https://doi.org/10.

1017/S0022109019000322
Downloaded from https://www.cambridge.org/core. IP address: 199.94.10.45, on 13 Dec 2019 at 20:01:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
JOURNAL OF FINANCIAL AND QUANTITATIVE ANALYSIS
COPYRIGHT 2019, MICHAEL G. FOSTER SCHOOL OF BUSINESS, UNIVERSITY OF WASHINGTON, SEATTLE, WA 98195
doi:10.1017/S0022109019000322

Leverage and the Beta Anomaly

Malcolm Baker, Mathias F. Hoeyer, and Jeffrey Wurgler *

Abstract
The well-known weak empirical relationship between beta risk and the cost of equity (the
beta anomaly) generates a simple tradeoff theory: As firms lever up, the overall cost of cap-
ital falls as leverage increases equity beta, but as debt becomes riskier the marginal benefit
of increasing equity beta declines. As a simple theoretical framework predicts, we find that
leverage is inversely related to asset beta, including upside asset beta, which is hard to ex-
plain by the traditional leverage tradeoff with financial distress that emphasizes downside
risk. The results are robust to a variety of specification choices and control variables.

I. Introduction
Millions of students have been taught corporate finance under the assump-
tion of the capital asset pricing model (CAPM) and integrated equity and debt
markets. Yet, it is well known that the link between textbook measures of risk
and realized returns in the stock market is weak, or even backward. For example,
a dollar invested in a low beta portfolio of U.S. stocks in 1968 grows to $70.50
by 2011, while a dollar in a high beta portfolio grows to just $7.61 (see Baker,
Bradley, and Taliaferro (2014)). The evidence on the anomalously low returns to
high-beta stocks begins as early as Black, Jensen, and Scholes (1972). In a re-
cent contribution, Bali, Brown, Murray, and Tang (2017) identify it as “one of
the most persistent and widely studied anomalies in empirical research of secu-
rity returns” (p. 2369). A large literature has come to view the beta anomaly,
and related anomalies based on total or idiosyncratic risk, as evidence of mis-
pricing as opposed to a misspecified risk model. In this paper, we consider these

*Baker, [email protected], Harvard Business School and NBER; Hoeyer, [email protected],

University of Oxford; and Wurgler (corresponding author), [email protected], NYU Stern
School of Business and NBER. For helpful comments we thank Heitor Almeida (the referee), Hui
Chen, Robin Greenwood, Sam Hanson, Lasse Pedersen, Thomas Philippon, Ivo Welch, and seminar
participants at the American Finance Association, Capital Group, George Mason University, George-
town University, the Ohio State University, NYU, the New York Fed, Southern Methodist University,
University of Amsterdam, University of Cambridge, University of Tokyo, University of Toronto, Uni-
versity of Miami, University of Utah, USC, and University of Warwick. Baker and Wurgler also serve
as consultants to Acadian Asset Management. Baker gratefully acknowledges financial support from
the Division of Research of the Harvard Business School.

1
 https://doi.org/10.1017/S0022109019000322
Downloaded from https://www.cambridge.org/core. IP address: 199.94.10.45, on 13 Dec 2019 at 20:01:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
2 Journal of Financial and Quantitative Analysis

explanations sufficiently plausible to consider their implications for leverage. In

doing so, we bring ideas from the anomalies literature into corporate finance,
which often assumes efficient and integrated securities markets to highlight other
frictions.
We show that a beta anomaly in equity markets, with the absence of any
other deviations from the Modigliani–Miller setup, leads to a simple tradeoff.
Intuitively, under the anomaly, beta risk is overvalued in equity securities, but not
in debt securities. Ideally, then, to minimize the cost of capital, risk is concentrated
in equity. A firm will always want to issue as much riskless debt as it can. This
lowers the cost of equity by increasing its beta without any “inefficient” transfer of
risk from equity to debt. But, as debt becomes risky, further increases in leverage
have a cost. Shifting overvalued risk in equity securities to fairly valued risk in
debt securities increases the overall cost of capital. For firms with high-beta assets,
this increase is high even at low levels of leverage. For firms with low-beta assets,
this increase remains low until leverage is high. Using the Merton (1974) model
to characterize the functional form of debt betas and the underlying transfer of
risk from equity to debt, we show that there is an interior optimum leverage ratio
that is inversely related to asset beta.
The theory’s immediate prediction is that leverage is negatively related to
asset beta. Empirically, we confirm that there is indeed a robust negative rela-
tionship between leverage and asset beta. The relationship remains strong when
controlling for overall asset variance (a control for distress risk), using alternative
measures of leverage and industry measures of asset beta and overall risk, and
including various other control variables.
While it is reassuring that the beta anomaly tradeoff generates a leverage-
asset beta prediction that is borne out in the data, an obvious alternative expla-
nation is based on the standard tradeoff theory, which also predicts an inverse
relationship between leverage and risk (most intuitively, total asset risk, not asset
beta). For example, Long and Malitz (1985) include systematic asset risk, not just
overall risk, in leverage regressions with the traditional tradeoff between taxes and
financial distress costs in mind. We do not wish to cast doubt on the relevance of
the standard tradeoff theory. Instead, we derive and test additional predictions of
the beta anomaly tradeoff to show that it has incremental explanatory power.
In particular, we prove that, under a beta anomaly, both “upside” and “down-
side” betas are inversely related to leverage. Financial distress costs, on the
other hand, clearly emphasize downside risk. Moreover, consistent with the beta
anomaly tradeoff, we find that leverage is indeed inversely related to upside as-
set beta as well. In some specifications, the upside beta relationship is actually
stronger than the downside beta relationship. While our results do not rule out the
traditional tradeoff theory, given that downside beta still matters, this prediction
and empirical relationship is most easily explained by a beta anomaly tradeoff.
We tentatively suggest that the beta anomaly tradeoff may also help to ex-
plain extremely high and low leverage levels that challenge the standard tradeoff.
For instance, Graham (2000) and others have pointed out that hundreds of prof-
itable firms, with high marginal tax rates, maintain essentially zero leverage. This
is often called the low-leverage puzzle. Conversely, a number of other profitable
firms maintain high leverage despite no tax benefit. The beta anomaly tradeoff
 https://doi.org/10.1017/S0022109019000322
Downloaded from https://www.cambridge.org/core. IP address: 199.94.10.45, on 13 Dec 2019 at 20:01:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
Baker, Hoeyer, and Wurgler 3

could help to explain these patterns. If low leverage firms have determined that
the tax benefit of debt is less than the opportunity cost of transferring risk to
lower-cost equity, low leverage may be optimal even in the presence of additional
frictions; a minor, realistic transaction cost of issuance could drive some firms to
zero leverage. Meanwhile, low asset beta firms with no tax benefits of debt still
resist equity because of its high risk-adjusted cost at low levels of leverage, and
instead use a very large fraction of debt finance.
In addition to providing a novel theory of leverage based on a single devi-
ation from the Modigliani–Miller setup, the beta anomaly approach has an at-
tractive and unifying conceptual feature. The traditional tradeoff theory, under
rational asset pricing, cannot fit both the leverage and asset pricing evidence on
the pricing of beta. If beta is truly a measure of risk relevant for capital structure,
then presumably it would also help to explain the cross section of asset returns,
which it does not. Investors, recognizing the associated investment opportunities,
would demand higher returns on assets exposed to the systematic risk of fire sales,
with high risk-adjusted costs of financial distress. If beta is not a measure of risk,
as the large literature that follows Fama and French (1992), (1993) has claimed,
then asset beta should not be a constraint on leverage, after controlling for total
asset risk. Although the beta anomaly is far from the only force at work in real-life
capital markets (following the standard approach in corporate finance theory, we
focus on a single “friction” (the beta anomaly) for simplicity), it is worth noting
that it offers an internally consistent explanation for both of these patterns.
Section I briefly reviews the beta anomaly and derives optimal leverage un-
der the anomaly. Section II contains empirical hypotheses and tests. Section III
concludes.

II. The Beta Anomaly Tradeoff

A. Motivation and Setup
Over the long run, riskier asset classes have earned higher returns, but the
historical risk-return tradeoff within the stock market is flat at best. The CAPM
predicts that the expected return on a security is proportional to its beta, but in
fact stocks with higher beta have tended to earn lower returns, particularly on
a risk-adjusted basis. As mentioned earlier, Bali et al. (2017) highlight the beta
anomaly as one of the most prominent of the many dozen anomalies studied in
the literature. We refer the reader to that paper, as well as Baker, Bradley, and
Wurgler (2011) or Hong and Sraer (2016) for an overview of the large literature
that shows the beta anomaly’s robustness across sample periods and international
markets and reviews some of its theoretical foundations.
The simple linear specification for the beta anomaly that we employ here is
(1) r = (β − 1)γ + r f + β r p ,
where r f is the risk-free rate, r p is the market risk premium, and γ < 0 measures
the size of the anomaly. The beta anomaly is that a stock has “alpha” and the alpha
falls with its beta.
A few comments on this simple functional form. In terms of the “neutral
point” where equity risk is fairly priced, equation (1) assumes this takes place at an
 https://doi.org/10.1017/S0022109019000322
Downloaded from https://www.cambridge.org/core. IP address: 199.94.10.45, on 13 Dec 2019 at 20:01:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
4 Journal of Financial and Quantitative Analysis

equity beta of 1.0. The equity premium puzzle suggests that, historically, location
might be at a higher level. But, even if equity were undervalued or overvalued
on average, the mean leverage behavior would change, but predictions about the
cross section of leverage, our focus here, would not.
If the beta anomaly extended in equal force into debt, Modigliani–Miller ir-
relevance would still hold. Neither the literature nor our own tests indicate a mean-
ingful beta anomaly in debt, however. In Frazzini and Pedersen (2014), short-
maturity corporate bonds of low risk firms have marginally higher bond-market-
beta risk-adjusted returns, but bond index betas have little relationship to stock
index betas, the basis of the beta anomaly. In fact, Fama and French (1993) show
that stock market betas are nearly identical for bond portfolios of various ratings,
and Baele, Bekaert, and Inghelbrecht (2010) find that even the sign of the corre-
lation/beta between government bond and stock indexes is unstable. Nonetheless,
just to be sure, we directly compared the returns on beta-sorted stock portfolios
with beta-risk-adjusted corporate bond portfolios and can easily reject an inte-
grated beta anomaly. These results are omitted for brevity but available on request.
See Harford, Martos-Vila, and Rhodes-Kropf (2015) for perspectives on financing
under (non-beta-based) debt mispricing.
With a beta anomaly in equity, the overall cost of capital depends not only
on asset beta but on leverage:

(2) WACC(e) = er + (1 − e)rd

= r f + βa r p + (βa − 1)y − (1 − e)(βd − 1)y,

where βa is asset beta and e is capital structure as measured by the ratio of equity
to firm value. The second to last term (the asset beta minus 1 times γ ) is the
uncontrollable reduction in the cost of capital that comes from having high-risk
assets. The last term is the controllable cost of having too little leverage while
debt beta remains low.
An advantage of using the CAPM to develop comparative statics is to see the
familiar textbook transfers of beta risk from equity to debt as leverage increases.
However, any asset pricing model that features a stronger beta anomaly in equity
will lead to the same qualitative conclusions.
B. Optimal Capital Structure
Next we outline a simple, static model of optimal capital structure with no
frictions other than a beta anomaly. There are no taxes, transaction costs, issuance
costs, incentive or information effects of leverage, or costs of financial distress. It
is interesting that unlike other tradeoff approaches, which require one friction to
limit leverage on the low side and another to limit it on the high side, this single
mechanism can drive an interior optimum.
The optimal capital structure minimizes the last term of equation (2) by satis-
fying the first-order condition for e. With the further assumption of a differentiable
debt beta, for a given level of asset beta the optimal capital ratio e∗ satisfies

∂βd [e∗ (βa ), βa ]

 
(3) − y 1 − βd [e (βa )] + [1 − e (βa )]
∗ ∗
= 0,
∂e
 https://doi.org/10.1017/S0022109019000322
Downloaded from https://www.cambridge.org/core. IP address: 199.94.10.45, on 13 Dec 2019 at 20:01:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
Baker, Hoeyer, and Wurgler 5

or, in terms of optimal debt beta,

∂βd [e∗ (βa ), βa ]
βd∗ [e∗ (βa ), βa ] = 1 + [1 − e∗ (βa )] .
∂e
The optimum leverage does not depend on the size of the beta anomaly, but this
is a bit of a technicality. If there were any other frictions associated with leverage,
such as taxes or financial distress costs, the anomaly’s size would become relevant.
Observation 1. Firms will issue some debt, and as much debt as possible as long
as it remains risk-free.
The first-order condition cannot be satisfied as long as the debt beta is 0. At
a zero debt beta, the left side of equation (3) is positive (recall γ < 0). In other
words, issuing more equity at the margin will raise the cost of capital. With zero
debt, the asset beta is equal to the equity beta and the WACC reduces to:

(4) WACC (1) = (βa − 1)y + r y + βa r p .

A first-order Taylor approximation around e = 1 shows that even marginal debt

will decrease the cost of capital:

(5) WACC(e) ≈ WACC(1) + (1 − e)λ < WACC(1).

At first blush, this would seem to deepen the low leverage puzzle of Graham
(2000). One might ask why nonfinancial firms do not increase their leverage ratios
further to take advantage of the beta anomaly: It is initially unclear how the low
leverage ratios of nonfinancial firms represent an optimal tradeoff between the tax
benefits of interest and the costs of financial distress, much less an extra benefit
of debt arising from the mispricing of low risk stocks.
The answer from equation (3) is that many low leverage firms (e.g., the
stereotypical unprofitable technology firm) already start with a high asset beta or
overall asset risk. Their assets are already quite risky at zero debt. Even at modest
levels of debt, meaningful risk starts to be transferred to debt. While equation (3)
cannot on its own explain why a firm would have exactly zero debt, it can explain
why some firms have low levels of debt, despite the tax benefits and modest costs
of financial distress.
Observation 2. Leverage has an interior optimum.
Zero equity is also not optimal. With all debt finance, the debt beta equals
the asset beta and equation (2) reduces to the traditional WACC formula without
the beta anomaly. This establishes that optimum leverage must be interior. The
intuition is that, with the assumption of fairly priced debt, the firm will be fairly
priced if it is funded entirely with debt (i.e. 100% leveraged). Can it increase
value by reducing its leverage ratio? Yes. This new equity, an out-of-the-money
call option, will be high risk, and hence overvalued. As a consequence, neither
0% nor 100% leverage are optimal, so there must be an interior optimum.
To further our understanding of optimal debt levels, we must characterize
the dynamics underlying the transfer of risk from equity to debt with increas-
ing levels of leverage, and in particular the dependence of debt beta on leverage.
 https://doi.org/10.1017/S0022109019000322
Downloaded from https://www.cambridge.org/core. IP address: 199.94.10.45, on 13 Dec 2019 at 20:01:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
6 Journal of Financial and Quantitative Analysis

A natural candidate for the functional form of debt betas is the Merton (1974)
model. Merton uses the isomorphic relationship between levered equity, a Euro-
pean call option, and the accounting identity D = V − E to derive the value of a
single, homogeneous debt claim, such that

(6) D(d, T ) = Be−r f t 8[x2 (d, T )] + V {1 − 8[x1 (d, T )]},

where V is firm value with volatility σ , D is the value of the debt with maturity in
−r f τ
τ and face value B. Let T = σ 2 τ be the firm variance over time, and d ≡ Be V the
debt ratio, where debt is valued at the risk-free rate, thus d is an upward biased
estimate of the actual market based debt ratio (Merton (1974), pp. 454–455). Here,
8(x) is the cumulative standard normal distribution and x1 and x2 are the familiar
terms from the Black–Scholes formula.
Following the approach of Black and Scholes (1973), we arrive at the debt
beta:
V
(7) βd = βa DV .
D
Here DV is the first derivative of the debt value given in equation (6) with respect
to V . In the Merton model, debt is equivalent to a riskless debt claim less a put op-
tion. It follows that the derivative DV is equivalent to the negative of the derivative
of the value of this put option. That is, the derivative (or delta) of the put option
on the underlying firm value is 1 put = −[1 − 8(x1 )], thus Dv ≡ −1 put = 1 − 8(x1 ).
Substituting for equation (2), the debt beta can be written as

(8) βd = X (d)βa ,

where X (d) = d8(x1−8(x 1)

2 )+1−8(x 1 )
is positive. Further, we have that the debt beta is
bounded:

lim{βd = X (d)βa } = 0, and lim {βd = X (d)βa } = βa

d→0 d→∞

in line with the boundary conditions of the debt and in support of the limiting
conditions necessary to establish the claim of an interior optimum leverage.
The factor 8(x1 ) is equivalent to the delta of an equity claim with spot price
equal to V and exercise price equal to face value of the debt B. In light thereof, the
debt beta can be seen to be driven by the increasing value loss in default. If βa > 0
then the debt beta will be continuous and strictly increasing in d. Now rewriting
equation (8),

V − V 8(x1 )
(9) βd = βa ,
D
and following the limits above it can be seen that 0 ≤ V − V 8(x1 ) ≤ D. Conse-
quently, V 8(x1 ) in equation (9) can be interpreted as the conditional expectation
of firm value given it is larger than the face value of debt, times the probability
of the firm value being larger than the face value of debt. This is effectively the
amount of firm risk carried by the debt.
 https://doi.org/10.1017/S0022109019000322
Downloaded from https://www.cambridge.org/core. IP address: 199.94.10.45, on 13 Dec 2019 at 20:01:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
Baker, Hoeyer, and Wurgler 7

On closer inspection, the debt beta in equation (8) can be written, showing its
full functional dependence, βd (d, βa , T ). In our framework, however, the measure
of leverage is not d, but rather the market-based capital ratio, e, that is given by

V −D
(10) e(d, T ) = = 8(x1 ) − d8(x2 ),
V
and is continuous and strictly decreasing in d.
By expressing the debt beta in equation (8) parametrically as a function of
the equity ratio in equation (10) with d as a shared parameter, holding all else
equal, the derivative of the debt beta in equation (3) is equivalent to:

∂βd (e(d, T ), βa , T ) ∂βd (d, βa , T )/∂d

(11) = ,
∂e(d, T ) ∂e(d, T )/∂d

and
∂βd (d, βa , T )/∂d
= −M(d)βa .
∂e(d, T )/∂d
The intuition of M(d) is not important. The first feature of M(d) is that it does
not depend on the asset beta βa itself. So, we can separate out the transfer of
risk (which depends on d and the distribution of firm values characterized by the
inputs of total risk, time to maturity, the riskless rate, and the face value of debt)
from the specific risk transferred:

φ(x1 ) φ(x2 )
8(x1 )8(x2 ) + 8(x2 ) √ + [1 − 8(x1 )] √ − 8(x2 )
σ τ σ τ
(12) M(d) =  .
φ(x2 ) φ(x1 )

[1 − 8(x1 ) + d8(x2 )]2 8(x2 ) − √ + √
σ τ dσ τ

A second feature of M is that it is positive, so the debt beta falls as the capital
ratio increases.
It is worth mentioning that a cost of equity anomaly in which average returns
vary but risk does not (e.g., Internet or blockchain assets are overvalued, or more
generally the type of mispricing described in Baker and Wurgler (2002)) does
not lead to an interior optimum leverage on its own. Without another friction
and without the issuance itself changing valuations, such as downward sloping
demand, one gets a corner solution of all equity if equity is overvalued, or all debt
if equity is undervalued.
Stein (1996) solves this problem by adding costs of financial distress. A mis-
pricing in which risk varies but average returns do not, such as the beta anomaly,
is fundamentally different because the act of levering or delevering itself changes
the risk of the equity, and hence how much it is mispriced. As just shown, overval-
ued firms may now want to issue a small amount of debt, and undervalued firms
will not fund themselves with 100% debt because as leverage rises, their equity is
no longer undervalued. This mechanism is not present in Stein, and allows us to
derive an interior optimum based on a beta anomaly alone.
Observation 3. The optimal leverage ratio is decreasing in asset beta.
 https://doi.org/10.1017/S0022109019000322
Downloaded from https://www.cambridge.org/core. IP address: 199.94.10.45, on 13 Dec 2019 at 20:01:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
8 Journal of Financial and Quantitative Analysis

This is a basic observation and an intuitive, testable prediction. With the

Merton assumptions, and with no frictions other than a beta anomaly in equity,
the optimal level of capital, e∗ in equation (3), can be expressed with equations (8)
and (11) as
1 − βd (d ∗ , βa )
 
1
(13) e∗ (d ∗ , βa ) = 1 −
M(d ∗ ) βa
 
1 1
= 1− − X (d ∗
) .
M(d ∗ ) βa
Holding the positive functions X and M constant (we are varying the asset beta
without changing the overall level of risk) the optimal level of capital is increasing
in asset beta:
∂e∗ (d ∗ , βa ) 1
(14) = ≥ 0.
∂βa M(d ∗ )βa2
We outlined the intuition in the introduction. Under the beta anomaly, beta
is overvalued in equity securities. Ideally, to minimize the cost of capital, risk
is concentrated in equity. This led to the first observation that firms will issue
as much risk-free debt as possible. This lowers the WACC through an increase
in the cost of equity due to its increased beta. Once debt becomes risky, further
increases in leverage, or reductions in equity capital e, have a cost. At that point,
M is greater than 0. Shifting overvalued risk in equity to fairly valued risk in
debt increases the cost of capital. For firms with high-beta assets, this increase is
high even at low levels of leverage. For firms with low-beta assets, this increase
remains low until leverage is high.
More formally, we can show that:
∂βd (e, βa , T )
(15) < 0,
∂e
∂βd (e, βa , T )
≥ 0,
∂βa
∂ 2 βd (e, βa , T )
< 0.
∂βa ∂e
The first 2 partial derivatives of the debt beta with respect to the equity ratio
and the asset beta follow directly from equation (9) and the assumption that
βa > 0. Furthermore, since the asset beta is simply a positive scalar, the cross par-
tial derivative with the equity ratio must have the same sign as the partial deriva-
tive with respect to the equity ratio.
We can now sign the change in the optimal capital ratio as a function of the
underlying asset beta in the more general case. Taking the derivative of e∗ with
respect to asset beta:
de∗ (βa ) ∂βd (βa ) ∂ 2 (e∗ , βa )
 
(16) = − + [1 − e (βa )]
∗
dβa ∂βa ∂e∂βa
∂βd (e (βa ), βa )
∗
∂ 2 βd (e∗ (βa ), βa )
 
× −2 ∗
+ [1 − e (βa )] .
∂e ∂e2
 https://doi.org/10.1017/S0022109019000322
Downloaded from https://www.cambridge.org/core. IP address: 199.94.10.45, on 13 Dec 2019 at 20:01:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
Baker, Hoeyer, and Wurgler 9

The first term is positive using the signs of the partial derivatives in equation (15).
The second term is the second order condition at the optimal leverage ratio defined
in equation (3). It follows from Cargo (1965) that the second order condition is
positive, if and only if both the debt beta in equation (8) and 1 minus the capital
ratio in equation (10) are strictly increasing in d and equation (11) is negative,
which we have already established.
Observation 4. Upside and downside asset beta are equally important in deter-
mining optimal leverage.
Although the mechanism is very different, the predictions so far are intu-
itively consistent with the standard tradeoff theory and associated evidence. This
is encouraging but it is also important to derive a testable prediction that allows
us to empirically distinguish the beta anomaly tradeoff from the costs of financial
distress and the traditional tradeoff. To do this, we make the simple observation
that in the traditional tradeoff, only downside beta, not upside beta, increases the
costs of financial distress and therefore reduces optimal leverage. In contrast, in
the beta anomaly tradeoff, the effect is symmetric. To show this formally, we re-
peat the previous exercise with a version of equation (1) that separates beta into
upside (β + ) and downside (β − ) components:

β + + β − − 1 γ + r f + (β + + β − )r p ,


(17) re =

which we define in the semivariance spirit of Markowitz (1959) and Hogan and
Warren (1974).1
We can now substitute this decomposition of the asset beta into (13),

(18) βa = βa+ + βa− ,

which implies that the optimal capital structure takes the following form:
 
1 1
(19) e∗ (d ∗ , βa ) = 1 − − X (d ∗
) .
M(d ∗ ) βa+ + βa−
This means that, once again holding the positive functions X and M constant, the
optimal level of capital is rising to the same degree in both upside and downside
asset beta, as in equation (14).
Again, there is a straightforward intuition for equation (19). The character
of total risk, as defined in the functions M and X , is what dictates the transfer of
risk from equity to debt. Holding this transfer of risk constant, the optimal level
of capital is increasing in any measure of mispriced asset risk, whether upside or
downside.
Observation 5. A negative empirical relationship between equity beta and lever-
age is sufficient to prove a negative empirical relationship between asset beta and
leverage.
The initial step in our empirical analysis is to confirm a negative relation-
ship between asset beta and leverage. However, to measure asset beta directly, all

β ≡ var1(rm ) n −1 [ j=1 (ri, j − µi )(rm, j − µm )Irm, j>T + j=1 (ri, j − µi )(rm. j − µm )Irm, j≤T ] = β + + β − .
1
Pn Pn
 https://doi.org/10.1017/S0022109019000322
Downloaded from https://www.cambridge.org/core. IP address: 199.94.10.45, on 13 Dec 2019 at 20:01:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
10 Journal of Financial and Quantitative Analysis

liabilities must be traded, which is rarely the case, so the traditional method of
estimating asset beta is simply to assume that debt betas are 0. Observation 5 is
housekeeping that allows us to avoid having to condition a negative leverage-asset
beta link on a zero debt beta assumption. If leverage rises as equity beta falls, then
asset beta must fall too through a simple Modigliani–Miller logic and absolute
priority of returns. If we find that the equity beta is lower and leverage is higher,
then the asset beta must be lower. Thus, a negative relationship between equity
beta and leverage further implies a negative relationship between asset beta and
leverage.
Further Remarks
Like all theoretical work on leverage, our framework clearly makes tradeoffs
between tractability and realism. But, one broader question is what do managers
really need to know, in practice, in order to drive the predictions made above?
In the model, managers must understand how their leverage choice affects their
cost of capital at the margin. In “reality,” a reasonable suggestion is that managers
estimate their cost of equity capital in the process of estimating the fundamental
value of their common stock. The manager’s valuation depends, as in textbooks,
on his or her best guess of future cash flows and the firm’s true, rational cost of
capital.
In such calculations, according to surveys by Graham and Harvey (2001),
“the CAPM is by far the most popular method of estimating the cost of equity
capital: 73.5% of respondents always or almost always use the CAPM.” So, these
calculations amount to an estimate of the cost of capital by CFOs. When the firm
is undervalued according to the CAPM, and their cost of capital is higher than it
ought to be, these firms are presumably less likely to issue equity and more likely
to rely on debt as a marginal source of finance.
The essential result for our purpose is that if the stock market contains a beta
anomaly in which the security market line is empirically too flat, firms with low
asset betas choose higher leverage. This goes through with a weaker assumption
that managers are assessing the value of their firm’s common equity with the
CAPM, as Graham and Harvey (2001) say that they do, as opposed to having full
knowledge of the return generating process and their influence on it.
We might also mention a possible extension to the case of three types of capi-
tal, such as investment grade debt, subordinated or junk debt, and equity, in which
a beta or risk anomaly is present in both investment grade debt and equity, but not
in junk debt. What would be challenging about such an extension is the computa-
tion of the endogenous, maximum quantity of pari passu investment grade debt, as
a function of asset volatility. We conjecture that for many firms there would be a
corner solution at the investment grade boundary (i.e. at the maximum amount of
investment grade debt and zero junk debt). We save such an extension for future
work.
C. A Calibration
Here we assume the model is correct, including its absence of other frictions,
to make some coarse calculations about the value of exploiting the beta anomaly.
To keep things simple, we use the Black–Scholes assumptions and a single
 https://doi.org/10.1017/S0022109019000322
Downloaded from https://www.cambridge.org/core. IP address: 199.94.10.45, on 13 Dec 2019 at 20:01:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
Baker, Hoeyer, and Wurgler 11

liquidation date, 5 years forward, with a contractual allocation of value between

debt and equity, and no costs of financial distress. For each level of leverage,
we compute the value of debt, the value of equity, and the equity beta using the
Merton model. Although this approach is a far cry from structural estimations
of dynamic capital structure models, it may provide a degree of insight into the
potential empirical relevance of the beta anomaly for leverage.
Figure 1 shows the cost of capital and firm value as a function of leverage
for a variety of asset risk levels. In the absence of a beta anomaly, cash flows both
grow and are discounted at the CAPM rate, so firm value is the same, at $10 under
our assumptions, at all risk levels. In the figure we modify the value of equity
using the beta anomaly in equation (1) of γ = −5% per year. This is conservative;
in Baker et al. (2011) sample of U.S. stocks between 1968 and 2008, the difference
in alphas between the top and bottom beta quintile portfolios is −7.5% per year.
Figure 1 shows how a high equity beta uses the anomaly to raise value, while
a low equity beta reduces value in passing it up. Because the only effects here are

FIGURE 1
Value Effects of Leverage When There Is a Beta Anomaly in Equities

In Figure 1, we compute firm value for firms with 5 levels of asset beta. Each firm has a normally distributed terminal value
of 5 years, hence, with a contractual distribution of value between debt and equity and no costs of financial distress or tax
effects. The value of each firm would be exactly $10, regardless of leverage, if there were no low-beta anomaly. Volatility
is equal to asset beta times the sum of a market volatility of 16% plus an idiosyncratic firm volatility of 20%. The risk-free
rate is 2%. We compute the value of equity, the value of debt, and the equity beta under the Merton model with no beta
anomaly. We compound this equity value using the CAPM expected return with a market risk premium of 8% over 5 years.
So, a firm with a beta of 0.25 (beta of 2.0) has a weighted average cost of capital of 4% (18%) in the absence of a beta
anomaly. We then present value this future equity value using the discount rate from equation (1) with a γ of −5%. This
is the adjusted equity value. The weighted average cost of capital uses the adjusted equity value and the value of debt
as weights, the cost of equity from equation (1), and the CAPM for debt. Firm value is the adjusted equity value plus the
value of debt. Leverage is computed using these market values.

Graph A. Weighted Average Cost of Capital Graph B. Absolute Firm Value

14% 14

12% 13
10%
12
Firm Value

8%
WACC

11
6%
10
4%

2% 9

0% 8
0% 20% 40% 60% 80% 100% 0% 20% 40% 60% 80% 100%
Leverage Ratio Leverage Ratio

Graph C. Firm Value Relative to the Maximum

0
Value Relative to Optimum

–4%

–8%

–12%

–16%

–20%
0% 20% 40% 60% 80% 100%
Leverage Ratio
 https://doi.org/10.1017/S0022109019000322
Downloaded from https://www.cambridge.org/core. IP address: 199.94.10.45, on 13 Dec 2019 at 20:01:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
12 Journal of Financial and Quantitative Analysis

through the weighted average cost of capital, with no cash flow effects, a weighted
average cost of capital minimum in Graph A is equivalent to a firm value max-
imum in Graph B. Finally, under a beta anomaly, high asset beta means higher
valuations at any level of leverage. Graph C removes this effect and shows value
relative to the maximum for each level of asset beta. Again, like any calibration
exercise, the conclusions are conditional on the validity of the underlying assump-
tions. What this exercise suggests is that under a stylized model with reasonable
parameters, not exploiting the anomaly can substantially reduce firm value.

III. Empirical Tests

A. Hypotheses
The observations above lead to three hypotheses associated with a beta
anomaly tradeoff. The first is basic, the second separates the beta anomaly tradeoff
from a traditional explanation, and the third provides an aspect of robustness.
Hypothesis 1. Leverage is negatively related to asset beta.
This hypothesis is motivated by Observation 3. Support for this hypothesis
is best viewed as an initial, necessary condition for the beta anomaly tradeoff to
be empirically relevant, in the sense that it cannot on its own rule out alternative
hypotheses based on financial distress. Long and Malitz (1985), among others,
propose this explanation for a negative link between leverage and asset risk. To
provide unique evidence for the beta anomaly channel, we put more weight on the
following hypothesis.
Hypothesis 2. Leverage is negatively related to both upside and downside beta.
Under a standard tradeoff where financial distress reduces the attractiveness
of debt, all else equal, one expects a negative relationship between leverage and
proxies for distress risk, such as, potentially, asset beta. Observation 4 helps us
rule this alternative explanation out.
Observation 4 is based on equation (17) which says that there is a linear
beta anomaly: The slope γ is constant across the full distribution of β, as it is in
the model of Brennan (1993) and as emphasized by Baker et al. (2011). More-
over, if anything, other theories suggests that the beta anomaly might be stronger
in the range of positive β + . Hong and Sraer (2016) focus on the interaction be-
tween overconfidence and high beta, or more speculative, stocks. These stocks
have greater disagreement and, under short sales constraints, are more overvalued.
A richer model might feature greater overconfidence in periods of rising market
returns and in principle deliver a stronger risk anomaly in upside betas.
This prediction is separate from the traditional tradeoff theory and the ideas
in Almeida and Philippon (2007) or Shleifer and Vishny (1992). The costs of fi-
nancial distress depend not only on the unconditional probability of default and
value lost in default, but also when distress occurs and value is lost. If asset beta,
holding all else constant, including total risk, dictates the market state when dis-
tress is likely to occur, then the present value of the costs of financial distress are
higher for assets with higher systematic risk. Almeida and Philippon argue that
risk-adjustment increases the cost of financial distress, while Shleifer and Vishny
 https://doi.org/10.1017/S0022109019000322
Downloaded from https://www.cambridge.org/core. IP address: 199.94.10.45, on 13 Dec 2019 at 20:01:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
Baker, Hoeyer, and Wurgler 13

offer the tangible example of refinancing risk and fire sales. If refinancing risks
and fire sale discounts are higher during market downturns, this would increase
the value lost in distress and lower optimal leverage for firms with higher levels
of systematic risk, although it is still hard to justify zero debt in the presence of
large tax benefits.
Put simply, in the traditional tradeoff, “downside risk” is the emphasis. Risk
matters because of bankruptcy costs (if anything, upside risk actually tends to
increase optimal leverage by increasing expected tax benefits), and this is a plau-
sible effect. For our purpose, the point is that if a beta risk version of the traditional
tradeoff drives an empirical link between high asset beta and leverage, it should
be concentrated in downside risk. Our concern is to distinguish the beta anomaly
tradeoff, in which upside and downside beta are equally relevant and both (most
tellingly, upside beta) should be negatively related to leverage.

Hypothesis 3. Leverage is negatively related to equity beta.

This is based on Observation 5, which notes that a negative empirical rela-

tionship between equity beta and leverage is sufficient to prove a negative empiri-
cal relationship between asset beta and leverage, our main point. Again, the utility
of this prediction is that it allows us to avoid conditioning any negative leverage-
asset beta empirical relationship on the assumption of zero debt beta which, as an
empirical matter, is a common assumption and almost unavoidable. In our tests,
we consider both the financial distress risk alternative explanation and the zero
debt beta assumption at once by relating leverage to upside and downside equity
beta.
B. Data
Our main variables are introduced in Table 1. Our basic sample is the portion
of the merged CRSP-Compustat sample for which marginal tax rates are avail-
able from John Graham. The data begin in 1980, when marginal tax rates are first
available, and end in 2014. They contain 1,038,097 firm-months and span all 50
Fama–French (1997) industries. Unlike much capital structure research, we in-
clude financial firms because they are not special under the beta anomaly theory,
but their exclusion does not affect the relevant results. In an average cross section
there are 2,181 profitable and 291 unprofitable firms.
Variable definitions are in the Appendix. Gross book leverage is long-term
debt and notes payable divided by the sum of long-term debt and notes payable
plus book equity. Net book leverage nets out cash and equivalents from the nu-
merator and denominator. Gross and net market leverage replace book equity with
the market value of common equity from CRSP.
The regressions control for traditional explanatory variables in Bradley,
Jarrell, and Kim (1984), Rajan and Zingales (1995), Frank and Goyal (2009),
Campello and Giambona (2013), and others. The fixed assets ratio, a proxy for fi-
nancial distress costs, is net property, plant, and equipment divided by total assets.
Profitability, which would be positively correlated with leverage under the stan-
dard tradeoff theory but inversely correlated under the Myers and Majluf (1984)
pecking order, is EBIT divided by total assets. Market-to-book assets is known to
be negatively related to leverage, consistent with the need for firms with strong
 https://doi.org/10.1017/S0022109019000322
Downloaded from https://www.cambridge.org/core. IP address: 199.94.10.45, on 13 Dec 2019 at 20:01:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
14 Journal of Financial and Quantitative Analysis

TABLE 1
Summary Statistics: CRSP and Compustat Data

Table 1 reports leverage ratios, asset beta and risk, and capital structure determinants over the period from 1980 to 2014.
We divide firms into profitable and unprofitable. A firm is defined as profitable if it has earnings before interest and taxes
(Compustat = EBITDA) greater than 0. Variable definitions are provided in the Appendix. There are 974,470 observations
in 50 industries across 420 months.
Profitable Firms Unprofitable Firms
Std. Std.
Avg. N Mean Dev. Avg. N Mean Dev.
BOOK_LEV_GROSS% 2,066 32.7 25.6 255 30.1 33.1
MARKET_LEV_GROSS% 2,066 26.8 24.5 255 21.5 25.8
BOOK_LEV_NET% 2,066 21.3 33.4 255 12.9 41.7
MARKET_LEV_ NET% 2,066 18.4 29.2 255 8.9 31.6
ASSET_BETA 2,066 0.71 0.54 255 0.89 0.81
UPSIDE_ASSET_BETA 2,066 0.58 0.56 255 0.58 0.77
DOWNSIDE_ASSET_BETA 2,066 0.83 0.61 255 1.15 0.99
IND_UPSIDE_ASSET_BETA 2,066 0.79 0.30 255 0.88 0.33
IND_DOWNSIDE_ASSET_BETA 2,066 0.84 0.30 255 0.93 0.32
UPSIDE_EQUITY_BETA 2,066 0.66 0.52 255 0.53 0.60
DOWNSIDE_EQUITY_BETA 2,066 0.96 0.49 255 1.06 0.62
PRE_INT_MGL_TAX_RATE% 2,066 33.4 10.3 255 14.0 14.0
FIXED_ASSETS_RATIO (%) 2,066 32.4 23.6 255 24.3 22.4
PROFITABILITY (%) 2,066 9.7 7.6 255 −21.5 20.4
MARKET_TO_BOOK_ASSETS 2,066 1.9 1.9 255 3.1 3.9
ln(ASSETS) 2,066 5.7 2.2 255 3.4 1.8
ASSET_GROWTH (%) 2,066 13.8 29.6 255 3.3 46.2

growth opportunities to avoid having to pass them up (Myers (1977)) as well a

more passive lack of adjustment of leverage to prior stock returns (Welch (2004)).
This variable is gross debt and market equity divided by the sum of gross debt
and book equity. Asset growth could be a proxy for growth opportunities, or cap-
ture size, or the profitability that helps to make debt-financed acquisitions. Size,
the natural log of book assets, may also proxy for multiple influences. Fama and
French (1992) use it to represent the greater cash flow volatility of smaller firms
and their higher expected costs of financial distress. It will also reflect their lesser
access to debt markets. Finally, Graham’s pre-interest marginal tax rates account
for many features of the tax code. As shown by Graham and Mills (2008), these
approximate the tax rates simulated with federal tax return data.
The leverage determinants that interest us most are constructed from stock
returns. Asset beta is unlevered equity beta, assuming debt is riskless. While betas
on corporate debt are very low, they are difficult to observe. To avoid the results
depending on this assumption we have Hypothesis 3. Total equity risk is the stan-
dard deviation of excess stock returns. Asset risk is the unlevered version. Industry
asset beta and risk are market equity-weighted averages.
C. Summary Statistics
Tables 2 and 3 report summary statistics and correlations. Profitable firms
are larger and have higher tax rates. Asset beta is somewhat higher for unprof-
itable firms, as is total risk, which we will use as a control variable for financial
distress costs. With respect to asset risk, a firm must be promising and at least
on a path to profitability to enter the CRSP-Compustat sample for the 24 months
that we require to compute beta. Becoming unprofitable may be associated with
 https://doi.org/10.1017/S0022109019000322
Downloaded from https://www.cambridge.org/core. IP address: 199.94.10.45, on 13 Dec 2019 at 20:01:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
Baker, Hoeyer, and Wurgler 15

TABLE 2
Correlations: CRSP and Compustat Data

Table 2 reports leverage ratios, asset beta and risk, and capital structure determinants over the period 1980–2014.
Variable definitions are provided in the Appendix. There are 974,470 observations in 50 industries across 420 months.
Panel A. Leverage Ratio

Book Leverage Market Leverage

Gross Net Gross Net

BOOK_LEV_GROSS% 1.00
MARKET_LEV_GROSS (%) 0.78 1.00
BOOK_LEV_NET (%) 0.93 0.77 1.00
MARKET_LEV_NET (%) 0.78 0.93 0.86 1.00

Panel B. Leverage Determinants

Own Asset Risk Industry Asset Risk Own Equity Risk

Std. Std. Std.
Beta Up Down Dev. Up Down Dev. Up Down Dev.
ASSET_BETA 1.00
UPSIDE_ASSET_BETA 0.87 1.00
DOWNSIDE_ASSET_BETA 0.93 0.65 1.00
ASSET_RISK (%) 0.58 0.34 0.66 1.00
IND_UPSIDE_ASSET_BETA 0.37 0.33 0.35 0.29 1.00
IND_DOWNSIDEASSET_BETA 0.34 0.27 0.35 0.30 0.91 1.00
IND_ASSET_RISK (%) 0.29 0.22 0.29 0.34 0.75 0.83 1.00
UPSIDE_EQUITY_BETA 0.62 0.82 0.37 −0.01 0.23 0.17 0.10 1.00
DOWNSIDE_EQUITY_BETA 0.63 0.45 0.69 0.18 0.26 0.28 0.18 0.53 1.00
RISK (%) 0.17 0.02 0.25 0.61 0.23 0.27 0.33 −0.03 0.28 1.00
PRE_INT_MGL_TAX_RATE% −0.07 0.01 −0.12 −0.37 −0.12 −0.12 −0.16 0.09 −0.03 −0.47
FIXED_ASSETS_RATIO (%) −0.22 −0.16 −0.23 −0.25 −0.28 −0.25 −0.24 −0.06 −0.10 −0.15
PROFITABILITY (%) −0.02 0.06 −0.07 −0.30 −0.06 −0.06 −0.09 0.09 −0.05 −0.42
MARKET_TO_BOOK_ASSETS 0.19 0.15 0.19 0.18 0.18 0.18 0.17 0.10 0.14 0.18
ln(ASSETS) 0.12 0.28 −0.02 −0.41 −0.11 −0.15 −0.12 0.47 0.16 −0.51
ASSET_GROWTH (%) 0.07 0.05 0.08 0.03 0.03 0.05 0.03 0.05 0.10 0.04

unexpectedly negative returns. Also, firms in variable industries are more likely
to find themselves unprofitable in a given period. The latter logic also applies to
beta, on the downside.
Other notable correlations in Table 2 are as follows: Gross and net leverage
measures are loosely correlated enough to consider both as a robustness exercise.
We follow tradition and consider both book and market leverage measures. Asset
beta, for the own firm or the industry, is negatively correlated with tax rates and
fixed assets and positively correlated with market-to-book and size. These cor-
relations are generally small relative to the correlations among the various risk
measures.
D. Extreme Leverage
Although firms at the leverage extremes are not uncommon, they are particu-
larly interesting to consider in light of the beta anomaly because they are where the
standard tradeoff theory is least compelling. In particular, a beta anomaly trade-
off could help to explain some of the low leverage puzzle of Graham (2000). As
an example, Linear Technology Corporation (NASDAQ: LLTC) produces semi-
conductors with a market capitalization of $7.7 billion as of Dec. 2012. Despite
profitable operations, a pre-interest marginal tax rate of 35% by the methodology
in Graham and Mills (2008), and a cash balance of $1 billion, Linear maintains
negative net debt. One potential explanation for this may be its high asset beta.
 https://doi.org/10.1017/S0022109019000322
Downloaded from https://www.cambridge.org/core. IP address: 199.94.10.45, on 13 Dec 2019 at 20:01:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
16 Journal of Financial and Quantitative Analysis

TABLE 3
Summary Statistics for Profitable Firms: CRSP and Compustat Data

Table 3 presents leverage ratios, asset beta and risk, and capital structure determinants over the period 1980–2014. We
divide the sample of profitable CRSP-Computstat firms into 6 groups, according to gross book leverage (in Panels A
and B) and according to pre-interest marginal tax rate (across 3 pairs of columns). A firm is defined as profitable if it has
earnings before interest and taxes (Compustat = EBITDA) greater than 0. The marginal tax rate is from John Graham,
computed using the methodology of Graham and Mills (2008). Variable definitions are provided in the Appendix.
Tax Rates
Avg. N MTR < 5% Avg. N Middle Avg. N MTR > 30%
Panel A. Low Leverage, <5 % Gross Book Leverage

BOOK_LEV_GROSS% 17 0.7 55 0.9 283 1.0

MARKET_LEV_GROSS% 17 0.7 55 0.8 283 0.7
BOOK_LEV_NET% 17 −20.4 55 −19.9 283 −20.0
MARKET_LEV_ NET% 17 −16.1 55 −17.0 283 −13.9
ASSET_BETA 17 1.22 55 1.14 283 1.05
UPSIDE_ASSET_BETA 17 0.89 55 0.83 283 0.85
DOWNSIDE_ASSET_BETA 17 1.50 55 1.40 283 1.21
IND_UPSIDE_ASSET_BETA 17 0.94 55 0.94 283 0.91
IND_DOWNSIDE_ASSET_BETA 17 0.99 55 0.98 283 0.95
UPSIDE_EQUITY_BETA 17 0.62 55 0.58 283 0.70
DOWNSIDE_EQUITY_BETA 17 1.04 55 0.97 283 0.96
PRE_INT_MGL_TAX_RATE% 17 1.9 55 17.8 283 36.6
FIXED_ASSETS_RATIO (%) 17 22.0 55 21.6 283 22.0
PROFITABILITY (%) 17 4.8 55 6.4 283 14.7
MARKET_TO_BOOK_ASSETS 17 2.82 55 2.13 283 2.98
ln(ASSETS) 17 4.06 55 3.97 283 4.86
ASSET_GROWTH (%) 17 17.6 55 11.3 283 15.1

Panel B. High Leverage, >50 % Gross Book Leverage

BOOK_LEV_GROSS% 19 75.9 78 74.3 371 67.9

MARKET_LEV_GROSS% 19 57.7 78 58.2 371 54.8
BOOK_LEV_NET% 19 71.1 78 69.8 371 63.1
MARKET_LEV_ NET% 19 54.1 78 54.3 371 50.9
ASSET_BETA 19 0.37 78 0.37 371 0.41
UPSIDE_ASSET_BETA 19 0.25 78 0.26 371 0.33
DOWNSIDE_ASSET_BETA 19 0.48 78 0.45 371 0.48
IND_UPSIDE_ASSET_BETA 19 0.75 78 0.73 371 0.68
IND_DOWNSIDE_ASSET_BETA 19 0.81 78 0.79 371 0.73
UPSIDE_EQUITY_BETA 19 0.54 78 0.55 371 0.65
DOWNSIDE_EQUITY_BETA 19 1.03 78 0.97 371 0.97
PRE_INT_MGL_TAX_RATE% 19 2.2 78 17.9 371 37.1
FIXED_ASSETS_RATIO (%) 19 37.3 78 37.2 371 35.3
PROFITABILITY (%) 19 2.1 78 4.2 371 7.7
MARKET_TO_BOOK_ASSETS 19 1.96 78 1.74 371 1.58
ln(ASSETS) 19 5.34 78 5.29 371 6.37
ASSET_GROWTH (%) 19 17.6 78 7.5 371 17.0

While rarer than inexplicably low-leverage firms, a number of profitable

firms maintain high leverage despite little tax benefit. Under a standard tradeoff
theory, this amounts to needlessly tempting a fate of financial distress. An example
is Textainer (NYSE: TGH), a firm that leases and trades marine cargo containers.
As of the end of 2012, its market capitalization was approximately $1.7 billion.
It has tangible assets of $3.4 billion and a cash balance of $175 million. Despite
a marginal tax rate close to 0%, as a result of front-loaded depreciation, modest
growth, and an offshore tax status, it maintains $2.7 billion in debt. A potential ex-
planation for this inconsistency with the standard tradeoff theory is the firm’s low
asset beta. Under a beta anomaly, equity is undervalued at low leverage, and its
value rises steadily as leverage increases to its correct valuation, and potentially
beyond.
The beta anomaly tradeoff may also be pertinent to a set of uniquely highly
leveraged firms (banks) which are often excluded from capital structure analyses.
 https://doi.org/10.1017/S0022109019000322
Downloaded from https://www.cambridge.org/core. IP address: 199.94.10.45, on 13 Dec 2019 at 20:01:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
Baker, Hoeyer, and Wurgler 17

Figure 1 suggests that a beta anomaly in equities means that regulating low asset
beta firms, in the sense of requiring them to delever significantly, can impose
large increases in the cost of capital and losses in shareholder value. Baker and
Wurgler (2015) find that banks’ asset betas are on the order of 0.10, and that
the beta anomaly within banks is at least as large as for all firms. While there are
numerous other forces at play in regulatory debates, the loss of the beta anomaly’s
benefits gives a coherent foundation for bankers’ common argument that reducing
leverage would increase their cost of capital.
Table 3 looks more closely within profitable firms, where we have 867,524
observations and the shortcomings of the standard tradeoff theory appear most
clearly. The panels separate profitable firms into low leverage (gross book lever-
age <5%) and high leverage (gross book leverage >50%) groups. What counts
as high leverage is subjective. We obviously cannot expect a mode at 100% that
resembles the mode at 0%, so we choose an arbitrary cutoff of 50%. The columns
then add another sort into low (MTR<5%), medium, and high (MTR>30%)
marginal tax rate groups.
The low leverage puzzle is represented in the large number of firm-months
that have chosen very low leverage despite positive profitability and high marginal
tax rates. In the average cross section, not far from half of firms in the highest tax
category, 43%, have chosen low leverage over high leverage (43% = 283/(283 +
371)). Firms like Linear Technology are here. The high leverage puzzle, if we
can call it that for sake of illustration, is the narrower but still noticeable fact that
nearly half of the firms in the low tax category, 47%, have chosen high leverage
(47% = 19/(17 + 19)). Firms like Textainer are in this bin.
Of course, this is certainly not the only potential driver of extreme leverage.
For example, Denis and McKeon (2012) explain it as an outcome of the evolu-
tion of operating needs and the desire to maintain financial flexibility (see also
Hackbarth and Mauer (2012)). Hackbarth (2009) suggests a role for managerial
optimism. Regression results follow below, but some initial support for the beta
anomaly tradeoff as an incremental influence comes from the much stronger dif-
ferences in asset risk across the leverage levels. Within the middle tax rate group,
for example, asset betas decline sharply with leverage. Firms with very low lever-
age have a median asset beta of 1.14, a median upside asset beta of 0.83, and
a median downside asset beta of 1.40. For high leverage firms this falls to 0.37,
0.26, and 0.45, respectively.
E. Regressions
Turning from extreme leverage observations to the broader cross section, the
first column in Panel A of Table 4 reports a baseline gross book leverage regres-
sion using typical covariates. We report marginal effects of Tobit regressions that
cluster on both firm and month. The first several variables’ signs and effects are
consistent with prior research, as is the poor overall R 2 . The marginal tax rate has
a positive coefficient, fixed assets a fairly strong positive coefficient, profitability a
negative coefficient, market-to-book a negative coefficient, and size a positive co-
efficient. Rajan and Zingales (1995) focus on the latter 4 variables and obtain the
same results. Asset growth has a positive coefficient, more consistent with the in-
terpretation that asset growth is a consequence of the ability and desire to finance
 https://doi.org/10.1017/S0022109019000322
Downloaded from https://www.cambridge.org/core. IP address: 199.94.10.45, on 13 Dec 2019 at 20:01:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
18 Journal of Financial and Quantitative Analysis

TABLE 4
Capital Structure and Asset Risk (1980–2014)

Table 4 reports Tobit regressions of gross book leverage on capital structure determinants. Gross leverage ratio is defined
as long-term debt (DLTT) plus notes payable (NP) divided by long-term debt plus notes payable plus book equity. Book
equity is computed in the same way as in Kenneth French’s data library. Robust t -statistics, with standard errors clustered
by month and by industry, are in brackets. Regressions labeled ‘‘Own Risk Measures’’ use firm measures of asset beta
and asset risk. Regressions labeled ‘‘Industry Risk Measures’’ use matched Fama–French industry measures of asset
beta and asset risk. Other variable definitions are in the Appendix.
Base Own Risk Measures Industry Risk Measures
Coef. t -Stat. Coef. t -Stat. Coef. t -Stat. Coef. t -Stat. Coef. t -Stat.
Panel A. Asset Beta
ASSET_BETA −11.6 [−13.0] −12.6 [−10.6] −15.8 [−4.5] −9.4 [−2.5]
ASSET_RISK (%) −2.74 [−16.7] −2.65 [−13.3] −2.39 [−2.6] −2.44 [−2.4]
PRE_INT_MGL_TAX_RATE% 0.11 [2.4] −0.11 [−3.0] 0.06 [1.4]
FIXED_ASSETS_RATIO (%) 0.19 [4.1] 0.05 [1.6] 0.14 [3.5]
PROFITABILITY (%) −0.40 [−8.2] −0.46 [−13.5] −0.39 [−8.2]
MARKET_TO_BOOK_ASSETS −1.4 [−5.7] −0.3 [−1.3] −1.1 [−5.0]
ln(ASSETS) 2.8 [8.8] 2.0 [6.4] 2.6 [10.3]
ASSET_GROWTH (%) 0.04 [4.4] 0.06 [9.4] 0.04 [4.7]
2-way clustering Yes Yes Yes Yes Yes
Industries 50 50 50 50 50
Months 420 420 420 420 420
N (000) 974 974 974 974 974

Panel B. Upside and Downside Asset Beta

UPSIDE_ASSET_BETA −6.4 [−10.7] −8.1 [−11.7] −14.7 [−5.2] −12.7 [−3.9]

DOWNSIDE_ASSET_BETA −8.2 [−12.4] −7.5 [−12.1] −0.3 [−0.1] 5.1 [1.5]
UPSIDE_ASSET_RISK (%) −1.71 [−4.7] −0.69 [−1.4] −3.55 [−1.0] 4.84 [1.1]
DOWNSIDE_ASSET_RISK (%) −3.15 [−6.3] −4.06 [−8.2] −3.07 [−0.9] −12.22 [−3.0]
Controls Yes Yes Yes Yes Yes
2-way clustering Yes Yes Yes Yes Yes
Industries 50 50 50 50 50
Months 420 420 420 420 420
N (000) 974 974 974 974 974

with debt, determined by other underlying sources, than an additional proxy for
growth opportunities.
1. Leverage and Asset Beta
We now add risk measures to test Hypothesis 1. Our special focus is on as-
set beta, which is what the beta anomaly tradeoff suggests, but we also control
for overall risk. In principle, any effect of total asset risk could reflect the beta
anomaly tradeoff; some explanations of the beta anomaly are specific to beta, oth-
ers are not. Total asset risk is a plausible proxy for the expected costs of financial
distress, especially compared to asset beta.
The middle columns of Panel A show that asset beta is a strong determinant
of leverage, consistent with Hypothesis 1. This is true controlling for overall asset
risk (and in unreported univariate regressions). Adding the control variables does
not significantly affect the coefficient or t-statistic on asset beta. With all controls,
a 1-unit increase in asset beta reduces leverage by 12.6%, a large effect by the
standards of the control variables.
The last columns of Table 4 show that the economic effects remain if we use
industry risk, which is an empirical solution to the issue of any mechanical nega-
tive link between leverage and asset beta created by using leverage itself to unlever
equity beta. It appears that any measurement error introduced by this switch does
 https://doi.org/10.1017/S0022109019000322
Downloaded from https://www.cambridge.org/core. IP address: 199.94.10.45, on 13 Dec 2019 at 20:01:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
Baker, Hoeyer, and Wurgler 19

not appear to greatly affect the coefficients on beta risk. These regressions provide
further support for Hypothesis 1.
2. Upside and Downside Asset Beta
We turn next to Hypothesis 2. It is clear that high asset beta is associated
with lower leverage. This is consistent with the beta anomaly tradeoff whereby
the cost of equity for high beta assets is lower and so less debt is optimal, but
also consistent with versions of the standard tradeoff theory to the extent that the
relationship is driven by downside risk. To examine whether downside risk alone
is driving the relationship, we estimate equity beta separately over months when
the market risk premium was positive and when it was negative. Unlevering these
and averaging by industry gives us an upside asset beta and downside asset beta
measure.
In Panel B of Table 4 we find that the upside and downside components of
asset beta have an equally strong relationship to gross book leverage when using
the firm’s own risk measures. Furthermore, the relationship actually favors the
upside asset beta under the preferred industry-based measures. Downside asset
beta has no statistical association with leverage and in one case a point estimate
of the wrong sign to support a version of the traditional tradeoff.
Reassuringly for the traditional tradeoff, and consistent with our use of asset
risk as a control variable for financial distress, the downside overall asset risk as-
sociation with leverage is generally stronger than the upside asset risk association.
This of course does not alter the conclusions relevant to Hypothesis 2, a prediction
involving asset beta risk. To repeat, our goal is not to cast doubt on the relevance
of the traditional tradeoff theory, but to show that there is a connection between
leverage and asset beta that is less tortured to explain by the beta anomaly.
Support for Hypotheses 1 and 2 also appears under other leverage measures,
including gross market leverage and net book and market leverage, still using
industry risk measures. Panel A of Table 5 shows results consistent with only a
beta anomaly tradeoff, in the form of an upside beta effect, remain strong, as well
as results consistent with both a beta anomaly tradeoff and a traditional tradeoff,
remain in the form of downside risk effects.
3. Upside and Downside Equity Beta
We now test Hypothesis 3, which addresses the required practical assump-
tion of a zero debt beta. While debt betas in practice are very low, Hypothesis 3,
which is based on the theoretical observation that a negative empirical relation-
ship between leverage and equity beta is sufficient to prove a negative empirical
relationship between leverage and asset beta, gives us a more grounded way to
avoid conditioning our asset beta results on this assumption.
The results in Panel B of Table 5 confirm a negative relationship between
leverage and equity beta. The result that upside beta risk is much stronger than
downside beta risk remains. It is worth noting that any mechanical link between
leverage and equity beta would bias results in a positive direction (i.e., away from
our hypotheses of interest), so it is particularly comforting that the results remain
consistent with the beta anomaly tradeoff.
 https://doi.org/10.1017/S0022109019000322
Downloaded from https://www.cambridge.org/core. IP address: 199.94.10.45, on 13 Dec 2019 at 20:01:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
20 Journal of Financial and Quantitative Analysis

TABLE 5
Alternate Leverage Ratios and Equity Beta (1980–2014)

Table 5 reports Tobit regressions of leverage on capital structure determinants. We repeat the regression of the third
column of Panel B in Table 4, using 4 different measures of leverage. Net leverage ratios deduct cash and equivalents
from debt. Market leverage ratios replace book equity with market capitalization, equal to price times shares outstanding
from CRSP. Panel B replaces asset measures of beta and risk with equity measures of beta and risk. Robust t -statistics,
with standard errors clustered by month and by industry, are in brackets.
Gross Leverage (%) Net Leverage (%)
Book Market Book Market
Coef. t -Stat. Coef. t -Stat. Coef. t -Stat. Coef. t -Stat.
Panel A. Upside and Downside Asset Beta

UPSIDE_ASSET_BETA −8.1 [−11.7] −8.4 [−12.0] −13.1 [−11.3] −12.6 [−13.7]

DOWNSIDE_ASSET_BETA −7.5 [−12.1] −8.0 [−13.3] −16.5 [−12.9] −16.3 [−15.5]
UPSIDE_ASSET_RISK (%) −0.69 [−1.4] −1.05 [−2.7] −3.99 [−7.8] −4.50 [−10.4]
DOWNSIDE_ASSET_RISK (%) −4.06 [−8.2] −3.36 [−7.9] −8.70 [−12.9] −7.86 [−15.9]
Controls Yes Yes Yes Yes
2-way clustering Yes Yes Yes Yes
Industries 50 50 50 50
Months 420 420 420 420
N (000) 974 974 974 974
Panel B. Upside and Downside Equity Beta

UPSIDE_EQUITY_BETA −7.3 [−7.4] −8.1 [−9.6] −10.3 [−6.4] −10.6 [−8.1]

DOWNSIDE_EQUITY_BETA 1.5 [1.5] 0.5 [0.7] 0.5 [0.4] −0.1 [−0.1]
UPSIDE_EQUITY_RISK (%) 5.80 [9.0] 4.84 [8.5] 6.61 [7.9] 5.51 [8.1]
DOWNSIDE_EQUITY_RISK (%) −0.99 [−1.5] −0.45 [−0.8] −1.03 [−1.3] −0.48 [−0.8]
Controls Yes Yes Yes Yes
2-way clustering Yes Yes Yes Yes
Industries 50 50 50 50
Months 420 420 420 420
N (000) 974 974 974 974

IV. Other Robustness Tests

We find qualitatively identical results using a variety of other asset beta es-
timation approaches. Acharya, Almeida, and Campello (2013) use a structural
model to estimate beta. They kindly shared their data, and we also replicated their
methodology in our full data set and jointly estimated asset volatility with as-
set beta. Choi (2013) uses an empirical measure of asset betas by combining the
returns of fixed income instruments with equity weighted by their market val-
ues. These approaches all lead to qualitatively identical results to our simple and
transparent estimate of asset betas. The detailed results are available on request.
Finally, in additional unreported robustness tests suggested by referees, we
found that switching to quarterly data (instead of clustering monthly), or expand-
ing the sample by dropping the requirement to include Graham’s marginal tax
rates, had little effect. Dropping total risk or equity risk in Tables 4 and 5 also had
no qualitative impact.

V. Conclusion
Many studies have shown that high-risk equities do not earn commensurately
high returns. This paper derives a novel explanation for leverage that is consistent
with this observation. We show that for firms with relatively risky assets, the cost
of capital is minimized at a low level of leverage. For firms with very low risk
assets, low leverage entails a substantial cost in the form of issuing undervalued
 https://doi.org/10.1017/S0022109019000322
Downloaded from https://www.cambridge.org/core. IP address: 199.94.10.45, on 13 Dec 2019 at 20:01:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
Baker, Hoeyer, and Wurgler 21

equity, and hence the cost of capital is minimized at much higher levels of lever-
age. In the data, leverage is indeed inversely related to systematic risk, supporting
the main prediction of the beta anomaly tradeoff.
Importantly, we derive and test a prediction of the beta anomaly tradeoff that
allows us to separate it from a standard tradeoff explanation in which asset beta
reflects financial distress costs. The beta anomaly tradeoff predicts that leverage is
inversely related to upside risk, not just downside risk. This is also confirmed. In
some specifications, the relationship between leverage and upside risk is stronger
than the relationship with downside risk.
More broadly, we suggest that the beta anomaly tradeoff may help to explain
the low and high leverage puzzles in which leverage choices cannot be easily ex-
plained by tax and financial distress considerations alone. Better understanding
the empirical limits of a beta anomaly tradeoff, and how it complements other re-
alistic explanations for leverage that receive support in the literature, is an area for
future research. Moreover, while the beta anomaly is perhaps the obvious starting
point, fruitful research might also come from investigating the interplay of capital
structure and numerous other stock market anomalies.

Appendix. Variable Definitions

All variables are winsorized at 1% and 99% as measured across the whole sample.
ASSET BETA: EQUITY BETA times 1 minus net market leverage.
ASSET GROWTH: The annual change in total assets (ASSETS) divided by total assets 1
year ago, in percentage terms.
ASSET RISK (%): EQUITY RISK times 1 minus net market leverage.
BOOK EQUITY: Shareholder’s equity minus preferred stock plus deferred taxes. Share-
holder’s equity (SEQ) or the sum of common equity (CEQ) plus preferred stock
(PSTK) if shareholder’s equity is missing or total assets (ASSETS) minus total li-
abilities (LT) if common equity is missing. Preferred stock is equal to the redemp-
tion value of preferred stock (PSTKRV) or the liquidating value of preferred stock
(PSTKL) or total preferred stock (PSTK) in that order and is set to 0 if still miss-
ing. Deferred taxes are equal to deferred tax and investment tax credit (TXDITC) or
balance sheet deferred tax (TXDB) in that order and 0 if missing.
BOOK LEV GROSS%: The sum of total long-term debt (COMPUSTAT = DLTT) and
notes payable (NP) divided by the sum of total long-term debt and notes payable
and book equity, in percentage terms.
BOOK LEV NET%: The sum of total long-term debt (COMPUSTAT = DLTT) and notes
payable (NP) less cash and equivalents (CHE) divided by the sum of total long-term
debt and notes payable and book equity less cash and equivalents, in percentage
terms.
DOWNSIDE ASSET BETA: DOWNSIDE EQUITY BETA times 1 minus net market
leverage.
DOWNSIDE ASSET RISK (%): DOWNSIDE EQUITY RISK times 1 minus net market
leverage.
DOWNSIDE EQUITY BETA: Market beta computed from CRSP returns (RET) net of
Treasury bill returns (YLDMAT) from CRSP regressed on the value-weighted mar-
ket return (VWRET), also net of the Treasury bill return, times an indicator variable
 https://doi.org/10.1017/S0022109019000322
Downloaded from https://www.cambridge.org/core. IP address: 199.94.10.45, on 13 Dec 2019 at 20:01:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
22 Journal of Financial and Quantitative Analysis

when the realized market return is positive and the value-weighted market return
times an indicator variable when the realized market return is negative, using 3-day
overlapping return windows. We require at least 750 overlapping windows of re-
turns and use at most 5 years of returns. The downside beta is the coefficient on
negative market returns.
DOWNSIDE EQUITY RISK (%): Standard deviation of CRSP returns (RET) net of
Treasury bill returns (YLDMAT), in percentage terms, conditional on the CRSP
return (RET) net of the Treasury bill return being negative.
EQUITY BETA: Market beta computed from CRSP returns (RET) net of Treasury bill
returns (YLDMAT) from CRSP regressed on the value-weighted market return
(VWRET), also net of the Treasury bill return, using 3-day overlapping return win-
dows. We require at least 750 overlapping windows of returns and use at most 5
years of returns.
EQUITY RISK (%): Standard deviation of CRSP returns (RET) net of Treasury bill re-
turns (YLDMAT), in percentage terms.
FIXED ASSETS RATIO (%): Plant, property, and equipment, net (PPENT) divided by
total assets (ASSETS), in percentage terms.
IND ASSET BETA: Market equity weighted average asset beta, computed for each
Fama–French industry classification. Market equity is equal to price (PRC) times
shares outstanding (CRSP) from CRSP. The 49 industry classifications are defined
in Ken French’s data library, with unclassified firms comprising a 50th group.
IND ASSET RISK (%): Market equity weighted average asset risk, computed for each
Fama–French industry classification. Market equity is equal to price (PRC) times
shares outstanding (CRSP) from CRSP. The 49 industry classifications are defined
in Kenneth French’s data library, with unclassified firms comprising a 50th group.
ln(ASSETS): The natural log of total assets (ASSETS).
MARKET TO BOOK ASSETS: Sum of total long-term debt (COMPUSTAT = DLTT)
and notes payable (NP) and market equity divided by the sum of total long-term
debt and notes payable and book equity. Market equity is equal to price (PRC) times
shares outstanding (CRSP) from CRSP.
MARKET LEV GROSS%: The sum of total long-term debt (COMPUSTAT = DLTT)
and notes payable (NP) divided by the sum of total long-term debt and notes payable
and market equity. Market equity is equal to price (PRC) times shares outstanding
(CRSP) from CRSP, in percentage terms.
MARKET LEV NET%: The sum of total long-term debt (COMPUSTAT = DLTT) and
notes payable (NP) less cash and equivalents (CHE) divided by the sum of total
long-term debt and notes payable and market equity less cash and equivalents. Mar-
ket equity is equal to price (PRC) times shares outstanding (CRSP) from CRSP, in
percentage terms.
PRE-INT MGL TAX RATE%: John Graham provided estimates of the pre-interest
marginal tax rate, computed using the methodology of Graham and Mills (2008),
in percentage terms.
PROFITABILITY%: Earnings before interest and taxes (EBIT) divided by total assets
(ASSETS), in percentage terms.
UPSIDE ASSET BETA: UPSIDE EQUITY BETA times 1 minus net market leverage.
UPSIDE ASSET RISK%: UPSIDE EQUITY RISK times 1 minus net market leverage.
 https://doi.org/10.1017/S0022109019000322
Downloaded from https://www.cambridge.org/core. IP address: 199.94.10.45, on 13 Dec 2019 at 20:01:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
Baker, Hoeyer, and Wurgler 23

UPSIDE EQUITY BETA: See DOWNSIDE BETA. The upside beta is the coefficient on
positive market returns.
UPSIDE EQUITY RISK%: Standard deviation of CRSP returns (RET) net of Treasury
bill returns (YLDMAT), in percentage terms, conditional on the CRSP return (RET)
net of the Treasury bill return being positive.

References
Acharya, V.; H. Almeida; and M. Campello. “Aggregate Risk and the Choice between Cash and Lines
of Credit.” Journal of Finance, 68 (2013), 2059–2116.
Almeida, H., and T. Philippon. “The Risk-Adjusted Cost of Financial Distress.” Journal of Finance,
62 (2007), 2557–2586.
Baele, L.; G. Bekaert; and K. Inghelbrecht. “The Determinants of Stock and Bond Return Comove-
ments.” Review of Financial Studies, 23 (2010), 2374–2428.
Baker, M.; B. Bradley; and R. Taliaferro. “The Low-Risk Anomaly: A Decomposition into Micro and
Macro Effects.” Financial Analysts Journal, 70 (2014), 43–58.
Baker, M.; B. Bradley; and J. Wurgler. “Benchmarks as Limits to Arbitrage: Understanding the
Low-Volatility Anomaly.” Financial Analysts Journal, 67 (2011), 40–54.
Baker, M., and J. Wurgler. “Market Timing and Capital Structure.” Journal of Finance, 57 (2002),
1–32.
Baker, M., and J. Wurgler. “Do Strict Capital Requirements Raise the Cost of Capital? Bank Regula-
tion, Capital Structure, and the Risk Anomaly.” American Economic Review, 105 (2015), 315–320.
Bali, T.; S. Brown; S. Murray; and Y. Tang. “A Lottery-Demand-Based Explanation of the Beta
Anomaly.” Journal of Financial and Quantitative Analysis, 52 (2017), 2369–2397.
Black, F.; M. C. Jensen; and M. Scholes. “The Capital Asset Pricing Model: Some Empirical Tests.”
Studies in the Theory of Capital Markets. M. C. Jensen, ed. New York, NY: Praeger (1972),
79–121.
Black, F., and M. Scholes. “The Pricing of Options and Corporate Liabilities.” Journal of Political
Economy, 81 (1973), 637–659.
Bradley, M.; G. Jarrell; and E. H. Kim. “On the Existence of an Optimal Capital Structure: Theory and
Evidence.” Journal of Finance, 39 (1984), 857–878.
Brennan, M. “Agency and Asset Pricing” Working Paper, University of California at Los Angeles
(1993).
Campello, M., and E. Giambona. “Real Assets and Capital Structure.” Journal of Financial and
Quantitative Analysis, 48 (2013), 1333–1370.
Cargo, G. T. “Comparable Means and Generalized Convexity.” Journal of Mathematical Analysis and
Applications, 12 (1965), 387–392.
Choi, J. “What Drives the Value Premium? The Role of Asset Risk and Leverage.” Review of Financial
Studies, 26 (2013), 2845–2875.
Denis, D., and S. B. McKeon. “Debt Financing and Financial Flexibility: Evidence from Proactive
Leverage Increases.” Review of Financial Studies, 25 (2012), 1897–1929.
Fama, E., and K. R. French. “The Cross-Section of Expected Stock Returns.” Journal of Finance, 47
(1992), 427–465.
Fama, E., and K. R. French. “Common Risk Factors in the Returns on Stocks and Bonds.” Journal of
Financial Economics, 33 (1993), 3–56.
Fama, E., and K. R. French. “Industry Costs of Equity.” Journal of Financial Economics, 43 (1997),
153–193.
Frank, M. Z., and V. K. Goyal. “Capital Structure Decisions: Which Factors Are Reliably Important?”
Financial Management, 38 (2009), 1–37.
Frazzini, A., and L. H. Pedersen. “Betting Against Beta.” Journal of Financial Economics, 111 (2014),
1–25.
Graham, J. “How Big Are the Tax Benefits of Debt.” Journal of Finance, 63 (2000), 1901–1941.
Graham, J., and C. Harvey. “The Theory and Practice of Corporate Finance: Evidence from the Field.”
Journal of Financial Economics, 60 (2001), 187–243.
Graham, J., and L. Mills. “Simulating Marginal Tax Rates Using Tax Return Data.” Journal of
Accounting and Economics, 46 (2008), 366–388.
Hackbarth, D. “Determinants of Corporate Borrowing: A Behavioral Perspective.” Journal of Corpo-
rate Finance, 15 (2009), 389–411.
Hackbarth, D., and D. C. Mauer. “Optimal Priority Structure, Capital Structure, and Investment.”
Review of Financial Studies, 23 (2012), 747–796.
 https://doi.org/10.1017/S0022109019000322
Downloaded from https://www.cambridge.org/core. IP address: 199.94.10.45, on 13 Dec 2019 at 20:01:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
24 Journal of Financial and Quantitative Analysis

Harford, J.; M. Martos-Vila; and M. Rhodes-Kropf. “Corporate Financial Policies in Misvalued Credit
Markets.” Working Paper, University of Washington (2015).
Hogan, W., and J. Warren. “Toward the Development of an Equilibrium Capital-Market Model Based
on Semivariance.” Journal of Financial and Quantitative Analysis, 9 (1974), 1–11.
Hong, H., and D. Sraer. “Speculative Betas.” Journal of Finance, 71 (2016), 2095–2144.
Long, M. S., and I. Malitz. “Investment Patterns and Financial Leverage.” Corporate Capital Struc-
tures in the United States. B. M. Friedman, ed. Chicago, IL: University of Chicago Press (1985),
325–352.
Markowitz, H. “Portfolio Selection.” Efficient Diversification of Investments. New York, NY: Wiley
(1959).
Merton, R. C. “On the Pricing of Corporate Debt: The Risk Structure of Interest Rates.” Journal of
Finance, 29 (1974), 449–470.
Myers, S. “Determinants of Corporate Borrowing.” Journal of Financial Economics, 5 (1977),
147–175.
Myers, S., and N. Majluf. “Corporate Financing and Investment Decisions When Firms Have
Information That Investors Do Not Have.” Journal of Financial Economics, 13 (1984), 187–221.
Rajan, R., and L. Zingales. “What Do We Know about Capital Structure? Some Evidence from
International Data.” Journal of Finance, 50 (1995), 1421–1460.
Shleifer, A., and R. Vishny. “Liquidation Values and Debt Capacity: A Market Equilibrium Approach.”
Journal of Finance, 47 (1992), 1343–1366.
Stein, J. “Rational Capital Budgeting in an Irrational World.” Journal of Business, 69 (1996), 429–455.
Welch, I. “Capital Structure and Stock Returns.” Journal of Political Economy, 112 (2004), 106–131.