage targets? Evidence from acquisitions.
Journal of Financial Economics 93, 1-14.
Hennessy, C., Livdan, D., Miranda, B., 2010. Repeated signaling and firm dynamics. Review of Financial
Studies, 1982-2023.
Hennessy, C., Whited, T., 2005. Debt dynamics. Journal of Finance 55, 1129-1165.
Hovakimian, A., Li, G., 2011. In search of conclusive evidence: How to test for adjustment to target
capital structure. Journal of Corporate Finance 17, 33-44.
Huang, R., Ritter, J., 2009. Testing theories of capital structure and estimating the speed of adjustment.
Journal of Financial and Quantitative Analysis 44, 237-271.
Kayhan, A., Titman, S., 2007. Firms’ histories and their capital structures. Journal of Financial
Economics 83, 1-32.
Kim, E., Morse, A., Zingales, L., 2009. Are elite universities losing their competitive edge? Journal of
Financial Economics 93, 353-381.
Kisgen, D., 2006. Credit ratings and capital structure. Journal of Finance 61, 1035-1072.
Kisgen, D., 2009. Do firms target credit ratings or leverage levels? Journal of Financial and Quantitative
Analysis 44, 1323-1344.
Lambrecht, B., Myers, S., 2012. A Lintner model of payout and managerial rents. Forthcoming Journal
of Finance.
Lemmon, M., Roberts, M., Zender, J., 2008. Back to the beginning: Persistence and the cross-section of
corporate capital structure. Journal of Finance 63, 1575-1608.
Mackay, P., Phillips, G., 2005. How does industry affect firm financial structure? Review of Financial
Studies 18, 1433-1436.
Malmendier, U., Tate, G., Yan, J., 2011. Overconfidence and early-life experience: The effects of
managerial beliefs and corporate financial policies. Forthcoming Journal of Finance.
Mayer, C., Sussman, O., 2004. A new test of capital structure. Working paper. University of Oxford.
Miller, M., 1977. Debt and taxes. Journal of Finance 32, 261-276.
Minton, B., Wruck, K., 2001. Financial conservatism: Evidence on capital structure from low leverage
firms. Working paper. The Ohio State University.
Modigliani, F., Miller, M., 1958. The cost of capital, corporation finance, and the theory of investment.
American Economic Review 48, 261-197.
Myers, S., Majluf, N., 1984. Corporate financing and investment decisions when firms have information
that investors do not have. Journal of Financial Economics 13, 187-221.
Parsons, C., Titman, S., 2008. Empirical capital structure: A review. Foundations and Trends in Finance,
1-93.
Rajan, R., Zingales, L., 1995. What do we know about capital structure? Some evidence from
international data. Journal of Finance 50, 1421-1460.
Rauh, J., Sufi, A., 2011. Explaining corporate capital structure: Product markets, leases, and asset
similarity. Forthcoming Review of Finance.
Scheffé, H., 1959. The Analysis of Variance. John Wiley & Sons, New York, NY.
Shyam-Sunder, L., Myers, S., 1999. Testing static tradeoff against pecking order models of capital
structure. Journal of Financial Economics 51, 219-244.
Taggart, R., 1985. Secular patterns in the financing of U.S. Corporations. In Friedman, B., ed., Corporate
Capital Structures in the United States, University of Chicago Press.
Uysal, V., 2011. Deviation from target capital structure and acquisition choice. Journal of Financial
Economics 102, 602-620.
Welch, I., 2004. Capital structure and stock returns. Journal of Political Economy 112, 106-131.