Common Flaws in Empirical Capital Structure Research * Ivo Welch Brown University and NBER October 2, 2006 Abstract This paper critiques three issues that commonly arise in empirical capital structure research. 1. Capital Structure Proxies: The financial-debt-to-asset ratio is flawed as a measure of leverage, because the converse of financial debt is not equity. Depending on speci- fication, the debt-to-asset ratio can explain only about 10-50% of the variation in the equity-to-asset ratio. This is because most of the opposite of the financial-debt-to-asset ratio is the non-financial-liabilities -to-asset ratio. This problem is easy to remedy— researchers should use a debt-to-capital ratio or a liabilities-to-asset ratio. The converse of either is an equity ratio. 2. Non-linearity: The intrinsic non-linearity of leverage ratios can render standard linear regressions even with perfect independent variables seemingly powerless. Fortunately, researchers can easily test whether variables have a linear or non-linear influence on equity value changes, debt value changes, or leverage ratios. 3. Selection Issues: There are large survivorship biases in the CRSP/Compustat data bases. About 10% of firms appear and 10% disappear in a single year. These birth and death rates are themselves functions of capital structure and other firm characteristics. This selection makes studying long-term capital structure changes difficult. Unfortunately, this problem is difficult to remedy. The paper does not claim that these three issues drive results in the existing literature. It does however claim that they are not so small as to allow ignoring them a priori. The paper also clarifies some theoretical issues, most of which are not new, but which are sufficiently often muddled that a clarification is useful. First the paper distinguishes between capital structure mechanisms and causes. Second, when it comes to causes, it clarifies that there is no dichotomy between the pecking order theory and the trade-off theory. A pecking order arises in a trade-off theory in which issuing more junior securities is relatively more expensive, or possibly prohibitively expensive. A pecking order is not synonymous with adverse selection, financial slack, or a financing pyramid, either. This draft is early. Comments are welcome. * This paper arose out of an NBER discussion of Lemmon, Roberts, and Zender (2006). I thank Malcolm Baker, Long Chen, Michael Roberts, Sheridan Titman, and Jeffrey Wurgler for comments. 1
33
Embed
Common Flaws in Empirical Capital Structure Research∗
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Common Flaws in Empirical Capital Structure Research∗
Ivo Welch
Brown University and NBER
October 2, 2006
Abstract
This paper critiques three issues that commonly arise in empirical capital structure
research.
1. Capital Structure Proxies: The financial-debt-to-asset ratio is flawed as a measure
of leverage, because the converse of financial debt is not equity. Depending on speci-
fication, the debt-to-asset ratio can explain only about 10-50% of the variation in the
equity-to-asset ratio. This is because most of the opposite of the financial-debt-to-asset
ratio is the non-financial-liabilities-to-asset ratio. This problem is easy to remedy—
researchers should use a debt-to-capital ratio or a liabilities-to-asset ratio. The converse
of either is an equity ratio. 2. Non-linearity: The intrinsic non-linearity of leverage
ratios can render standard linear regressions even with perfect independent variables
seemingly powerless. Fortunately, researchers can easily test whether variables have a
linear or non-linear influence on equity value changes, debt value changes, or leverage
ratios. 3. Selection Issues: There are large survivorship biases in the CRSP/Compustat
data bases. About 10% of firms appear and 10% disappear in a single year. These
birth and death rates are themselves functions of capital structure and other firm
characteristics. This selection makes studying long-term capital structure changes
difficult. Unfortunately, this problem is difficult to remedy.
The paper does not claim that these three issues drive results in the existing literature.
It does however claim that they are not so small as to allow ignoring them a priori.
The paper also clarifies some theoretical issues, most of which are not new, but
which are sufficiently often muddled that a clarification is useful. First the paper
distinguishes between capital structure mechanisms and causes. Second, when it
comes to causes, it clarifies that there is no dichotomy between the pecking order
theory and the trade-off theory. A pecking order arises in a trade-off theory in which
issuing more junior securities is relatively more expensive, or possibly prohibitively
expensive. A pecking order is not synonymous with adverse selection, financial slack,
or a financing pyramid, either.
This draft is early. Comments are welcome.
∗This paper arose out of an NBER discussion of Lemmon, Roberts, and Zender (2006). I thank MalcolmBaker, Long Chen, Michael Roberts, Sheridan Titman, and Jeffrey Wurgler for comments.
1
The principal phenomenon that the capital structure literature tries to explain is variation
in corporate indebtedness. (Closely related literatures are the payout distribution and
repurchase literatures, and are often also grouped with the capital structure literature.) The
capital structure literature is interested both in the cross-section of capital structure—why
do some firms have high ratios today and others do not—and in the time-series—how do
capital structures evolve. Although the goal of the literature is straightforward, there are
many variations in the details. For example, different papers interpret different theories and
findings differently, seek to explain different variables, and use alternative specifications
This paper presents three simple but common pitfalls that empiricists should be aware of.
1. Capital Structure Proxies: The financial-debt-to-asset ratio is flawed as a measure of
leverage, because the converse of financial debt is not equity. Depending on spec-
ification, the debt-to-asset ratio can explain only about 10-50% of the variation in
the equity-to-asset ratio. This is because most of the opposite of the financial-debt-
to-asset ratio is the non-financial-liabilities-to-asset ratio. This problem is easy to
remedy—researchers should use a debt-to-capital ratio or a liabilities-to-asset ratio.
The converse of either is an equity ratio.
2. Non-linearity: The intrinsic non-linearity of leverage ratios can render standard linear
regressions even with perfect independent variables seemingly powerless. Fortunately,
researchers can easily test whether variables have a linear or non-linear influence on
equity value changes, debt value changes, or leverage ratios.
Some variables are particularly prone to exert linear influences on leverage ratio
constituents. For example, (retained) earnings and depreciation tend to influence the
book values of equity, market price changes tend to influence market values of equity.
Interest rate changes and payments tend to influence the value of debt. It is a priori
unclear whether the influence of these variables (and variables strongly correlated
with them) is linear or non-linear.
3. Selection Issues: There are large survivorship biases in the CRSP/Compustat data bases.
About 10% of firms appear and 10% disappear in a single year. These birth and death
rates are themselves functions of capital structure and other firm characteristics. This
selection makes studying long-term capital structure changes difficult. Unfortunately,
this problem is difficult to remedy.
This paper does not show that these issues are responsible for results in the existing
literature. Indeed, it could be that if research designs take them into account, previously
2
reported empirical regularities would improve in significance. (They could also pull in
different directions, and by happenstance cancel one another.) However, we do not know
this a priori. The paper does show that these three issues are important. It is unsettling
that the empirical literature today relies primarily on capital structure studies that suffer
from one or more of these three issues. Thus, it would not only be desirable for future
studies taking these issues into account, but also to confirm the results of earlier studies.1
The paper also makes some other empirical specification and theoretical interpretation
observations. It distinguishes between mechanisms and deeper causes. The two theories
most prominently perceived as deeper explanations are the pecking order theory and the
trade-off theory. Although most of the insights themselves are not novel, they are sufficiently
frequently left ambiguous to make it worthwhile to lay them out clearly. The main point is
that a pecking order arises in a trade-off theory, in which issuing more junior securities
is relatively more expensive, or possibly prohibitively expensive. The pecking-order and
trade-off theories are not converse, but (normally) facets of the same theory. It is also
important to recognize that the pecking order is not synonymous with adverse selection,
financial slack, or a financing pyramid. Finally, this paper offers a speculative view on where
future progress in our understanding of capital structure could come from.
I Issues With the Financial Debt To Asset Ratios
Although other leverage measures are also in common use, the single most common debt-
ratio variable in this literature is financial debt (often the sum of long-term debt and debt
in current liabilities, Compustat #9 plus #34) divided by assets (Compustat #6). The assets
are usually quoted in book value, though sometimes translated into market value. This is
accomplished by subtracting off the book value of equity and adding back the market value
of equity.2 The authors’ normal desired interpretation is that this financial debt-to-asset
ratio is a measure of leverage, the converse of which are presumably more junior financial
1To a referee (not for publication): I originally considered replicating some earlier studies. Ultimately,I decided this would be counterproductive. I do not wish to “pick on” one particular study. Showing thatthese three issues are problems for one study would say little about whether or not they are problems forother studies. Thus, I believe it is more productive to instead quantify how important the three issues are, asmy current paper does.
2The majority of paper in this literature use such a debt-to-asset ratio. This includes such classic papersas, for example, Rajan and Zingales (1995), Shyam-Sunder and Myers (1999), Baker and Wurgler (2002), andGraham (2003). Note that these papers use debt-to-asset ratios at least in some of their specifications, butthey often do entertain other measures, too. (The reader can easily confirm from current finance journalsthat the debt-to-asset ratio remains the most common dependent variable in this literature today.)
3
securities, such as an equity-to-asset ratio. Increases in debt-to-asset ratios are presumably
leverage increases, and equivalent to decreases in the equity-to-asset ratio.
However, this interpretation is flawed, because the opposite of financial debt is not equity.
Instead, the opposite of financial debt are non-financial liabilities plus equity.
Because the LHS dependent variable is an unambiguous ratio measure of leverage (equity is
junior to both financial and non-financial liabilities), these regressions can be interpreted
as relating the commonly-used leverage ratio (RHS) to the correct leverage ratio. If γ1 ≈ 1,
then firms do not substitute other liabilities for financial debt on average. If the R2 of the
regression is close to 100%, then the omission of non-financial liabilities is inconsequential.
(It is the R2s of non-financial liabilities plus liabilities that must add up to 1 in explaining
equity ratios, not the correlations, so the reader should think in terms of R2s.) In levels,
firms with higher debt-to-asset ratios can then be viewed as more levered. In changes,
increases in the financial-debt-to-asset ratio can then be viewed as moving the firm towards
a capital structure with more senior and fewer junior claims.
The regression results in Table 2 convey a similar message as the means in Table 1. In
the cross-sectional regressions, among the largest 500 firms (which comprise most of the
5
Table 1: Means of Financial Asset-Normalized Financial Debt, Non-Financial Liabilities, andEquities
Market or Minimum MeansBook Value Inclusion D/A NFL/A E/A N
Average Year-by-Year Mean BV 8 27% 48% 25% 505Pooled Mean 27% 47% 26% 11,625
Average Year-by-Year Mean MV 8 23% 42% 35% 505Pooled Mean 34% 43% 23% 11,625
Average Year-by-Year Mean BV 0.2 23% 42% 35% 2,905Pooled Mean 0.2 34% 43% 23% 66,826
Explanation: D/A are financial liabilities (Compustat D = #9 + #34) divided by assets (A = #6). Marketvalues subtract the book value of equity (E = #60) and add back the market value (#199·#54). NFL are theremaining non-financial liabilities. All ratios were truncated at 0 and 1.For a firm-year to be included, it had to have had market and book levels of assets greater than x times theS&P500 level the year prior to inclusion, where x is either 8 (for about 500 firms) or 0.2 (for about 3,000firms). For example, in the top four rows, a firm-year was included in 2000 if its book value of assets was atleast 8 · $1.469 = $11,752 billion at the end of 1999, because the S&P in 1999 finished at 1,469.Interpretation: Non-financial liabilities are not inconsequentially small.
market’s capitalization, and thus would dominate any value-weighted study), there is little
correlation between financial-debt-to-asset and equity-to-asset ratios in levels. In book
values, the R2 is below 5%; in market values, it is around 10%. It follows that most of
the cross-sectional heterogeneity in the true leverage ratio (i.e., one minus the dependent
variable, the equity-to-asset ratio) is due to non-financial liabilities. With explanatory power
this low, any regression among these firms using the debt-to-asset ratio suffers from a pure
error problem, more than it suffers from an error-in-variables problem. One cannot simply
equate large firms with high financial-debt-to-asset ratios as firms that are highly levered.
In a broader set of 3,000 firms, depending on whether a Fama-MacBeth type or pooled
regression is used, around 34% to 43% of the level heterogeneity in leverage ratios is
due to variation in financial debt-to-asset ratios. Thus, two thirds of the cross-sectional
heterogeneity in equity-to-asset ratios comes from the non-financial liabilities. Non-financial
liabilities have a stronger influence on true leverage than the financial debt variable that is
so commonly used.
In changes, among the largest 500 firms, less than one third of the variation in equity-
to-asset ratio changes is explained by financial-debt-to-asset ratio changes. Therefore,
about three-quarters of the variation in the true leverage ratio changes must come from
6
Table 2: Financial Debt-To-Asset and Financial Equity-To-Asset Ratios
In Levels E/A = γ0 + γ1 ·D/A In Differences ∆E/A = γ0 + γ1 ·∆D/A
500 Firms (V Factor 8), Market Values of Equity and Assets
Coef γ1 R2 NFM −0.43 0.10 505
Pooled −0.42 0.08 11,625
Coef γ1 R2 NFM −0.71 0.39 505
Pooled −0.75 0.41 11,625
Explanation: The regression explains the equity-to-asset ratio (book-value in the top two panels, market-valuein the bottom panel) with the corresponding financial debt-to-asset ratio. Inclusion criteria and truncationsare explained in Table 1. FM (“Fama-Macbeth”) is the mean of the annual coefficients.Interpretation: A large part of the variation in equity-to-asset ratios comes from changes in the non-financialliabilities-asset ratio, and not from changes in the financial-debt-to-asset ratio. Put differently, a large part ofchanges in financial leverage is picked up (undone) by changes in non-financial liabilities.
7
non-financial liability ratio changes. In other words, financial-debt-to-asset ratio changes
are not even the primary driver of leverage ratio changes, non-financial debt-to-asset ratio
changes are. In a broader subset of 3,000 firms, the relation between debt-to-asset and
equity-to-asset ratios becomes stronger, but it is not overwhelming, either. Non-financial
liability ratio changes continue to explain about half of the equity-to-asset (true leverage
ratio) changes.
It is not necessarily comforting that the correlation between true and used leverage ratios
is higher when more firms are included. (Most existing studies would include a sample
akin to the broader set.) Aside from the fact that the explanatory power still often remains
below 50%, the fact that different firms have different relationships between financial debt
and equity on the one hand, and non-financial liabilities on the other hand, further advises
caution: The proxy quality is not just simply noisy, but noisier for some specific firms with
specific attributes (e.g., size) than it is for others. This will bias the coefficients.
Broader Implications: In existing studies, the use of a financial-debt-to-asset ratio may
have just added noise. For example, if a factor that influences the financial debt-to-asset
ratio does not influence the non-financial liabilities-to-asset ratio, especially if does not do
so differentially across different types of firms, then the results would remain the same. The
regressions would merely be less powerful. Nevertheless, we cannot know which existing
results in the literature, if any, are sensitive to the leverage definition. Studies having used
financial debt-to-asset ratios may require reexamination and confirmation. Third variables
hypothesized to drive the division between debt and equity could instead merely influence
the division of the firm’s liabilities between financial and non-financial ones.
Fortunately, this problem is easy to correct. It is sensible to offer as a prescription for future
research to avoid the financial-debt-to-asset ratio. Instead, researchers should entertain
only either
1. financial debt divided by financial capital (the sum of financial debt plus financial
equity);
2. or total liabilities divided by total assets. (Other liabilities can include pension obliga-
tions, payables, etc., because they create an asset at the same time, just as issuing
activity does.)
Of course, these measures have different economic meaning, but they are internally consis-
tent, in the sense that a higher measure implies a higher leverage. The reason is that the
8
converse of either indebtedness ratio is an equity ratio.3 A third alternative that would sort
non-financial liabilities into those that are of higher priority and those that are of lower
priority than financial debt seems generally not feasible. A fourth alternative would be use
a measure based on flows, such as interest coverage, as a measure of debt burden. However,
because operating cash flows can be negative and because they are strongly business-cycle
dependent, and because interest payments can be zero (and some firms even report negative
interest payments), such a measure is not easy to use, either.
As a final note, the problem that leverage is often misdefined as a financial-debt-to-asset
ratio is not exclusive to the capital structure literature. The debt-to-asset ratio also often
appears as an independent variable (rather than as the dependent variable) in other financial
literatures. In this case, such a debt-to-asset definition can cause a potentially equally
serious error-in-variables problem. The measure is unlikely to measure well what the
researcher originally had in mind.
II Issues with the Econometric Specification:
Non-Linearity and Changes in Leverage Ratios
Heterogeneity in capital ratios among different industries and types of firms is a well-
known empirical regularity. For example, Table 5 below shows that disproportionally many
technology firms have no debt. This section shows that this heterogeneity complicates
translating value changes into capital structure effects. The reason is that debt ratios are
highly non-linear in their components, debt and equity. For example, consider the effect of
a firm that doubles its equity value (either through net equity issuing or through a value
change). If the firm was originally financed 50-50 by debt and equity, it will now be financed
33-67 by debt and equity. This firm would experience a decline in its debt-equity ratio. If
it was financed 0-100 by debt and equity instead, it will experience no change in its debt
ratio.
3I would also claim that the common use of book values rather than market values is a mistake—butwhich to use is already a well-known controversy. I know of no literature, other than the capital structureliterature, in which researchers prefer a book value to a readily available market value. There are systematiccross-sectional differences (e.g., in age and size) in how book-values relate to market-values. In this paper, Ihave used the book value only because I wanted to clarify that the issues critiqued are not due to the use ofmarket-value based measures of leverage.
9
A More Counterintuitive Illustration
Figure 2 illustrates yet another less intuitive aspect of the non-linearity. The firm’s base
capital structure is fixed at an equity value of $10 at the outset, and a debt value that is
indicated by the value on the right of each graphed function. In the figure, there is no
change in debt value, and there are zero other liabilities. The independent variable is the
size of the equity value change (e.g., caused by an equity issue or a stock return). The
dependent variable is the change in the equity/capital ratio (CER), defined as
∆CERt−1,t ≡ Et +∆Et−1,t
Dt + Et +∆Et−1,t− EtDt + Et
. (2)
Of course, the more common change in the capital leverage ratio (CLR) is ∆CLRt−1,t ≡1 − ∆CERt−1,t . The figure shows that for firms which start out with no debt, an equity
value change causes no leverage ratio change. The equity value change is progressively
more effective in changing the capital structure ratio until the original debt reaches about
D0 =√(E0 +∆E) · E0 = $24—and then the effect of an equity value change on leverage
ratio changes declines again.
B The Standard Linear Research Design
This paper can illustrate how the non-linearity can matter in an empirical linear-estimation
context. The empirical research hypothesis is that changes in debt and equity value change
leverage-to-capital ratios. Of course, this relationship between these variables is a tautology—
to change leverage, it is precisely and only changes in debt and equity values that can
matter.
The specific research design now entertained will however use a more naive approach. The
goal is to assess the quality of this standard linear model specification in the context of
these perfect independent variables.
1. Define the financial capital as the sum of debt and equity,
Ct ≡ Et +Dt (3)
where E is the firm’s book equity (Compustat #60), and D is the sum of book financial
debt (long-term debt [Compustat #9] and debt in current liabilities [Compustat #34]).
To keep the variance low, this uses the book value of equity rather than the market
value of equity.
10
Figure 2: Non-Linearity
0 10 20 30 40 50
0.0
0.1
0.2
0.3
0.4
0.5
Change in Equity Value
Ch
an
ge i
n E
qu
ity/A
sset
Rati
o
D0=$0
D0=$1
D0=$2
D0=$8
D0=$25
D0=$50
D0=$150
D0=$200
E0=$10
This figure plots (E0+∆E)/(D0+E0+∆E) − E0/(D0+E0). The original equity value, E0 = $10. The originaldebt value, D0, is indicated on the right. The total equity value change, ∆E, is plotted on the y-axis.
2. Compute the change in the capital leverage ratio (CLR),
∆CLRt+1 ≡ Dt+1
Ct+1− Dt
Ct. (4)
Except for the fact that the denominator is financial capital rather than total assets,
this variable is the most common dependent variable in the capital structure literature.
3. Compute the equity change (including issuing and value changes), normalized by the
start-of-period capital,
∆Et+1 ≡Et+1 − Et
Ct, (5)
and do the same for debt,
∆Dt+1 ≡Dt+1 −Dt
Ct. (6)
As noted, these equity and debt change variables rely on ex-post knowledge of all
changes in value. Thus, they are the best independent variables that an empiricist
could ever hope to have. In a non-linear fashion, together with the starting capital
11
structure, these two changes can fully explain the change in the capital leverage ratio,
because
∆CLRt+1 = Dt +∆Dt+1
Dt +∆Dt+1 + Et +∆Et+1− DtDt + Et
. (7)
4. Instead of estimating the non-linear tautology, we now estimate the relationship
between ∆CLRt+1 as the dependent variable and equity changes ∆Et+1 and debt
changes ∆Dt+1 as independent variables using a linear regression:
∆CLRt = γ0 + γ1 ·∆Dt + γ2 ·∆Et + Noiset . (8)
Such a “linearized reduced form” is the most common method of estimation in the
literature. The “linearized theory” would suggest a strong positive γ1 coefficient on
∆Dt+1 (debt value increases should be associated with increases in leverage ratios), and
a strong negative γ2 coefficient on ∆Et+1 (equity value increases should be associated
with decreases in leverage ratios). More importantly, such a study would likely judge
the meaning of debt and equity value changes as explanators of leverage ratios by the
R2 of this regression.
I obtained all Compustat debt and equity values for the last two years to which I had easy
data access, 2002 and 2003.4 The empirical estimate of regression (8) is
Thus, even a researcher who was “fortunate” enough to have truncated this way would still
conclude that the explanatory power of the regression remains low. Intuitively, 75% of the
4Brown University has not yet subscribed to Compustat, so I could not update the regression. However,this is a nuisance and not important for the substance of the illustration.
5Unreported, many of the mutual correlations are very sensitive to the truncation employed. For example,there is strong mean reversion without the truncation, but the correlation between lagged capital structureand changes falls dramatically if capital ratio differences are truncated at −100% and +100%. If the regressionreported in the text is repeated with an alternative truncation rule of −300% and +300%, the regression R2
falls to 8%.
12
change in leverage ratios remains unexplained—even though this is a regression whose
variables embody perfect and complete foresight.
In sum, both regressions illustrate that a low R2 in a linear regression predicting leverage
changes need not be due to poor independent variables. Instead, it can be due to the linear
prediction of a non-linear dependent variable.6
C Rebalancing?
It was pointed out by Welch (2004, p.112) that changes in debt or equity value, and changes in
debt ratio are very different, though only in passing. This can matter in tests of rebalancing
especially in the cross-section. Some studies have explored issuing activity (one part of
value changes) as the dependent variable, and a measure of deviation from a target debt
ratio as the independent variable. These studies are particularly vulnerable to non-linearity
concerns. Chen and Zhao (2006) point out the simple yet convincing fact that a firm that is
90-10 debt-equity financed and which issues four times as much debt as equity (80-20) is
still rebalancing towards equity, not towards debt! It follows that it is not enough to show
that certain firms, which should have more equity, tend to issue more equity than debt in
order to demonstrate that they are in fact rebalancing. Chen and Zhao show empirically
that seemingly natural inference on other variables (specifically on a target-debt ratio) can
reverse: on average, a third variable can relate positively to net debt issuing activity and/or
negatively to net equity issuing, and yet relate negatively to debt ratio changes.
D A Functional Alternative
The solution to this problem would be easy if capital structure theory did provide clear
functional specifications of how third variables influence debt ratios, debt levels, and equity
levels. It does not. Some variables are likely to influence the ratio directly. Other variables
may have primarily a linear influence on debt and equity value changes. The book value of
equity essentially increases with retained earnings and equity issuing activity, decreases
with depreciation and repurchasing and dividend payout activity. Thus, these variables may
be good candidates for a linear influence specification. (If equity is measured in terms of
market value, then stock returns replace retained earnings minus depreciation.) Similarly,
6Of course, in ordinary capital structure studies, the researcher usually does not have access to variablesas powerful as the advance-knowledge ∆D and ∆E. Instead, the research usually specifies some lagged firmvariables (attributes), which in turn may influence either equity value changes, or debt value changes, or both.
13
collateral or interest and principal payments may have a linear influence on the book
value of debt. The same applies in turn to variables that have a linear influence on these
components.
The non-linearity issue is harder for a researcher to properly address than the debt-to-asset
ratio definition explained in Section I. The dependent variable must remain a leverage ratio,
because it is the variable of interest in this literature. However, there is an alternative
diagnostic and modeling option. When the empirical relations can work linearly through debt
and equity value changes, a multiple equation system can explore whether hypothesized
variables predict empirically either debt or equity year-to-year value changes, or whether
they predict the leverage ratio directly.
(D
D + E
)= t0 + t1 ·
(D
D + E
)+ t2 · x1 + t2 · x2 + · · · + noise
D = d0 + d1 · x1 + d2 · x2 + · · · + noise
E = e0 + e1 · x1 + e2 · x2 + · · · + noise
Such a specification can indicate which x variables are better modeled as linear components
of debt and equity, and which variables are better modeled as direct linear calibrators of the
non-linear leverage ratio. The correlation of the top equation’s hatted (predicted) capital
ratio and the observed capital ratio could also be used as a measure of forecasting success
for the lower two equations. (This is similar to the procedure in Welch (2004)).)
III Issues with Appearance and Disappearance of Firms
(Survivorship)
There is a general consensus that simple trade-off readjustment theories cannot explain
capital structure behavior over annual horizons. The research has thus begun to shift
towards measuring how strong and timely capital structure readjustment is. Inevitably,
the focus then is shifting to explaining capital structure changes over longer horizons.
Among papers that attempt this are Welch (2004) and Kayhan and Titman (2006) over 5-year
horizons, and Lemmon, Roberts, and Zender (2006) since firm inception.
14
A Average Selection Bias
Figure 3: Firms First Appearing in 1980 on Compustat/CRSP Merged Tape
1980 1985 1990 1995 2000
10
20
30
40
50
Last Compustat Year
Nu
mb
er
of
IPER
MS
331 IPERMs starting in 1980
Average Existence: 10 years, truncated
15%
sti
ll a
live i
n 2
003
Most research, including my own, has assumed that the population of firms remains
fairly constant, i.e. that selection (survivorship bias) of firms is not an issue of first-order
importance. Figures 3 and 4 describe the persistence of PERMNOs on Compustat/CRSP,
which proves otherwise. CRSP changes PERMNOs at mergers or deaths, which is the right
experiment in this context: acquisitions, delisting, and bankruptcy result in fundamental
capital structure change, which would then not be picked up by simple regressions that
rely on consecutive firm observations that are on the tapes for multiple years. If anything,
for the study of capital structure, this experiment may still be too conservative. Compustat
excludes many volatile firms that drop in and out of their annual selection criteria.
Figure 3 shows how many PERMNOs on Compustat that initiated in 1980 (331 firms) have
died in each year, and how many remain in the final year, 2003. The figure shows that
the disappearance rate is not trivial. After 23 years, only 15% of all firms remain. The
average length of existence of firms that appeared in 1980 was 10 years. Figure 4 shows
the converse: how long individual PERMNOs present in 2003 have been alive. 20% of all
firms have been on the tapes for fewer than 5 years. Only half of all firms remain after ten
15
Figure 4: Firms Present on Compustat/CRSP Merged Tape in 2003
First Compustat Year
Nu
mb
er
of
IPER
MS
01
00
20
03
00
40
05
00
60
07
00
7,732 IPERMs of 2003Average Age: 10 years
0.0
0.2
0.4
0.6
0.8
1.0
Cu
mu
lati
ve P
ct
of
Fir
ms E
arl
ier
Th
an
Year
X
1950 1960 1970 1980 1990 2000
years. In sum, this data suggests that survivorship bias is a first-order concern for studies
that explore capital structure over longer horizons.7
B Selection By Characteristic
The problem is even more difficult if birth and survival is itself related to the variables being
studied, such as the dependent variable (some form of debt ratio) or some independent
variables (such as firm size). Table 3 shows how different characteristics influence the
probability that a particular firm-year is the last or first year for this firm on the merged
Compustat-CRSP tapes. Firms which appear on Compustat seem disproportionally more
fragile at the outset, and probably tend to die sooner. The birth and death rates are
differentially high enough even on an annual basis to raise concern whether selection issues
can distort inference in capital structure work. Unfortunately, the issue of survivorship
bias and sample selection is easier to diagnose than it is to correct. Depending on the
specific study and inference sought, the solution must lie in modeling what firms are likely
7This evidence may explain why the average firm alive today has a reasonably high debt ratio, even thoughthe stock market has increased dramatically over the last 50 years. That is, it may not require a specifictrade-off theory to explain this pattern.
16
Table 3: Annual Probability of Appearance and Disappearance on Compustat/CRSP
Explanation: Each cell reports frequencies for between 42,420 and 42,480 firm-years. To be included, thefirm-year had to have a value for assets, and an PERMNO on merged CRSP/Compustat tape. (Note thatCRSP assigns new PERMNOs after big mergers.) About 9.5% of all firm-year observations on Compustat arefirst year observations, and about 9.5% are last-year observations. The upper tabular is based on first-yearfirm-years, the lower on last-year firm-years. The Liability ratios are based on E/D sorts, and reported inreverse order. (A D/E based variable or sort is non-sensible, because book equity values can be negative.)Interpretation: Birth and death rate relate systematically both to dependent and independent variables inthe capital structure literature.
to have appeared and disappeared. This can be done, e.g., by using a “worst-case scenario”
to establish bounds, or by using a first-stage model (e.g., a Heckman procedure) to adjust
coefficients.
IV Theory Background of Empirical Capital Structure Tests
The remainder of the paper clarifies some issues in the theory that serves as background to
this literature. It is useful to distinguish between “mechanisms” and “causes” of capital
structure change, although they are really issues on a spectrum:
• At a relatively shallow level of causality (which can be called mechanisms), one can be
interested in how firms have gotten to where they are. This is a question of whether
17
today’s capital structure has come about via debt issues, via debt repurchases, via
equity issues, via stock returns, via shares issued during M&A transactions, etc.
• At a deeper level of causality, one can be interested in why firms have gotten to where
they are. This is a question of whether firms did so because they wanted to save on
taxes, avoid issuing equity, retain financial flexibility, time the market, etc.
The “how” question is informative about the “why” question, but not fully so. For example,
acquirers may issue debt and/or equity, but knowing which payment method they chose
does not tell us why they used one form of payment over one another, much less why
they acquire another firm. On the other hand, a theory that states that firms do not issue
equity because they are afraid of the negative inference of the market (either about current
projects or future use of cash) could have specific implications about the “how.”
The shallower level is easier to research than the deep level. Nevertheless, remarkably little
is known here, too. Fama and French (2005) describe equity changes in Fortune-100 firms,
and find that an astonishingly large fraction (3.68% out of 3.77%) appears in the context of
M&A activity. Another 1.05% fraction appears in the context of employee compensation,
obtained primarily through repurchases. Simple equity issuing activity without an M&A
background represents only 0.09% out of this 3.77%. However, FF have little to say either
about debt dynamics or about how equity changes in smaller firms. Welch (2004) offers
a non-linear variance decomposition of mechanisms that shows that capital structure is
greatly influenced by long-term debt net issuing activity and stock returns. However, Welch
offers no distinction between the context in which these equity appears, or between issuing
and repurchasing activity. Both papers find that firms experience active and dynamic capital
structure changes year-to-year.
Most of the focus of the literature has been on exploring deeper causes, and two in
particular:8 the trade-off theory (TO) and the pecking order theory (PO). The main point of
this section is to outline that these two theories are not mutually exclusive, but merely that
they highlight two different empirical predictions of the same overall theoretical framework.
Again, there is no claim that this part of the paper offers anything new. The insights can be
found in the set of earlier papers. They are interesting only because there are also a good
number of papers, in which the distinctions that I believe to be important are either blurred
or outright misinterpreted.
8One paper cannot adequately describe the nuances of capital structure theories—there is a plethora ofmodels that offer rich sets of implications (e.g., Harris and Raviv (1991)). Many of these papers are interestedonly in a particular aspect of capital structure, or in some information-theoretic phenomenon instead. Theseinformation theories are not the focus of my paper here.
18
A The Trade-Off Theory
The trade-off theory is prominently associated with work by Merton Miller, Sheridan Titman,
Ronald Masulis, and others. It merely posits that firms desire to trade off the costs and
benefits of debt and equity (plus debt and equity adjustments), but leaves the identification
of the factors for later.
The most common versions of the trade-off theory identify the benefit of debt as tax savings,
and the cost as financial distress. A number of papers have attempted to calibrate such
friction-free static models, perhaps none so more prominently than Graham (2000). Other
papers (e.g., Almeida and Philippon (2006)) have critiqued his specific estimates as too
naïve, either because of assumptions in Graham’s estimates of the tax benefits or in the
estimates of financial distress costs.
One important extension of the static model is the degree to which frictions prevent firms
that want to follow the trade-off theory from doing so. A number of papers (e.g., Strebulaev
(2003)) have shown that the objective function can be so flat that deviations from the
optimum barely matter, especially in the presence of transaction costs. Even modest
capital structure adjustment costs can then induce firms not to readjust. Frequently, these
papers then declare victory for the trade-off theory—that the trade-off theory can explain
the empirical evidence. Yet this is a phyrric victory. An alternative and equally valid
interpretation of a flat objective function with adjustment costs is that the trade-off theory
is not useful. If the costs of adjusting are so high and the benefits of adjusting so low that
“anything may go,” then the trade-off theory is practically irrelevant. In some sense, a flat
objective function could even be seen as the NULL against which a tradeoff theory is tested.
An even more general version of the trade-off theory allows the target to be dynamic or
at least different across firms. For example, the trade-off adjustment models in Fischer,
Heinkel, and Zechner (1989), Leland (1998), Leland and Toft (1996), and Hennessy and
Whited (2004) are sophisticated attempts to derive an optimal dynamic capital structure.
These models’ advantages is that if their assumptions are met, the models offer accurate
quantitative predictions, the likes of which are not easily obtainable in the context of simpler
and stylized models. However, the models are no panacea. Their detailed approaches
do not easily allow an empiricist to embed specific alternatives, require certain strong
assumptions (e.g., what kind of debt can be issued and when), and make it difficult for the
casual reader to judge where the degrees of freedom (which the models fit) lie. Moreover,
they are difficult to solve and understand, and it is usually left to the reader to judge their
robustness and specificity—or even how flat the objective functions are, both in a static
19
and in a dynamic sense. Generally, very few structural models have enjoyed wide success
and use, either in corporate finance or asset pricing. In the absence of a generally agreed
successful standard, it is simply not clear what the right model to use is.
In any case, there is consensus in the profession that a simple friction-free trade-off
model cannot explain the evidence. Firms do not readjust their capital structures on short
horizons—say 1 to 2 years. There are also few who would doubt that firms try to readjust
at least partially over longer horizons, say 5 to 10 years. Estimating the extent and speed
of readjustment is now becoming one focus of this literature. However, Leary and Roberts
(2004) point out that firms may adjust capital structure in lumps, rather than continuously,
which presents interesting theoretical and empirical complications.
B The Pecking Order (Adverse Selection, Slack, and the Financing Pyramid)
The pecking order theory first arose in Donaldson (1961), but has since been most promi-
nently associated with Stewart Myers (Myers (1984), Myers and Majluf (1984)). It was at
first narrowly interpreted to mean that firms never issue risky instruments, because in-
vestors infer negative information about existing projects when a firm wants to issue a risky
instrument—a pure adverse selection argument. In later and broader interpretations, it was
interpreted to be, as it names suggest, a funding preference theory, in which firms fund
new projects first internally with retained earnings, then with debt, and only finally with
equity. In such a world, financial slack becomes valuable because it can provide internal
funding for future projects when external equity is too expensive.
Over time, the pecking-order theory has also acquired an identification with a number of
related phenomena. Consequently, it is useful to clarify the phenomena being discussed.
The pecking order: The preference to fund new projects with more senior claims.
Adverse selection: The fact that insiders know more than potential new investors.
Financial slack: An internal cash reserve that firms can tap.
The financing pyramid: A capital structure that contains more senior than junior claims.
A pecking order can arise in any trade-off theory in which issuing junior claims is more
expensive than issuing senior claims. It can be introduced not only through adverse
selection costs to equity (the traditional method), but also through agency costs to equity
(e.g., Morellec (2004)), or even through higher physical distribution costs of new equity
20
shares.9 For example, investors may fear that equity issues might give managers more
access to unrestricted capital, which reduces the firm value. Fearing such negative reactions
enough (i.e., if they outweigh the benefits to issuing equity), managers then avoid issuing
equity, even in the absence of adverse selection in the sense of negative inference about
existing projects. The complement, benefits of debt, can also generate pecking orders (for
example in Auerbach (1979)). In general, it is not difficult to create models in which internal
equity is cheap (e.g., Lewellen and Lewellen (2004)). Any such theory can then also justify a
negative announcement effect, because the attempt to issue informs investors of managerial
intent and/or that the firm had to resort to such costly methods of equity financing.
Figure 5 illustrates the relationships between these phenomena. The most important aspect
of the figure is that it considers them to be distinct. Each can occur without the other,
although the presence of a phenomena higher up in the figure can help cause phenomena
higher down. For example, adverse selection can cause a pecking-order—but so can other
costs of equity. Financial slack is a natural corporate response when there are future costs
of obtaining new capital. And a financing pyramid can result from a pecking order over
time, but it can also be influenced by other factors, such as corporate value changes. The
figure also clarifies that one cannot conclude from the presence of a pecking order that
adverse selection is at work, and that one cannot conclude from the presence of a financing
pyramid that a pecking order is at work. It is only true that in the absence of other forces,
adverse selection causes a pecking order.
On a historical note, when Stewart Myers proposed the pecking-order theory, among his
intentions was to provide a NULL hypothesis against which the trade-off theory could
be tested. One problem of such a perspective, however, is that the trade-off theory is
really only a statement of optimization—such as value maximization or managerial utility
maximization. There is no clear upfront identification of the trade-off forces. Therefore,
if it were to be viewed as the alternative, the pecking-order would have to be a rejection
of generic optimization, which was surely not Myers’ intent. This has been confounded
by the fact that the two theories have indeed often been mistakenly presented as being
9This has also recently been derived in Frank and Goyal (2005). Like Bolton and Dewatripont (2005, p.119f),they also point out that it is possible to have a different form of adverse selection create an inverse peckingorder—a preference for equity.
21
Figure 5: Related But Distinct Phenomena
Traditional Grouping (PO)
Adverse SelectionManagers know more about projects
Financial SlackFirms are underlevered toavoid future equity issues
@@R
Slow Active ReadjustmentFirms remain underlevered
@@R
Financing PyramidFirms are more debt financed
Other Conceivable Causes
Moral Hazard, Transaction CostsIssuing is costly a Markets react badly.
������
Potential Causes:
Agency IssuesManagers dislike high debt levels
or “forget” to pay out earnings.
Potential methods:
Broader ActivityEarnings retention.
����
Potential Causes:
Agency IssuesManagers dislike high debt levels
or “forget” to pay out earnings.
Potential methods:
Broader ActivityDebt/Payout Policy.
��
Potential Causes:
Value ChangesStock/Bond Returns (MV). Ret. Earnings, Deprec. (BV)
Potential methods:
Broader ActivityDebt/Payout Policy.
��
Broader activity includes repurchasing activity, payout policy, debt activities, etc. Multiple arrows indicate acloser and more general connection, while single arrows mean a possible relation in some, though not allcompanies.
mutually exclusive—on two opposing ends of one spectrum.10,11 This is not at all the case.
Instead, one can view the pecking-order theory as the funding phenomenon that is the
result of a set of forces that postulate high costs to issuing more junior securities and
which are balanced by the firm against their benefits. If the marginal cost of more equity
is high enough, then firms tend to fund projects with more senior securities—a pecking
10It is sometimes suggested that the difference between pecking-order and tax benefits of debt is thatthe former is about issuing changes in debt and equity, while the latter is merely about the level of debtand equity. Yet, it would seem natural even in a pecking-order to think of an equity repurchase as goodinformation, the same way one would think of equity issue as negative information. In this interpretation,the difference between the PO and the TO is [a] about the fact that the pecking-order does not as explicitlyentertain a benefit to equity; and [b] that the pecking-order is about an asymmetric reaction to good and badinference.
11Some papers show that simulated runs of either the trade-off theory or of a pecking order can explainthe empirical evidence. This is correct, but it does not mean a rejection of the other theory.
22
order which is the outcome of a plain capital structure cost-vs-benefit trade-off. Coercing
the two theories to be viewed as mutually exclusive, one being the converse of the other, is
confusing.
This does not mean that PO tests are necessarily TO tests. The PO tests are about specific
empirical predictions, even though they are based on the same theoretical structure. They
are about how firms fund new projects with additional capital, while tests of the tradeoff
theory can be about the relative benefits of debt and equity in a more general context,
including situations in which there is no need to fund new projects.
C A Short Clinical Analyses
Much empirical capital-structure work has been conducted through the lenses of these
theories. This has confirmed that both trade-off and pecking order behavior are present, at
least in some firms. However, it is not clear whether these empirical tests have captured
the first-order determinants of capital structure, or merely marginal effects.12
In fact, with the exception of one working paper (Frank and Goyal (2003)), the literature has
not even answered such basic questions as to whether large firms or small firms tend to
have higher debt ratios. Even without a good theory, it would be interesting to learn more
about whether and how such basic characteristics as firm size or industry (MacKay and
Philips (2004)) are capital structure determinants—and then to build a theory why this is
so. Thus, it could be useful to describe the first-order differences across firms that have
high or low debt ratios, that are then to be explained by theories.
In this vein, Tables 4 and 5 present a short clinical analysis of current capital structure.
They show the firms that have the lowest and highest debt ratios among S&P 500 members
in February 2006, as obtained from Yahoo!Finance. It does not require a rigorous analysis
to notice a number of strong regularities (most of which have appeared in the literature):
1. Industry seems to play an important role.
• Among the most highly levered companies, there are a large number of financial
services and automobile related companies. These can be fundamentally different
from industrial firms. For instance, FNMA has 36 times as much debt as equity.
12Looking at these theories at such a broad canvas suggests a higher hurdle. With a large set of possibleproxies, tens of thousands of firm-years, and fairly general specifications, the goal should not be merely toobtain marginal statistical significance. Instead, the theoretical proxies should explain a reasonably largefraction of capital structure variation. Losely speaking, it is not just the t-statistic, but the R-square that weneed to identify first-order important phenomena.
23
Tab
le4:
Cap
ital
Stru
ctu
re:
Hig
hly
Leve
red
S&P500
Com
pan
ies
inFe
bru
ary
2006
Tic
ker
Nam
eIn
du
stry
D/E
-MV
L/A
-MV
L/A
-BV
MV
-EB
V-E
Totl
iab
LTD
CD
TT
axR
DIn
tO
CF
ICF
FCF
NR
atin
gR
R5
R1
0
FNM
Fan
nie
Mae
F3
59
7%
#N
/A#
N/A
$5
0.8
#N
/A#
N/A
#N
/A#
N/A
#N
/A#
N/A
#N
/A#
N/A
#N
/A#
N/A
5N
LS−
30
%−
37
%2
82
%
TX
UT
XU
U2
82
0%
98
%$
0.5
$2
5.1
$1
1.4
$2.3
$0.6
-$
0.8
$2.5
−$
1.0
−$
1.6
7B
BB
-
SLM
SLM
F2
42
5%
81
%9
6%
$2
2.7
$3.8
$9
5.5
$8
8.1
$3.8
$0.7
-$
3.1
−$
0.7
−$
15.7
$1
5.5
11
A5
%1
60
%2
02
7%
FRE
Fred
die
Mac
F2
24
0%
#N
/A#
N/A
$4
4.8
#N
/A#
N/A
#N
/A#
N/A
#N
/A#
N/A
#N
/A#
N/A
#N
/A#
N/A
5A
A-
−1
0%
8%
55
5%
GT
Good
year
C7
42
9%
83
%1
00
%$
3.1
$0.1
$1
5.6
$4.7
$0.4
$0.3
-$
0.4
$0.9
−$
0.4
−$
0.2
80
B+
19
%−
18
%−
34
%
LULu
cen
tIT
18
36
%5
7%
98
%$
11.9
$0.4
$1
6.0
$5.1
$0.4
−$
0.2
$1.2
$0.3
$0.7
−$
1.3
−$
0.4
30
B−
29
%−
76
%-
MS
Morg
anSt
anle
yF
17
92
%9
3%
97
%$
67.5
$2
9.2
$8
69.3
$2
98.0
$4
15.4
$1.9
-$
24.4
−$
31.4
−$
4.1
$3
2.1
53
A+
4%−
22
%8
7%
GM
Gen
eral
Moto
rsC
17
08
%9
8%
97
%$
10.8
$1
4.6
$4
61.5
$2
85.8
-−
$5.9
-$
15.8
−$
16.9
$8.6
$3.5
32
7B
−4
8%−
51
%−
10
%
BSC
Bea
rSt
ern
sF
14
30
%9
6%
$1
0.8
$2
81.8
$6
6.0
$1
21.2
$0.7
-$
4.1
−$
14.0
−$
0.2
$1
5.9
11
A
LEH
Leh
man
Bro
sF
12
96
%9
1%
96
%$
40.3
$1
6.8
$3
93.3
$2
14.1
$1
19.1
$1.6
$0.2
$1
7.8
−$
7.5
−$
0.4
$7.4
22
A+
48
%9
6%
18
45
%
UST
UST
C1
28
0%
95
%$
0.1
$1.3
$0.8
#N
/A$
0.3
-$
0.1
$0.6
−$
0.0
−$
0.8
5A
AES
AES
Corp
.U
12
63
%7
3%
94
%$
10.4
$1.6
$2
8.0
$1
6.8
$1.8
$0.5
-$
1.9
$2.2
−$
0.8
−$
1.2
30
B+
16
%−
71
%2
24
%
FFo
rdC
11
91
%9
5%
95
%$
14.2
$1
3.0
$2
56.5
$1
54.3
-−
$0.5
-$
7.6
$2
1.7
$7.4
−$
20.7
30
0B
B-
−4
5%−
61
%1
3%
GS
Gold
man
Sach
sF
11
70
%9
1%
96
%$
69.4
$2
8.0
$6
78.8
$1
23.3
$3
53.3
$2.6
$0.4
$1
8.2
−$
12.4
−$
1.1
$1
9.4
31
A+
24
%2
5%
-
CFC
Cou
ntr
yFi
n’l
F8
60
%9
3%
$1
2.8
$1
62.3
$7
6.2
$3
6.4
$1.6
-$
5.6
−$
11.7
−$
41.8
$5
3.7
54
A
CN
PC
ente
rpoin
tU
70
0%
92
%$
1.3
$1
5.8
$8.6
$0.7
$0.2
-$
0.7
$0.0
$0.0
$0.0
9B
BB
CIT
CIT
Gro
up
F6
87
%8
4%
89
%$
10.8
$7.0
$5
6.4
$4
7.9
-$
0.5
-$
1.9
$2.9
−$
5.0
$3.2
6A
15
%-
-
AM
ZN
Am
azon
IT6
16
%1
5%
93
%$
19.7
$0.2
$3.5
$1.5
-$
0.1
-$
0.1
$0.7
−$
0.8
−$
0.2
12
BB
-6
%2
03
%-
NA
CN
avis
tar
C5
65
%7
8%
94
%$
2.0
$0.5
$7.1
$2.0
$0.8
$0.1
$0.2
$0.1
$0.2
−$
0.1
$0.1
14
BB
-−
35
%9
%8
9%
EPEl
Pas
oI
52
0%
89
%$
3.4
$2
8.4
$1
7.0
$1.2
−$
0.3
-$
1.4
$0.3
−$
0.5
$0.2
5B
+
ETE*
Tra
de
F5
10
%#
N/A
#N
/A#
N/A
#N
/A#
N/A
#N
/A#
N/A
#N
/A#
N/A
#N
/A#
N/A
3B
+
AX
PA
mer
ican
Exp
ress
F4
40
%6
2%
91
%$
63.4
$1
0.5
$1
03.4
$3
0.8
$1
5.6
$1.0
$0.0
$0.9
$8.0
−$
17.3
$6.4
65
A+
5%
12
%6
24
%
CZ
NC
itiz
ens
Com
m.
IT4
06
%5
8%
84
%$
3.9
$1.0
$5.4
$4.0
$0.2
$0.1
-$
0.3
$0.8
−$
0.2
−$
0.5
6B
B+
−4
%2
2%
41
%
PB
IPit
ney
C3
60
%8
8%
$1.3
$9.3
$3.8
$0.9
$0.3
$0.2
$0.2
$0.5
−$
0.5
−$
0.1
34
A+
TH
CT
enet
H3
50
%9
0%
$1.0
$8.8
$4.8
$0.0
−$
0.1
-$
0.4
$0.8
−$
0.4
$0.3
53
B
AZ
OA
uto
zon
eS
34
0%
91
%$
0.4
$3.9
$1.9
-$
0.3
-$
0.1
$0.6
−$
0.3
−$
0.4
29
BB
B+
GE
Gen
eral
Elec
tric
*3
39
%9
6%
84
%$
24.0
$1
09.4
$5
64.0
$2
12.3
$1
58.2
$3.9
-$
15.2
$3
7.6
−$
35.0
−$
6.1
31
6A
AA
−1
%1
8%
−8
4%
CA
TC
ater
pil
lar
I3
05
%5
0%
82
%$
38.6
$8.4
$3
8.6
$1
5.7
$1
0.1
$1.1
$1.1
$1.0
$3.1
−$
3.5
$1.2
85
A2
1%
17
6%
44
5%
DJ
Dow
Jon
esS
29
1%
36
%9
1%
$2.9
$0.2
$1.6
$0.2
$0.2
$0.0
-$
0.0
$0.2
−$
0.5
$0.3
7B
BB
+−
15
%−
30
%6
6%
CM
SC
MS
Ener
gy
U2
99
%8
1%
85
%$
3.2
$2.3
$1
3.4
$7.3
$0.4
−$
0.2
-$
0.5
$0.6
−$
0.5
$0.1
8B
B3
9%−
48
%−
8%
DE
Dee
reI
27
2%
63
%8
0%
$1
5.9
$6.9
$2
6.8
$1
1.7
$6.9
$0.7
$0.7
$0.8
$1.2
−$
5.2
$3.1
47
A-
−7
%6
4%
33
2%
HN
ZH
JH
ein
zC
27
2%
42
%7
5%
$1
1.2
$2.6
$8.0
$4.1
$0.6
$0.3
-$
0.2
$1.2
−$
0.3
−$
1.1
41
A-
−1
1%−
15
%1
37
%
CL
Colg
ate-
Pal
moli
veC
25
5%
20
%8
4%
$2
9.4
$1.4
$7.2
$2.9
$0.5
$0.7
-$
0.1
$1.8
−$
0.2
−$
1.5
35
AA
-1
0%
−8
%3
61
%
Sou
rce:
Yah
oo.
F=Fi
nan
cial
Serv
ices
.C
=C
on
sum
erG
ood
s(i
ncl
Au
to).
U=
Uti
liti
es.
H=
Hea
lth
care
.I=
Ind
ust
rial
.IT
=In
fote
ch.
BT
=B
iote
ch.
Ple
ase
note
that
D/E
and
L/A
her
ew
ere
pro
vid
edth
era
nk
ord
er,
asof
Feb
ruar
y2
00
6.
How
ever
,th
isca
nn
ot
be
easi
lyre
fill
edn
ow
.D
/Eis
the
mar
ket
-bas
edd
ebt-
equ
ity
rati
o.
L/A
isth
eli
abil
ity-
asse
tra
tio,m
arket
-val
ue
bas
ed.
Totl
iab
=to
talli
abil
itie
s,LT
D=
lon
g-t
erm
deb
t,C
DT
=d
ebt
incu
rren
tli
abil
itie
s.T
axar
eIn
com
eta
xes,
RD
isR
&D
exp
ense
,In
tis
inte
rest
exp
ense
.O
CF
isop
erat
ing
cash
flow
,IC
Fis
inve
stm
ent
cash
flow
,FC
Fis
fin
anci
ng
cash
flow
.D
oll
arfi
gu
res
are
inb
illi
on
s.N
isn
um
ber
of
emp
loye
es,
inth
ou
san
ds.
Rat
ing
isfr
om
S&P.R
,R5
,an
dR
10
are
on
e,fi
ve,a
nd
10
year
rate
sof
retu
rn—
tob
ere
com
pu
ted
.
24
Tab
le5:
Cap
ital
Stru
ctu
re:
Zer
o-L
ever
edS&
P500
Com
pan
ies
inFe
bru
ary
2006
Tic
ker
Nam
eIn
du
stry
D/E
-MV
L/A
-MV
L/A
-BV
MV
-EB
V-E
Totl
iab
LTD
CD
TT
axR
DIn
tO
CF
ICF
FCF
NR
atin
gR
R5
R1
0
MSF
TM
icro
soft
IT0
%8
%3
2%
$2
69.3
$4
8.1
$2
2.7
--
$4.4
$6.2
-$
16.6
$1
5.0
−$
41.1
61
−1
%−
10
%6
80
%
GO
OG
Google
IT0
%1
%8
%$
12
3.3
$9.4
$0.9
--
$0.7
$0.6
$0.0
$2.5
−$
3.4
$4.4
51
15
%-
-
CSC
OC
isco
IT0
%8
%2
7%
$1
05.3
$2
5.8
$9.8
--
$2.0
$3.2
-$
7.1
$0.5
−$
7.8
38
A+
−1
1%−
55
%7
78
%
QC
OM
Qu
alco
mm
IT0
%2
%1
1%
$7
1.2
$1
1.1
$1.4
--
$0.7
$1.0
$0.0
$2.7
−$
0.8
−$
1.1
93
%7
%2
97
1%
AA
PL
Ap
ple
IT0
%6
%3
5%
$6
1.0
$7.5
$4.1
--
$0.5
$0.5
-$
2.5
−$
2.6
$0.5
14
12
3%
86
6%
64
6%
EBA
YEb
ayIT
0%
3%
15
%$
60.8
$1
0.0
$1.7
--
$0.5
$0.3
$0.0
$2.0
−$
2.5
$0.5
11
−2
6%
42
4%
-
WA
GW
algre
enC
0%
39
%$
8.9
$5.7
--
$0.9
--
$1.4
−$
0.4
−$
0.8
13
1A
+
GIL
DG
ilea
dH
0%
20
%$
3.0
$0.7
$0.2
$0.1
$0.3
$0.3
$0.0
$0.7
−$
0.7
$0.4
1
BR
CM
Bro
adco
mIT
0%
4%
16
%$
16.5
$3.1
$0.6
--−
$0.0
$0.7
-$
0.4
−$
0.2
$0.3
44
7%−
44
%-
AD
BE
Ad
ob
eIT
0%
3%
24
%$
22.2
$1.9
$0.6
--
$0.2
$0.4
-$
0.7
−$
0.3
−$
0.2
51
8%
13
9%
94
7%
GEN
ZG
enzym
eB
T0
%2
5%
$5.1
$1.7
$0.8
$0.0
$0.2
$0.5
$0.0
$0.7
−$
1.2
$0.3
8B
BB
AC
EA
ceF
0%
81
%$
11.8
$5
0.6
$4.5
$0.3
$0.3
-$
0.2
$4.3
−$
5.6
$1.3
10
BB
B+
BII
Bio
gen
BT
0%
17
%$
6.9
$1.5
$0.0
-$
0.1
$0.7
$0.0
$0.9
$0.4
−$
0.9
3B
BB
ERT
SEl
ectr
on
ics
Art
sIT
0%
20
%$
3.5
$0.9
--
$0.1
$0.8
-$
0.6
−$
0.1
−$
0.5
7
FRX
Fore
stLa
bs
H0
%1
2%
$3.1
$0.4
--
$0.3
$0.3
-$
0.9
$0.1
−$
1.0
5
PA
YX
Pay
chex
IT0
%6
8%
$1.4
$3.0
--
$0.2
--
$0.5
−$
0.2
−$
0.2
10
AD
IA
nal
og
Dev
ices
IT0
%1
9%
$3.7
$0.9
--
$0.2
$0.5
$0.0
$0.7
−$
0.0
−$
0.6
8
MX
IMM
axim
IT0
%3
%1
4%
$1
1.6
$2.6
$0.4
--
$0.3
$0.3
-$
0.7
−$
0.5
−$
0.2
7−
14
%−
22
%7
48
%
NT
AP
Net
work
Ap
pli
ance
IT0
%3
0%
$1.7
$0.7
$0.0
-$
0.1
$0.2
$0.0
$0.5
−$
0.4
−$
0.0
4
LLT
CLi
nea
rT
ech
IT0
%2
%1
2%
$1
1.0
$2.0
$0.3
--
$0.2
$0.1
-$
0.5
−$
0.1
−$
0.3
3−
6%−
19
%5
30
%
BB
BY
Bed
Bat
hB
eyon
dC
0%
31
%$
2.2
$1.0
-$
0.1
$0.3
--
$0.6
−$
0.4
−$
0.3
33
BB
B
KLA
CK
lA
-Ten
cor
IT0
%2
4%
$3.0
$0.9
--
$0.2
$0.3
$0.0
$0.5
−$
0.2
−$
0.1
5
TR
OW
TR
ow
ePri
ceF
0%
12
%$
2.0
$0.3
--
$0.2
-$
0.0
$0.5
−$
0.1
−$
0.1
4
APO
LA
poll
oS
0%
5%
46
%$
10.4
$0.7
$0.6
--
$0.3
--
$0.6
$0.2
−$
0.8
11
AA
A−
25
%1
77
%-
XLN
XX
ilin
xIT
0%
4%
12
%$
8.7
$2.7
$0.4
--
$0.1
$0.3
-$
0.3
−$
0.0
−$
0.1
3−
14
%−
45
%4
19
%
INT
UIn
tuit
IT0
%9
%3
2%
$9.3
$1.8
$0.9
$0.0
-$
0.1
$0.3
-$
0.6
−$
0.2
−$
0.5
72
1%
35
%3
79
%
AD
SKA
uto
des
kIT
0%
5%
43
%$
9.9
$0.6
$0.5
--
$0.0
$0.2
-$
0.4
$0.2
−$
0.3
41
3%
55
6%
38
4%
NV
DA
Nvi
dia
IT0
%2
8%
$1.2
$0.5
-$
0.0
$0.0
$0.3
$0.0
$0.1
−$
0.2
$0.0
2B
B-
VR
SNV
eris
ign
IT0
%2
%3
6%
$5
3.5
$2.0
$1.1
--
$0.1
$0.1
-$
0.5
$0.1
−$
0.5
4−
35
%−
70
%-
AB
IA
pp
lera
(Bio
)T
0%
14
%2
6%
$4.9
$2.3
$0.8
--
$0.0
$0.3
$0.0
$0.2
$0.1
$0.0
4A
-2
8%−
71
%-
APC
CA
PC
IT0
%2
1%
$1.6
$0.4
--
$0.1
$0.1
-$
0.2
$0.1
−$
0.0
7
CPW
RC
om
pu
war
eIT
0%
39
%$
1.5
$1.0
--
$0.0
$0.2
-$
0.2
−$
0.2
$0.0
7
QLG
CQ
logic
IT0
%7
%$
1.0
$0.1
--
$0.1
$0.1
-$
0.2
−$
0.0
−$
0.1
0
TEK
Tek
tron
ixIT
0%
2%
32
%$
23.3
$1.0
$0.5
--
$0.0
$0.2
$0.0
$0.1
$0.1
−$
0.2
4B
B+
−6
%−
15
%1
94
%
PM
TC
Par
amet
ric
Tec
h.
IT0
%3
%5
9%
$1
7.0
$0.3
$0.5
--
$0.0
$0.1
-$
0.1
−$
0.2
−$
0.0
34
%−
55
%−
29
%
AM
CC
Ap
pli
edM
icro
IT0
%1
1%
$1.0
$0.1
-$
0.0
$0.0
$0.1
-−
$0.0
−$
0.2
−$
0.0
0
Sou
rce:
Yah
oo.
F=Fi
nan
cial
Serv
ices
.C
=C
on
sum
erG
ood
s(i
ncl
Au
to).
U=
Uti
liti
es.
H=
Hea
lth
care
.I=
Ind
ust
rial
.IT
=In
fote
ch.
BT
=B
iote
ch.
Ple
ase
note
that
D/E
and
L/A
her
ew
ere
pro
vid
edth
era
nk
ord
er,
asof
Feb
ruar
y2
00
6.
How
ever
,th
isca
nn
ot
be
easi
lyre
fill
edn
ow
.D
/Eis
the
mar
ket
-bas
edd
ebt-
equ
ity
rati
o.
L/A
isth
eli
abil
ity-
asse
tra
tio,m
arket
-val
ue
bas
ed.
Totl
iab
=to
talli
abil
itie
s,LT
D=
lon
g-t
erm
deb
t,C
DT
=d
ebt
incu
rren
tli
abil
itie
s.T
axar
eIn
com
eta
xes,
RD
isR
&D
exp
ense
,In
tis
inte
rest
exp
ense
.O
CF
isop
erat
ing
cash
flow
,IC
Fis
inve
stm
ent
cash
flow
,FC
Fis
fin
anci
ng
cash
flow
.D
oll
arfi
gu
res
are
inb
illi
on
s.N
isn
um
ber
of
emp
loye
es,
inth
ou
san
ds.
Rat
ing
isfr
om
S&P.R
,R5
,an
dR
10
are
on
e,fi
ve,a
nd
10
year
rate
sof
retu
rn—
tob
ere
com
pu
ted
.
25
• Some firms have slipped into this category due to poor performance, such as
Goodyear, GM and Ford.
• Among the least levered companies, there are disproportionally many information
technology companies.
2. There are a number of firms which have very low equity book values and much higher
market value. For example, Amazon’s equity was $20 billion in market value but only
$0.2 billion in book value. (Theoretically, the book value of equity can be negative.)
3. Among highly levered firms, total liabilities can be quite different from financial debt.
For example, Ace (an insurance firm) listed $50.6 billion in total liabilities but no
financial debt.
4. Taxes consume about a third of the operating cash flows of many zero-leverage firms.
Many highly levered firms have no or do not report R&D expenses. R&D, taxes, and
interest expenses can be large in terms of operating cash flows, but are often only a
small fraction of firm asset value.
5. The distribution of credit ratings provides some insight about the reason why firms
and CMS have junk bond status. This is also partly reflected in their historical stock
market rates of return.
6. Highly indebted firms are older and have more employees than zero-levered firms.
These seem to be first-order effects. Clinical analyses can sometimes point out the pitfalls
in more formal tests of capital structure theory. For example, a number of theories (e.g.,
Titman (1984)) predict that firms with intangible assets that would dissipate in financial
distress should have more equity. Some tests have used R&D expenditures as a proxy. It is
indeed correct that firms with high R&D expenditures and in the IT industry have low debt
ratios. Yet, I would argue that the question remains whether this should be interpreted as
good evidence that the intangibility of assets has induced these firms not to take on leverage.
In my opinion, the table suggests that this may not be the case. Microsoft, Google, Cisco,
Qualcomm, Apple, etc., are probably not unlevered because they chose such structures
in order to avoid bankruptcy at all cost. Given their capitalization and product lines, a
10% debt-ratio (rather than 0%) would almost surely not increase their expected costs of
bankruptcy enough to outweigh the immediate gain in lower tax obligations. This would
suggest that when R&D expenses are used in terms of proxying for distress costs, they are
not only noisy but also have high type-I error: The tests could suggest that the theory were
26
true when it is false (when low debt is driven by other considerations, instead). A more likely
explanation is that these particular firms in R&D intensive industries have experienced high
operating cash flows in the past. A full control for such third factors may not be easy—and
it is not clear to me what the role of R&D is. A better research design to test whether firms
with high R&D expenses have low debt ratios because they need to avoid financial distress
would be to search for the impact of R&D principally among firms that have a reasonable
probability of encountering financial distress.
V Future Directions in Empirical Capital Structure Research
In a working paper critiquing the past methods of this literature, it seems not entirely out of
place to use two pages to speculate about future promising directions in this literature—even
if this section may ultimately not be suitable for publication.
Despite years of capital structure research, it is not clear to me what the important empirical
findings are. As to first-order effects, it is my impression that we are on solid ground with
respect to the following phenomena: most firms are reluctant to finance new projects with
new equity (a pecking order), except in the context of acquisitions; that (long-term) and other
debt net issuing activity is common; that there are industry differences (possibly caused
by industry performance differences); that acquisitions play an important role, and that
historical stock returns are important factors, because are not immediately counteracted
by firms. It is not clear to me whether the following are first-order or second order effects:
Firms seem to want to be similar to their peers. There is also some evidence in favor of
the trade-off theory (especially with respect to tax effects), and some evidence in favor of
a capital structure that can be explained by an agency conflict (the desire of managers to
avoid debt and have cash or equity to take over other companies). Other evidence seems
more preliminary—for example, we are just learning more about dynamic financing policy
and market timing.
As to future research, the following appear to be promising avenues to me:
1. Back to Basics: It would be useful to learn the first-order differences among firms’ capital
structure, even if not guided by specific theory tests. This may help develop future
theories that explain the most important aspects of capital structure better.
2. M&A Activity: Fama and French (2005) have documented that M&A activity can be an
order of magnitude more important in explaining capital structure than ordinary
27
equity issuing activity. We also know that mergers and acquisitions are responsible
for large changes in debt. It will require a more unified perspective of capital structure
and acquisition activity to better understand capital structure. An interesting first
paper exploring methods of payment and M&A is Baker, Coval, and Stein (2004).
3. Survey Evidence: Graham and Harvey (2001) initiated the first modern survey of CFOs.
If we want to learn why corporations end up with certain capital structures, learning
how the main decision makers think is essential evidence. Alas, the weakness of
this literature is that it has not linked actual behavior to survey responses. For
example, many managers claim to pursue a target debt ratio—but the empirical
evidence in this respect is not strong. To make progress, we must be able to correlate
individual managers’ survey responses with their observed behavior. To make even
more progress, we should confront CFOs claiming one type of behavior and behaving
differently with the evidence in order to learn whether their survey response was a
mistake, their capital structure behavior was a mistake, whether times have changed,
or whether we have misunderstood them altogether.
4. Behavioral Finance: Linking the survey evidence to behavior could lead to a more produc-
tive use of behavioral finance in corporate finance. After all, capital structure decisions
are made by managers and with relatively modest forces capable of arbitraging away
mistakes.
Here, I mean not general motivation, but direct identification of particular firms’ CFO
answers with their actual behavior, so that it can be checked, e.g., whether managers
who claim they rebalance all the time actually do so. Still, even without good direct
survey links, this area has produced some interesting findings.
Baker and Wurgler (2002) were generally motivated by the survey evidence and find
that managers attempt to time their equity issues, although the role of stock prices
in predicting issuing activity was mentioned by Marsh (1982), Taggart (1977), Myers
(1984) and others. They argue that timing effects are persistent enough to create
strong path dependence, and explain a good part of the contemporaneous capital
structure variation in firms. However, their findings are not undisputed, either. For
example, Welch (2004),13 Roberts and Leary (2005) and Kayhan and Titman (2006)
13Some papers (e.g., Roberts and Leary (2005)) group Welch (2004) with Baker and Wurgler (2002). However,the two papers could not be more different. The former is about a passive change of capital structure due tostock returns, regardless of active managerial change. The latter is about active managerial behavior in theface of different stock returns, regardless of mechanical, passively induced changes. The two effects are alsoeasy to disentangle.
28
wonder how strong the effect is—companies can issue both debt and equity, and pay
out and repurchase, in response to value increases. The jury is still out.
Another promising behavioral research strand has arisen in Bertrand and Schoar
(2003). They ask to what extent CFO behavior is driven by non-financial factors,
such as education, relationships, etc. Similarly, in a series of paper, Malmendier (e.g.,
Malmendier and Tate (2005)) has begun to measure managerial overconfidence and
its impact on corporate behavior.
4. Natural Experiments In the last decade, the econometric tool-set and the method of
thinking about econometric issues has improved considerably. Instrumental variable
and regression discontinuity techniques have helped to recast and analyze a number
of phenomena as “natural experiments.” There have been some attempts in corporate
finance, e.g., Opler and Titman (1994), Zingales (1998), Chevalier and Scharfstein
(1995), Lamont (1997), and others. In the capital structure arena, phenomena waiting
to be better exploited are tax law changes which only affect some companies, the
sudden and unforeseen destruction of assets of some companies, discontinuity in
credit ratings (Kisgen (2006)), and so on.
5. Event Studies In the late 1980s, there were a large number of event studies. The
technique was perhaps over-used. But the pendulum has swung back too far, and they
have so fallen out of favor that they are now under-used.14 Event studies are a form of
a “natural experiment.” The exogenous variable is clear (the corporate behavior) as is
the endogenous variable, the financial market response at the announcement. A good
event study can help illuminate whether value effects of capital structure changes are
positive or negative, especially when combined with other evidence.
Many good event studies were published in the 1980s and thus were based on data
from the 1960s and 1970s. It is often not known whether or how the financial markets
have changed in the assessment of value changes associated with capital structure
events.
14Admittedly, a second reason for their fall in popularity has come with the advent of behavioral finance.We are more agnostic today as to whether limits of arbitrage prevent the financial markets from fullyincorporating information upon announcement. We understand that financial markets are more noisycreatures than we once thought. Nevertheless, event studies are powerful tools.
29
VI Conclusion
The paper’s main contribution was to critique three issues, and offer suggestions for future
specifications.
1. Proxy Choice: The opposite of financial debt is not equity, but in large part non-financial
liabilities. Researchers should not use a financial-debt-to-asset ratio. Instead, they
should use either a financial-debt-to-financial-equity ratio or a liability-to-asset ratio.
2. Specification: Linear regressions predicting capital structure ratios can have low ex-
planatory power, when they are confronted with non-linearities in the dependent
variable. Debt-capital ratios are inherently non-linear in debt and capital. The example
in this paper showed that even with perfect foreknowledge of equity and debt changes,
a linear regression may pick up less than 10% (25% if variables are truncated) of the
variation in leverage ratios. Interpreting fit as a measure of how well variables explain
capital structure can then be misleading.
Researchers could estimate a system that allows for both linear and non-linear effects
of any third variables.
3. Survivorship: Corporate birth and death rates suggest significant selectivity biases,
even on annual horizons. About one in ten firms disappears and/or appears in any
given year. Moreover, there are systematic differences (in capital structure and other
characteristics) as to what types of firms appear and disappear. This bias renders
multi-year long-run explorations progressively more difficult.
Researchers can use Heckman techniques to model the birth and death processes of
the observations that flow into their capital structure estimation.
Finally, the paper has taken some unusual liberties clarifying and opining on theoretical
interpretations and future work. First, it has argued that it is useful to draw a distinction
between capital structure mechanisms and causes. Remarkably little is known about
mechanisms. Second, when it comes to causes, the empirical literature has often suggested
that the takeover and a pecking order theory are opposites or mutually exclusive. However,
they are merely different implications of the same theory. Third, the paper suggested
that it would be useful to distinguish better between the pecking order, adverse selection,
financial slack, and the financing pyramid as discrete empirical phenomena. Fourth, the
paper “pontificated” as to some promising avenues for further research.
30
References
Almeida, H., and T. Philippon, 2006, “The Risk-Adjusted Cost of Financial Distress,” working paper, NewYork University.
Auerbach, A., 1979, “Wealth Maximization and the cost of capital,” Quarterly Journal of Economics, 93,433–446.
Baker, M., and J. Wurgler, 2002, “Market Timing and Capital Structure,” The Journal of Finance, 57(1), 1–32.
Baker, M. P., J. D. Coval, and J. C. Stein, 2004, “Corporate Financing Decisions when Investors take the path ofleast resistance,” working paper, Harvard.
Bertrand, M., and A. Schoar, 2003, “Managing with Style: The Effect of Managers on Corporate Policy,”Quarterly Journal of Economics, 118(4), 1169–1208.
Bolton, P., and M. Dewatripont, 2005, Contract Theory. The MIT Press, Cambridge, MA.
Chen, L., and X. Zhao, 2006, “Mechanical Mean Reversion of Leverage Ratios,” Economic Letters.
Chevalier, J. A., and D. S. Scharfstein, 1995, “Liquidity constraints and the cyclical behavior of markups,”American Economic Review, 85(2), 390–396.
Donaldson, G., 1961, “Corporate Debt Capacity: A Study of Corporate Debt Policy and the Determination ofCorporate Debt Capacity,” working paper, Harvard GSBA.
Fama, E. F., and K. French, 2005, “Financing Decision: Who Issues Equity,” Journal of Financial Economics, 76,549–582.
Fischer, E. O., R. Heinkel, and J. Zechner, 1989, “Optimal Dynamic Capital Structure Choice : Theory andTests,” Journal of Finance, 44, 19–40.
Frank, M. Z., and V. K. Goyal, 2003, “Capital Structure Decisions: Which Factors are Reliably Important?,”working paper, University of Minnesota.
, 2005, “Trade-off and Pecking Order Theories of Debt,” in Handbook of Corporate Finance: EmpiricalCorporate Finance, ed. by B. E. Eckbo. Elsevier.
Graham, J. R., 2000, “How big are the tax benefits of debt?,” Journal of Finance, 55, 1901–1941.
, 2003, “Taxes and Corporate Finance: A Review,” Review of Financial Studies, 16, 1074–1129.
Graham, J. R., and C. R. Harvey, 2001, “The Theory and Practice of Corporate Finance: Evidence from theField,” Journal of Financial Economics, 60, 187–243.
Harris, M., and A. Raviv, 1991, “The Theory of Capital Structure,” Journal of Finance, 46, 297–355.
Hennessy, C., and T. Whited, 2004, “Debt Dynamics,” Journal of Finance, 60(3), 1129–1165.
Kayhan, A., and S. Titman, 2006, “Firms’ Histories and Their Capital Structures,” working paper, Universityof Texas at Austin, JFE, forthcoming.
Kisgen, D. J., 2006, “Credit Ratings and Capital Structure,” Journal of Finance, 61(3), 1035–1072.
Lamont, O., 1997, “Cash Flow and Investment: Evidence from Internal Capital Markets,” Journal of Finance,52.
Leary, M. T., and M. R. Roberts, 2004, “Do Firms Rebalance Their Capital Structures,” The Journal of Finance,60(6), 2575–2619.
Leland, H., 1998, “Agency Costs, Risk Management, and Capital Structure,” Journal of Finance, 53, 1213–1244.
Leland, H. E., and K. B. Toft, 1996, “Optimal Capital Structure, Endogenous Bankruptcy, and the Term Structureof Credit Sprreads,” Journal of Finance, 51, 987–1019.
Lemmon, M. L., M. R. Roberts, and J. F. Zender, 2006, “Back to the Beginning: Persistence and the Cross-Sectionof Corporate Capital Structure,” working paper, University of Utah, Wharton, and UC/Boulder.
Lewellen, J., and K. Lewellen, 2004, “Taxes and Financing Decisions,” working paper, Tuck School at Dartmouth.
MacKay, P., and G. M. Philips, 2004, “How Does Industry Affect Firm Financial Structure,” Review of FinancialStudies, 18(4), 1433–1466.
31
Malmendier, U., and G. Tate, 2005, “Does Overconfidence Affect Corporate Investment? CEO OverconfidenceMeasures Revisited,” European Financial Management, 11(5), 649–659.
Marsh, P. M., 1982, “The Choice Between Equity and Debt: An Empirical Study,” Journal of Finance, 37,121–144.
Morellec, E., 2004, “Can Managerial Discretion Explain Observed Leverage Ratios,” Review of Financial Studies,pp. 257–294.
Myers, S. C., 1984, “The Capital Structure Puzzle,” Journal of Finance, 39, 575–592.
Myers, S. C., and N. S. Majluf, 1984, “Corporate Financing and Investment Decisions when firms haveinformation investors do not have,” Journal of Financial Economics, 13, 187–221.
Opler, T. C., and S. Titman, 1994, “Financial Distress and Corporate Performance,” Journal of Finance, 49(3),1015–1040.
Rajan, R., and L. Zingales, 1995, “What do we know about capital structure? Some evidence from Internationaldata,” Journal of Finance, 50, 1421–1460.
Roberts, M. R., and M. T. Leary, 2005, “Do Firms Rebalance their Capital Structures?,” Journal of Finance, 60,2575–2619.
Shyam-Sunder, L., and S. C. Myers, 1999, “Testing Static Tradeoff against Pecking Order Models of CapitalStructure,” Journal of Financial Economics, 51, 219–292.
Strebulaev, I. A., 2003, “Do Tests of Capital Structure Theory Mean What They Say?,” working paper, StanfordUniversity.
Taggart, R., 1977, “A Model of Corporate Financing Decisions,” Journal of Finance, 32, 1467–1484.
Titman, S., 1984, “The Effect of Capital Structure on a Firm’s Liquidation Decision,” Journal of FinancialEconomics, 13(1), 137–151.
Welch, I., 2004, “Capital Structure and Stock Returns,” Journal of Political Economy, 112(1), 106–131.
Zingales, L., 1998, “Survival of the Fittest or the Fattest? Exit and Financing in the Trucking Industry,” Journalof Finance, 53(3), 905–938.