Financial Leverage, Corporate Investment, and Stock

Returns
Ali K. Ozdagli

November 16, 2010
Abstract
This paper rationalizes empirical patterns of market leverage, book leverage, book-to-
market ratios, and stock returns across different book-to-market portfolios, using a model of
rm nancing and investment. The model shows analytically that tax-deductibility of inter-
est payments increases effective investment irreversibility and that investment irreversibility
weakens the relationship between book-to-market values and returns. This provides a clear
and novel mechanism showing how nancial leverage affects stock returns beyond the stan-
dard Modigliani-Miller paradigm. The paper argues that operating leverage or investment
irreversibility alone cannot generate the cross-sectional stock return patterns, and that market
leverage is the main source of the value premium.

Federal Reserve Bank of Boston, Ali.Ozdagli@bos.frb.org. This paper is a modied and improved version of my
Ph.D. thesis with the same title. I am grateful to Fernando Alvarez, Lars Hansen, Anil Kashyap, and Robert Lucas
for their support. I also thank Federico Diez, Francois Gourio, Satadru Hore, Tim Johnson, Sergey Koulayev, Oz
Shy, Christina Wang, and the seminar participants at the Federal Reserve Bank of Chicago, Federal Reserve Bank of
Boston, Boston University, AEA 2010 Atlanta meetings and The University of Chicago, in particular, Gene Fama and
Jarda Borovicka, for helpful comments. Sarojini Rao provided excellent research assistance. The usual disclaimer
applies.
1
Firms with a high ratio of book value of equity to market value of equity (value rms) earn higher
expected stock returns than rms that have a low book-to-market equity ratio (growth rms). How-
ever, as Grinblatt and Titman (2001) point out, conventional wisdom tells us that growth options
should be riskier than assets-in-place.
1
Therefore, as Zhang (2005) stresses, growth rms, which
derive their value from growth options, should have higher expected stock returns than value rms,
which derive their value from assets-in-place.
In addition, Fama and French (1992) show that portfolios of stocks with different book-to-
market ratios have similar riskiness as measured by the standard Capital Asset Pricing Model
(CAPM) of Sharpe (1964), Lintner (1965), and Black (1972). This phenomenon, known as the
“value premium puzzle,” helped the Fama and French model replace the CAPM as the benchmark
in the asset pricing literature.
This paper aims to explain the differences between the stock returns of value rms and growth
rms. For this purpose, I extend the investment irreversibility model of Abel and Eberly (1996)
by incorporating investors' risk preferences, risk-free debt contracts, and debt adjustment costs.
With this framework, I show that market leverage, dened as debt-to-assets ratio using market
values, can explain the major share of the value premium, while investment irreversibility without
leverage generates a growth premium rather than a value premium. This nding is different from
the recent literature that emphasizes investment irreversibility as the reason for the value premium,
such as Zhang (2005) and Cooper (2006). However, investment irreversibility is still an important
1
From p. 392 of Grinblatt and Titman (2001): "Consider Wal-Mart, for example. The value of this rm's assets
can be regarded as the value of the existing Wal-Mart outlets in addition to the value of any outlets that Wal-Mart may
open in the future. The option to open new stores is known as a growth option. Because growth options tend to be
most valuable in good times and have implicit leverage they contain a great deal of systematic risk."
2
ingredient that improves the model's t with the data, by generating a wide range of book-to-market
values.
The nancing decisions in this model are similar, but not identical, to those of Fischer, Heinkel,
and Zechner (1989) and Gomes and Schmid (2010). These papers assume that existing debt is
retired at the time of the new issues and add capital restructuring costs to the standard tradeoff
theory of capital structure whereby a rm chooses its nancing policy by balancing the costs of
bankruptcy against the benets of debt, such as tax shields due to interest payments. My paper
also assumes that rms benet from the tax shield of debt as in the tradeoff theory and that they
face additional costs at the time of debt restructuring.
However, in this paper, debt has two properties distinct from its properties in previous papers:
it is risk-free and endogenously limited by the lenders to a certain fraction of capital. Since interest
payments are tax deductible, the rm prefers debt nancing to equity nancing and would rather
have an innite amount of debt. But this would lead to negative equity value in some states where
the rm would rather go bankrupt than pay its debt. Therefore, for debt to remain risk-free, lenders
will limit the amount of debt. They can accomplish this by accepting the resale value of capital as
collateral and ensuring that this value is not lower than the amount of debt, so that they can recover
their money in case of bankruptcy.
2
Alternatively, lenders may limit the amount of debt in order to
ensure that the market value of equity is always non-negative so that bankruptcy is suboptimal for
the rm. I show that the market value of equity is strictly positive when the debt capacity equals
the resale value of capital. Therefore, the collateralized debt agreement is more restrictive than
2
This is a common assumption in the papers that model risk-free debt. A recent example is Livdan, Sapriza, and
Zhang (2009).
3
the debt agreement that induces a non-negative market value of equity and the suboptimality of
bankruptcy. So, the latter policy provides the rm with a higher debt capacity and the rm prefers
this latter debt policy, while the lenders are indifferent.
One important property of the model is that the rm's book leverage, that is, the fraction of total
capital supplied by lenders, is not state-dependent. The book leverage is determined in a manner
that ensures that the rmvalue is non-negative even in the worst-case scenario, to avoid bankruptcy.
I show that this worst-case scenario is independent of the state variables, and hence a revision of
the debt agreement at a later date would lead to the same level of leverage. Thus, it is not optimal
for a rm to change its book leverage once it is set. As a result, the book leverage remains the
same across rms with different ratios of book-to-market equity, whereas market leverage differs
signicantly. Figure 1 plots averages of book and market leverage within different book-to-market
portfolios and provides support for this argument. Moreover, because the level of debt is constant
in the inaction region (when the rm does not invest), the rm's market debt-to-equity ratio varies
closely with uctuations in its own stock price. This implication of the model is in line with the
results of Welch (2004), who nds that U.S. corporations do little to counteract the inuence of
stock price changes on their capital structures.
My analysis shows that investment irreversibility without leverage causes a growth premium
rather than a value premium in the model. The rm's investment opportunity is a call option,
because the rm has the right, but not the obligation, to buy a unit of capital at a predetermined
price. As we know from the nancial options literature, when the price of the underlying security
rises and falls, the price of the call option rises and falls at a greater rate. This suggests that the value
4
Figure 1: Market and book leverage across book-to-market (BE/ME) portfolios. This gure
presents book leverage and market leverage across different book-to-market value portfolios. The
numbers on the horizontal axis give the average book-to-market equity value in each portfolio. The
numbers on the vertical axis give the average market and book leverage in each portfolio. At the
beginning of each year portfolios are formed on the basis of ranked values of BE/ME, where each
portfolio covers a decile. Book equity (BE) is the book value of common equity (item 60) plus
deferred taxes and investment tax credit (item 35) from annual COMPUSTAT les. Market value
of equity (ME) is price of each share times the number of shares outstanding from CRSP database.
Book leverage is total assets (COMPUSTAT item 6) minus book equity divided by total assets.
Market leverage is total assets minus book equity divided by total assets minus book equity plus
market equity. The accounting data for all scal year-ends in calendar year t 1 is matched with
market equity at the end of December of year t 1, as in Fama and French (1992). Source: The
Center for Research in Security Prices, CRSP-COMPUSTAT merged database, from 1963 to 2008
for t.
5
of a growth option, that is, the call option to invest, should be more responsive to economic shocks
than the assets-in-place. Therefore, growth options increase the riskiness of the rm.
3
Similarly,
the disinvestment opportunity is a put option, because the rm has the right, but not the obligation,
to sell a unit of capital at a predetermined price. The value of this put option is negatively related to
the value of the underlying asset, because the gain from exercising it is higher for less productive
rms. Therefore, the disinvestment option provides the value rms that have low productivity with
insurance against downside risk and hence reduces their riskiness. This proposition is different
from the results of recent literature, for example, Zhang (2005) and Cooper (2006), which presents
investment irreversibility as the source of the value premium.
4
In my model, nancial leverage affects stock returns directly, through its effect on equity risk as
in Modigliani and Miller (1958), and indirectly, through its effect on business risk, by inuencing
investment decisions. I nd that these two channels have opposing effects on the relationship
between book-to-market ratios and stock returns. However, the Modigliani-Miller effect strongly
dominates the investment channel and explains the major share of the value premium.
The Modigliani-Miller (1958) effect of debt comes from the fact that the book-to-market ratio
and market leverage are closely related when book leverage is constant, as we observe within the
context of this model and in Figure 1. In particular, if we let 11. `1, 11, and `1 be the book
3
This pattern is in line with Bernardo, Chowdhry, and Goyal (2007) who show that the beta of growth opportunities
is greater than the beta of assets-in-place for the unlevered rm.
4
Hall (2004) estimates the adjustment cost parameter for capital in a quadratic adjustment cost model without
debt and nds that adjustment costs are relatively small and are not an important part of the explanation of the large
movements in company values.
6
value of equity, market value of equity, book leverage, and market leverage, respectively, we have
11
`1
=
`1
1 `1
1 11
11
. (1)
Because book leverage is constant across value and growth rms, this equation implies that, in the
model, market leverage is the main source of variation in book-to-market values and that value
rms have higher market leverage than growth rms. Therefore, they have greater equity risk
according to the Modigliani and Miller theorem.
Financial leverage also affects investment and hence the business risk in the model, because
it inuences the effective degree of investment irreversibility faced by the owners of the rm.
When investment can be nanced with leverage, the effective price of capital is reduced by the tax
savings associated with debt nancing at the time of investment. On the other hand, at the time
of disinvestment, the rm has to pay back its debt, in line with the debt agreement, and therefore
has to give up the tax savings associated with the debt nancing of that particular investment.
Because the purchase price is greater than the resale price and both should be adjusted by the same
value of tax savings, their ratio increases as a result of debt nancing. This increases the effective
irreversibility perceived by the owners of the rm. Since irreversibility reduces the value premium,
so, too, does the investment channel of leverage.
The risk-free debt assumption is not critical, at least qualitatively, for the main results of the
paper; in particular it is not critical to understanding how market leverage affects stock returns
through the Modigliani and Miller channel and the investment channel, as the intuition for these
7
channels does not depend on the risk-free debt assumption. The Modigliani and Miller channel
simply states that equity of rms with higher market leverage is riskier and the investment chan-
nel intuition is based on the tax advantage of debt. The Modigliani and Miller paradigm and tax
deductibility of interest payments would still be present in a model with risky debt. The relation-
ship between market leverage and book-to-market value would also be unaffected as long as book
leverage does not vary signicantly across rms in different portfolios but rather obeys the data
as shown in Figure 1. Nevertheless, the capital structure choice in my model has two advantages.
First, it provides consistency between theoretical and empirical analysis. Because we do not ob-
serve the market value of debt, many studies that relate risky debt to returns use the book value
of debt as a proxy for the market value of debt. However, this contradicts the assumption of risky
debt because the value of risky debt is not equal to the book value of debt. Second, in a tradeoff
model with risky debt, more productive rms, which also have lower book-to-market values, tend
to choose higher debt relative to their total capital because they have lower bankruptcy probabil-
ity for a given debt level. This implies that book leverage would not be constant across different
book-to-market portfolios, in contrast with the data in Figure 1.
This paper is closely related to the growing literature that links corporate decisions to asset
returns. Like Zhang (2005) and Cooper (2006), Carlson, Fisher, and Giammarino (2004) have a
model with investment irreversibility but link the value premium to operating leverage rather than
investment irreversibility. Gomes and Schmid (2010) link leverage and growth options to asset
returns when issuing equity is costly and Livdan, Sapriza, and Zhang (2009) look at the effect of
exogenous risk-free debt capacity on stock returns.
8
My paper contributes to this literature in many ways. First, unlike Livdan, Sapriza, and Zhang
(2009), the risk-free debt capacity of the rm is endogenously determined. Second, the closed-
form solution of the model identies explicitly how investment irreversibility, nancial leverage,
and their interaction affect the cross-section of stock returns. Gomes and Schmid (2010) emphasize
that nancial leverage affects growth options, which in turn affect stock returns, creating a channel
beyond the Modigliani and Miller paradigm. However, they do not focus on how leverage changes
investment decisions and how this channel affects the relationship between book-to-market val-
ues and stock returns. My paper provides a clear and novel mechanism showing that nancial
leverage increases effective investment irreversibility and that investment irreversibility weakens
the relationship between stock returns and book-to-market values in a model that incorporates not
only a growth option but also a disinvestment option. Third, the paper provides a novel calibration
exercise that ts the model to the distribution of book-to-market values via maximum likelihood
and shows that the model captures the empirical patterns of book leverage, market leverage, book-
to-market values, and stock returns across different book-to-market portfolios reasonably well.
5
Fourth, the paper provides both theoretical and empirical analysis consistent with the claim that
market leverage can explain the major portion of the value premium.
Finally, the mechanism that relates nancial leverage to stock returns through the investment
channel in the model suggests two new testable implications: Because nancial leverage increases
effective investment irreversibility through the interest tax-shield, rms with a greater interest tax-
5
To the best of my knowledge, no other paper in the literature makes an effort to match the distribution of book-
to-market values and leverage, although this distribution is important in generating the cross-sectional distribution of
returns. The implications of omitting this fact are crucial and are discussed in the Calibration section.
9
shield invest as if they faced greater investment irreversibility.
6
Moreover, the relationship between
investment irreversibility and stock returns in the model suggests that industries with a higher
degree of investment irreversibility, after taking leverage and the interest tax shield into account,
should have a weaker relationship between book-to-market values and stock returns.
The next section introduces data and variables, and presents supporting evidence that favors
market leverage over other channels in explaining the value premium. The second section presents
the problem of the rm in a continuous time setting. The third section then discusses the optimal
investment policy and the market value of equity. The fourth section presents optimal nancing
policy and its relationship with investment. The fth section links stock returns with investment
irreversibility and nancial leverage. The sixth section presents the calibration of the model and
the comparison of simulation results with the data. The seventh section provides a discussion of
the results and reconciles evidence from earlier literature with the paper's claims. The last section
concludes.
I. Data and Supporting Evidence
The choice of data closely follows Fama and French (1992). The only exception is that I include
all rms in the NYSE, AMEX and NASDAQ in the monthly CRSP database, whereas Fama and
French (1992) include only nonnancial rms. Many recent papers, such as Carlson, Fisher, and
Giammarino (2004) and Zhang (2005), use the data series published on Kenneth French's website
and, unlike the data in Fama and French (1992), these data series include all rms. Therefore, in-
6
The effect of irreversibility on investment is discussed by Abel and Eberly (1994, 1996) in detail.
10
cluding all rms makes my results comparable to previous literature, although excluding nancial
rms does not change any result signicantly. As in Fama and French (1992), this dataset is then
merged with the annual balance sheet les from Compustat database. The dataset covers the period
1962 to 2008.
The creation of variables also closely follows Fama and French (1992). Accordingly, I match
the accounting data for all scal year-ends in calendar year t 1 (1962-2007) with the returns for
the period starting July of year t to June of year t + 1. Moreover, I use a rm's market equity at
the end of December of year t 1 to compute its book-to-market and market leverage. Market
value of equity is dened as the price of each share times the number of outstanding shares from
the CRSP database. Book value of equity is dened as the sum of book value of common equity
(COMPUSTAT code CEQ/ item number 60) plus deferred taxes and and investment tax credit
(COMPUSTAT code TXDITC/ item number 35) from Compustat database.
7
Therefore, the book
value of debt is equal to total assets (COMPUSTAT code AT/ item number 6) minus book value
of equity, and book leverage is equal to book value of debt divided by total assets. Finally, as in
Gomes and Schmid (2010), market leverage is equal to book value of debt divided by the sum of
book value of debt and equity. The dates of the numerator and denominator in the book-to-market
ratio and market leverage are not aligned, and the implications of this are discussed in Fama and
French (1992).
In order to support my claim that market leverage is the main reason for the value premium, I
unlever equity returns using the Modigliani-Miller relationship :
¹
= `1 :
1
+ (1 `1) :
1
7
Fama and French (1992) add only deferred taxes, whereas Fama and French (1993) add both deferred taxes and
investment tax credit, a practice I follow.
11
where :
¹
is the return on total assets (i.e., unlevered returns), `1 is the market leverage, :
1
is
the return on debt, and :
1
is the return on equity. For return on debt I use monthly returns on the
T-Bill, taken from CRSP. The purpose of this exercise is to see what is left over from the value
premium after accounting for the direct effect of nancial leverage. The results are provided in
Figure 2. I have also used an alternative measure of return on debt, given by the ratio of the interest
expense to total debt. The results are similar.
Figure 2 shows that the major chunk of the value premium vanishes after this unlevering ex-
ercise, which is in line with the paper's claim that market leverage is the main reason behind the
value premium. In particular, note that the value premium between the top and bottom deciles re-
duces from 16% to 4%. The result is even more drastic if we compare the value premium between
the second decile and the top decile; the value premium between these deciles reduces from 14%
to 2% after unlevering. Hecht (2001) performs a similar analysis using market values for bond
data, and a new working paper by Choi (2010, Table VI, panel A) uses both bond and loan data
to unlever returns. Their analysis also shows that unlevered returns vary much less than levered
returns across book-to-market portfolios.
Of course, both my approach and the approaches of Choi and Hecht have their shortcomings:
My unlevering exercise calculates market leverage by replacing the market value of debt with the
book value of debt and uses the risk-free rate for the return on debt. Hecht's and Choi's approaches
use market values, but bonds and bank loans are traded infrequently in the secondary markets and
they do not consider accounts payable as a source of debt. In addition, their sample includes rms
for which they have bond and loan data covering only half of the Compustat rms and a shorter
12
Figure 2: Stock returns and unlevered returns across book-to-market portfolios. This gure
presents actual and unlevered stock returns across different book-to-market value portfolios. The
numbers on the vertical axis give the time series average of equally weighted cross-sectional
means of returns for each portfolio in annualized percentages. At the beginning of each year
portfolios are formed on the basis of ranked book-to-market values, where each portfolio covers
a decile. Book-to-market value, market leverage, and book leverage are dened in the caption
of Figure 1. The accounting data for all scal year-ends in calendar year t 1 is matched with
market equity at the end of December of year t 1 and the returns for the period starting July
of year t to June of year t + 1, as in Fama and French (1992). The stock returns come from the
monthly CRSP series and the unlevered return for each month is calculated using the weighted
average of stock returns and monthly T-Bill returns from CRSP, where the weight of the T-Bill
return is market leverage. Source: The Center for Research in Security Prices (CRSP), CRSP-
COMPUSTAT merged database, from July 1963 to June 2008.
13
time period. Nevertheless, it is striking that all these approaches yield to similar conclusions.
This exercise also provides evidence that favors market leverage channel over the operating
leverage channel in explaining the value premium. Operating leverage is the leverage of the all-
equity nanced (unlevered) rm, as illustrated in Carlson, Fisher, and Giammarino (2004). There-
fore, operating leverage cannot claim to capture more than a small portion of the value premium
left over from this unlevering exercise.
II. The Model
My model extends the investment irreversibility model of Abel and Eberly (1996) with corporate
taxes, debt, and a stochastic discount factor to capture investors' risk preference. While debt ca-
pacity and investment and nancing decisions are endogenous, investors' preferences are captured
by an exogenous discount factor, as in Zhang (2005), Cooper (2006), and Carlson, Fisher, and
Giammarino (2004), among others.
The rms choose their investment and nancing policy in order to maximize the market value
of equity. Investment is subject to partial irreversibility, that is, the purchase price of one unit
of capital is 1 and the resale price is j < 1. Each rm produces output at time t using capital,
1
t
, and takes the level of productivity, A
t
, and the stochastic discount factor of investors, o
t
, as
14
exogenously given. Both A
t
and o
t
follow geometric Brownian motions
dA
t
A
t
= j
A
dt + o
¹
dn
¹
+ o
i
dn
i
= j
A
dt + odn (2)
do
t
o
t
= :dt o
S
dn
¹
. (3)
where 1
t
[do
t
,o
t
] = :dt is the interest rate and o
S
is the price of risk. The Brownian increments
dn
¹
and dn
i
represent systematic and idiosyncratic shocks and are independent of each other.
They can be aggregated using o =
_
o
2
i
+ o
2
¹
and dn = (o
i
,o) dn
i
+ (o
¹
,o) dn
¹
. Moreover, if
we let l
t
and 1
t
denote total capital purchases and total capital sales, respectively, up to time t, we
can write the net change in the stock of capital as
d1
t
= dl
t
d1
t
. (4)
where dl
t
0 and d1
t
0.
The net income of the rm is given by the operating cash ows net of the cost of maintenance
and cash ows to debtholders plus tax shields from maintenance and interest payments:
:
t
: (1
t
. A
t
. /
t
) = (1 t)
_
/
1 ¸
A
¸
t
1

t
o1
t
:/
t
1
t
_
. (5)
where t is the tax on corporate income, / 0 is the productivity multiplier, and 0 < ¸ < 1
is the returns-to-scale parameter of the production function. On the cash outow side, o is the
maintenance cost per unit of capital, : is the risk-free rate on debt, and /
t
is the the ratio of the
15
book value of debt to the book value of assets, or book leverage. In accordance with tax laws,
the net income incorporates the tax deductiblity of interest payments. Except for taxes, debt, the
investors' stochastic discount factor, and maintenance costs, all the assumptions are the same as in
Abel and Eberly (1996). The assumptions regarding maintenance cost and the stochastic discount
factor are similar to those in Carlson, Giammmarino, and Fisher (2004). The tax structure is similar
to that in Miao (2005), among others. I also assume, for the sake of simplicity, that the accounting
salvage value of the capital is the same as the actual salvage value, so that the rm does not pay
any tax on the resale price of capital, as in Carlson, Giammmarino, and Fisher (2004), Gomes and
Schmid (2010), and Miao (2005), among others.
I model nancial leverage as risk-free debt extended through a credit line, where the debthold-
ers agree to nance a certain fraction of operating capital. Intuitively, the lenders can keep the debt
risk-free by a collateralized debt agreement and can limit the amount of debt by the resale price of
capital so that j /. Alternatively, they can set a limit on debt that guarantees that the rm always
has non-negative market equity and hence honors its debt rather than going bankrupt. The rm has
the option to renegotiate this fraction later, but debt restructuring requires that the existing debt be
retired altogether and that new debt be issued at a cost proportional to the amount of new debt, c,
as in Fisher, Heinkel, and Zechner (1989).
As a result of this debt agreement, the rm will invest when the marginal value of capital to
equity holders is 1 /, as this is the fraction of new investment that should be nanced with
equity. Moreover, the rm will disinvest when the marginal value of capital is j /, because the
rm gets j for each unit of capital sold but has to give back / to debtholders in order to keep the
16
book leverage constant, in accord with the debt agreement. In the following analysis, A
l
(1. /)
and 1
l
(A. /) denote the investment boundary along which the marginal value of capital is 1 /,
while A
1
(1. /) and 1
1
(A. /) denote the disinvestment boundary along which the marginal value
of capital is j /.
8
These two boundaries enclose the inaction region, where the marginal value
of capital is between 1 / and j / and net investment is zero. This investment policy will be
discussed in more detail in the next section.
Because the lenders are indifferent between a collateralized debt agreement and the no-bankruptcy
debt agreement, it is up to the rm which agreement to choose. The following proposition estab-
lishes that the rm will never go bankrupt under a collateralized debt agreement.
Proposition 1 The market value of equity is strictly positive if debt is limited by the resale price
of capital.
Proof. We have j / if debt is limited by the resale price of capital. The market value of
equity is bounded by (j /) 1 0, because this is what the shareholders will get after pay-
ing the lenders if they decide to dissolve the rm. Let J (A. 1. /) be the market value of equity
and 1
l
(A. /) be the investment boundary. Then J (A. 1. /) (j /) 1 must hold at the in-
vestment boundary, 1
l
(A. /), since otherwise the rm would dissolve immediately, leaving the
shareholders with capital (j /) 1 in return to their investment (1 /) 1 (j /) 1. There-
8
Because the investment boundary is a surface in the (X; K; b) space, we can express it either as K
U
(X; b) or as
X
U
(K; b). The same holds for the disinvestment boundary.
17
fore, J (A. 1
l
(A. /) . /) 0. Finally, we can write the market value of equity as
J (A. 1. /) = J (A. 1
l
(A. /) . /) +
_
1
1
U
(A,b)
J
1
(A. /. /) d/. (6)
where 1 / J
1
(A. 1. /) j / 0, because the marginal value of capital is bounded
due to investment and disinvestment options. We also have 1 1
l
(A. /) for any given A and /,
because the rm will invest to prevent the value of capital from falling below1
l
(A. /). Therefore,
the integral on the right side of this equation should be positive. Since the sum of the two positive
terms is positive, we have J (A. 1. /) (j /) 1 0, and this completes our proof.
This proposition essentially tells us that even if the rm had the option to go bankrupt, it
would never exercise this option if debt were limited by the resale price of capital, because the
disinvestment boundary would be hit before bankruptcy became optimal. It follows immediately
that the debt agreement with a no-bankruptcy condition is less restrictive. In particular, it provides
a greater debt limit because the market value of equity is still non-negative if the lenders lend the
rm more than the resale value of its capital.
9
Since bankruptcy is suboptimal under both lending
policies, I omit bankruptcy in the rest of the paper.
9
The choice of the rm between risk-free debt policies does not affect the results of the paper qualitatively. How-
ever, the condition < b creates a greater degree of irreversibility by increasing the wedge between the investment
and disinvestment boundaries and hence improves the t of the model-implied book-to-market distribution to data.
18
The rm maximizes shareholder value by choosing its investment and nancing plans:
J (1
t
. A
t
. /
t
) = 1
t
max
fol
t+s
,o1
t+s
,ob
t+s
g
_
1
0
o
t+c
o
t
[ :
t+c
d: (1 /
t+c
)dl
t+c
+ (j /
t+c
) d1
t+c
]
+
1

c2fc:ob
t+s
6=0g
o
t+c
o
t
[d/
t+c
c(/
t+c
+ d/
t+c
)] 1
t+c
. (7)
where the term d/
t+c
is the change in book leverage after debt adjustment, and
_
1
0
dl
t+c
and
_
1
0
d1
t+c
are Stieltjes integrals. Note that the stochastic discount factor does not appear as a state
variable in the value function J, because o
t+c
,o
t
is log-normally distributed with parameters :t
and o
2
S
t and this distribution does not depend on any state variable. The rst term, the integral, in
equation (7) is the discounted value of after-tax prots net of cash inows and outows through
the purchase and resale of capital. The second term in equation (7) is the discounted value of debt
adjustment costs.
The debt limit imposed by the lenders adds an additional constraint to the problem. If debt is
limited by the resale value of capital, then this constraint is simply /
t+c
j. If, on the other hand,
debt is limited by the no-bankruptcy condition then we have
0 J (1
t+c
. A
t+c
. /
t+c
) for all 1
t+c
. A
t+c
. /
t+c
. (8)
Although the debt policy with the no-bankruptcy condition provides a greater debt capacity than
the collateralized debt policy, we still cannot rule out the latter until I show that debt nancing is
preferred to equity nancing. Proposition 3 in section IV serves this purpose.
19
Because of investment and debt adjustment costs, it is not optimal for the rm to adjust capital
and debt frequently. The Hamilton-Jacobi-Bellman equation (HJB) in the inaction region, where
the rm does not make any adjustments, is given by
:J (1. A. /) = : (1. A. /) + jAJ
A
(1. A. /) +
1
2
o
2
A
2
J
AA
(1. A. /) . (9)
where j = j
A
o
S
o
¹
is the risk-adjusted drift of the productivity process. This is practically the
same as substituting the stochastic discount factor with the risk-free rate and taking expectations
under the risk-neutral measure. When we divide both sides of this equation by the market value
of equity, J, this equation tells us that the required rate of return from buying the rm should
be equal to the dividend yield (the rst term) plus expected capital appreciation under the risk-
adjusted productivity process (the second and third terms). The solution of the partial differential
equation (9) requires that we specify some boundary conditions, the so-called value matching and
smooth pasting conditions. These conditions, presented in the appendix, guarantee the continuity
and optimality of the value function.
10
10
Dixit (1993) is a good introduction to the derivation of these conditions.
20
III. Optimal Investment Policy and the Valuation of Equity
Since equation (9) holds identically in 1, we can take the derivative of both sides with respect to
1 to get
:J
1
(1. A. /) = :
1
(1. A. /) + jAJ
1A
(1. A. /) +
1
2
o
2
A
2
J
1AA
(1. A. /) . (10)
Because all terms in the rm's problem are homogeneous of degree one in A and 1, the value
of the rm should also be homogeneous of degree one in A and 1. As a result, the marginal
value of capital should be homogeneous of degree zero in A and 1. Therefore, we can dene
the ratio of productivity to capital, ¸ A,1, and the marginal value of capital to shareholders,
¡ (¸. /) J
1
(1. A. /), to express the last equation as
:¡ (¸. /) =


¸
: + j¸¡
j
(¸. /) +
1
2
o
2
¸
2
¡
jj
(¸. /) . (11)
where

/ = (1 t) / and :(/) = (1 t) (o + :/) is the tax adjusted marginal cost of maintenance
and nancing. Then, the boundary conditions at the upper and lower investment bounds are given
by the following equations.
11
¡ (¸
l
(/) . /) = 1 / and ¡
j

l
(/) . /) = 0 (12)
¡ (¸
1
(/) . /) = j / and ¡
j

1
(/) . /) = 0. (13)
11
We can see that the additional smooth pasting conditions for b, that is, q
b
(y
L
(b) ; b) = q
b
(y
L
(b) ; b) = 1; are
automatically satised once we take the derivative of the value-matching equations with respect to b and apply the
smooth pasting conditions for y. Therefore, we omit these conditions for the rest of the analysis.
21
X
(productivity)
K (capital)
Slope = y
U
(b)
Slope = y
L
(b)
disinvestment
region
investment
region
inaction region
b q b − < < − 1 η
b q − =1
b q − =η
Figure 3: Projection of investment and inaction regions on the K-X plane. The line with slope
¸
l
(/) gives the investment boundary where the marginal value of capital, ¡, is equal to 1 /, the
net cash outow per unit purchase of capital, where / is book leverage. The line with slope ¸
1
(/)
gives the disinvestment boundary where the marginal value of capital, ¡, is equal to the net cash
inow per unit sale of capital, j /, where j is the resale value capital. These two boundaries
enclose the inaction region for investment, where the marginal value of capital is bounded by 1 /
and j /.
This reduces the original HJB equation to an ordinary differential equation; solving this involves
nding two constants of integration and the boundary values for ¸. Figure 3 displays the projection
of the investment and inaction regions implied by the boundary conditions on the (1. A) plane.
The appendix shows that solving these equations and integrating the marginal value of capital
leads to
J (1. A. /) =

HA
¸
1

+

1
1
(/) A
c
P
1
1c
P
+

1
.
(/) A
c
N
1
1c
N

:(/)
:
1. (14)
22
where c
1
1 ¸ 0 c
.
, and

1
1
(/) and

1
.
(/) are functions of book leverage that take
only positive values. The four terms are the value of assets in place (before costs), growth options,
disinvestment options, and the present value of maintenance and nancing costs. Note that book
leverage, /, is constant in the inaction region due to debt adjustment costs. Therefore, any effect
of debt adjustment on the market value of equity should be captured by constants of integration

1
1
(/) and

1
.
(/) . and hence equation (14) does not have a separate term for the debt adjustment
option.
IV. Financial Leverage and Investment
We now turn our attention to optimal nancing policy and its relationship with investment. The fol-
lowing proposition shows that the tax advantage of leverage makes the rm choose its investment
policy as if it faced greater irreversibility. Then I show that the optimal nancing policy for the
rm is to exhaust its debt capacity. Therefore, the rm prefers the no-bankruptcy condition because
it provides greater debt capacity. Finally, I show that the debt capacity under the no-bankruptcy
condition is independent of state variables.
Proposition 2 When interest payments are tax deductible, the gap between the investment bound-
ary and the disinvestment boundary as measured by G(/) ¸
l
(/) ,¸
1
(/) increases with book
leverage /.
Proof. See appendix
23
Intuitively, the gap between the investment and disinvestment boundaries increases as the ratio
of the purchase to the resale price increases, because it is this discrepancy between prices that cre-
ates investment irreversibility. Then, we should answer why the purchase price increases relative to
the resale price. When investment is nanced with leverage, the shareholders care not only about
the actual price of capital but also about the nancing costs. In the absence of tax deductibility of
interest payments, there would be no net gain from borrowing because the amount of investment
nanced through borrowing would be paid back to lenders through interest. When there is tax
deductibility, the rm still pays interest to lenders but it has to pay less tax to the government. At
the time of investment, the net purchase price of capital from the shareholders' perspective is the
actual price net of any tax savings due to debt nancing. At the time of disinvestment, the net
resale price of capital is the actual price minus the loss of tax deductions due to debt repayment.
Since the purchase and resale prices of capital decrease by the same amount of tax savings, their
ratio should increase. This increases the effective irreversibility perceived by the shareholders.
The next two propositions show that the optimal behavior for the rm is to use all of its debt
capacity at once if the cost of issuing debt is sufciently small so that the tax savings due to
interest payments dominate the cost of debt nancing and never to adjust its book leverage after
that. I assume that the cost of issuing debt is below this limit.
12
Proposition 3 Let G(/) ¸
l
(/) ,¸
1
(/) 1. Then, there is a critical level for the cost of issuing
12
The parameter estimates for the calibration of the model imply that c

is 2.3 percent, which is above the level used
in the literature. For example, Fischer, Heinkel, and Zechner (1989) use 1 percent for the corresponding parameter in
their model.
24
debt, given by
c

= t
_
1
1
1 c
.
_
0 (15)
below which the rm strictly prefers debt to equity.
Proof. See appendix
Proposition 4 It is never optimal to readjust debt.
Proof. The appendix shows that J
b
(A. 1. /) + 1 0. Therefore, the smooth pasting conditions
required at debt restructuring are not satised.
The intuition for this proposition follows from the envelope theorem. Section A. of the appen-
dix shows that J
b
(A. 1. /) + 1 = 0 should hold at the debt adjustment boundary for debt adjust-
ment to be optimal and discusses how this condition can be interpreted as an envelope condition.
In particular, if the before-adjustment book leverage, /, changes by a small amount, d/, the amount
of debt the rm has to repay at the time of adjustment increases by (d/)1. If debt adjustment
were optimal, i.e., if debt adjustment were to maximize the market value of equity, J(A. 1. /),
then the shareholders should have been compensated by the same amount, which would imply that
J
b
(1. A. /)d/ + 1d/ = 0 or J
b
(1. A. /) + 1 = 0 at the debt adjustment boundary. Because
J
b
(1. A. /) + 1 0 for all (A. 1. /), this optimality condition is violated, i.e., debt adjustment
cannot be an optimal policy, for any choice of adjustment boundary.
The following proposition shows that the leverage limit set by the debtholders is the same for
all rms regardless of their productivity and capital levels.
Proposition 5 The debt limit implied by the no-bankruptcy condition is not state-dependent.
25
Proof. Using equation (14) we can write the no-bankruptcy condition J (1. A. /) 0 as
J (1. A. /) ,1 = J (1. ¸. /) 0, that is,
J (1. ¸. /) = H¸
¸
+ 1
1
(/) ¸
c
P
+ 1
.
(/) ¸
c
N

:
:
0. (16)
Moreover, J (1. ¸. /) should be increasing in ¸ because, given capital and leverage, more productive
rms should have higher market value, that is, J
A
(1. A. /) 0. Therefore, J (1. A. /) 0 for
all (A. 1. /) if and only if J (1. ¸
1
(/) . /) 0. As a result, the debt limit is given by the equation
J (1. ¸
1
(/) . /) = 0, whose solution is independent of state variables.
This proposition tells us that the book leverage in this model should not be state-dependent
because the debt limit is determined by the worst-case scenario, which is not state-dependent due
to the homogeneity of the rm's value. This result is important for two reasons: First, it strengthens
the result that it is not optimal to adjust debt once it is set, because it is costly to adjust and the new
limit would be the same as the old one. Second, because the debt limit as a fraction of total capital
is the same for all rms, book leverage is the same across rms with different book-to-market
ratios. Figure 1 shows that this implication of the model ts the data.
Propositions 4 and 5 also imply that the book leverage is constant over time. This simplies
the analysis of the model. However, we can easily generalize the model with time-varying book
leverage: Suppose that the resale value of capital, j, is changing over time. The debt limit can be
either set via / < j or via the no-bankruptcy condition, so that in each case it can be conditioned
on the resale value of capital. Then, since the homogeneity of the problem in A and 1 still holds,
26
the investment boundaries would be given by A,1 = ¸
l
(/. j) and A,1 = ¸
1
(/. j). Following
the proof of proposition 5, the no-bankruptcy debt limit would be given by J (1. ¸
1
(/. j) . /) = 0,
implying that the debt limit is a function only of the resale value, / (j). As a result, although book
leverage is time varying with j, it would still be constant across different book-to-market portfolios
without changing the other properties of the model.
V. Stock Returns
A. Investment Irreversibility and Stock Returns
In order to isolate the pure effect of investment irreversibility on stock returns, I focus on a rm
that does not have any operating costs and nancial leverage. In this case, the market value of
rm's equity is given by
J (1. A) = HA
¸
1

+ 1
1
A
c
P
1
1c
P
+ 1
.
A
c
N
1
1c
N
. (17)
where c
1
1 ¸ 0 c
.
and H,1
1
, and 1
.
are positive constants. These three terms
capture the market value of the assets-in-place, the growth options, and the disinvestment options,
which I denote J
¹1
. J
G
, and J
1
. respectively.
Using Ito calculus and some algebra, we can derive the (conditional) expected excess stock
27
return as
1
dt
1 (d1) : =
1
dt
1
_
:dt + dJ
J
_
: = o
S
o
¹
J
A
A
J
= o
S
o
¹
_
J
¹1
J
¸ +
J
G
J
c
1
+
J
1
J
c
.
_
= o
S
o
¹
(:
¹1
¸ + :
G
c
1
+ :
1
c
.
) . (18)
Therefore, the expected excess stock return is a value-weighted average of expected excess returns
that come from the three sources of value. Since the book-to-market ratio can be expressed as
(1 /) 1,J (1. A) = (1 /) ,J (1. ¸), the ratio of productivity to capital, ¸, is negatively related
to the book-to-market ratio. The appendix shows that the expected stock return increases in ¸
and hence decreases in book-to-market values, producing a growth premium rather than a value
premium.
The result presented in this section is intuitive, once we realize the similarities of growth and
disinvestment options with nancial options. The rm's investment opportunity is a call option
because the rm has the right, but not the obligation, to buy a unit of capital at a predetermined
price. As we know from the nancial options literature, as the price of the underlying security rises
and falls, the price of the call option rises and falls at a greater rate than the underlying security,
because the strike price of the call option acts like an implicit leverage. This suggests that the value
of the growth option, that is, the call option to invest, should be more responsive to protability
shocks, and hence riskier, than the assets-in-place. This is captured by c
1
¸ in this model.
This result is consistent with the nding of Bernardo, Chowdry, and Goyal (2007) that the beta of
28
growth opportunities is greater than the beta of assets-in-place for the unlevered rm. As a result,
growth rms, which derive their value mainly from growth options, should have higher expected
returns than value rms.
Similarly, the disinvestment opportunity is a put option, because the rm has the right, but not
the obligation, to sell a unit of capital at a predetermined price. The value of this put option is
negatively related to the value of the underlying asset, because the gain from exercising the option,
that is, from disinvestment, is higher for less productive rms. Therefore, the disinvestment option
provides the value rms with insurance against downside risk and hence reduces their riskiness. In
this model, this is captured via c
.
< 0.
This result is in contrast with the intuition of several recent papers, such as Zhang (2005) and
Cooper (2006), that present investment irreversibility as the source of the value premium. These
papers argue that investment adjustment costs make it harder for value rms to deploy their excess
capital when the economy faces bad shocks, whereas growth rms do not face the same problem,
as they do not have too much excess capital. As a result, assets-in-place should be riskier than
growth options and hence value rms should be riskier than growth rms. However, these papers
include xed operating costs in the prot function of the rm, which would affect the business risk,
possibly creating a value premium, as Carlson, Fisher, and Giammarino (2004) suggest.
13
To show that the relationship between investment irreversibility and the cross-section of stock
returns does not depend on the assumptions of the model, the following proposition generalizes the
13
Zhang (2005) provides (in Table IV) a sensitivity analysis that shows that a 10 percent reduction in xed operating
costs reduces the difference between stock returns of the rms in the highest and lowest book deciles by 1 percent.
If we assume that the elasticity of return differences to xed costs is constant, eliminating operating leverage should
lead to a 10 percent decrease in the value premium between the highest and lowest deciles and hence would nullify the
stock return differences.
29
argument that growth options are riskier than assets-in-place, by providing a proof that does not
depend on the properties of the adjustment cost or on the properties of the processes for productiv-
ity or the stochastic discount factor. The proposition focuses on total irreversibility of investment
because, if irreversibility were the main reason for the value premium, it would create the greatest
value premium if rms were not able to disinvest.
Proposition 6 In the absence of leverage and under perfect investment irreversibility, growth op-
tions are riskier than assets-in-place.
Proof. See Appendix
It follows from this proposition that growth rms that derive their value from growth options
should have higher expected returns, so we have a growth premium rather than a value premium
under investment irreversibility without leverage. Despite the negative relationship of the value
premium and investment irreversibility, I keep irreversibility in my model because it is useful to
generate a wide range of book-to-market values and market leverage that improves the model's t
with the data.
B. Financial Leverage and Stock Returns
Using Ito calculus and the Hamilton-Jacobi-Bellman equation (9) from the previous section, we
can write the expected excess stock return as
1 [d1
i
] :dt =
: (1
i
. A
i
) dt + 1 [dJ (1
i
. A
i
)]
J (1
i
. A
i
)
:dt = o
S
o
¹
A
i
J
iA
(1
i
. A
i
)
J
i
(1
i
. A
i
)
dt. (19)
30
where o
S
is the price of risk and o
¹
A
i
J
iX
(1
i
,A
i
)
J
i
is the risk exposure. The appendix shows that we
can rewrite this as
1
dt
1 (d1
i
) : =
_
1 +
:(/) ,:1
J
_
(¸:
¹1
+ c
1
:
G
+ c
.
:
1
) o
S
o
¹
. (20)
where the rst factor captures the Modigliani-Miller effect and the second factor decomposes the
total business risk (as if the rm were entirely nanced by equity) into assets-in-place, a growth
option, and a disinvestment option.
Financial leverage affects expected returns in two ways. The rst effect, the Modigliani-Miller
channel, is obvious in equation (20). Firms with higher market leverage, /1,J, also have higher
book-to-market values, (1 /) 1,J, when book leverage, /, is constant. This makes the equity of
rms with higher book-to-market value riskier.
The second effect comes from the interaction of nancial leverage and investment. We have
seen in Section IV that nancial leverage increases the effective degree of investment irreversibility
faced by the owners of the rm, and in Section V.A that irreversibility causes a growth premium,
rather than a value premium. Therefore, the effect of leverage on business risk, which is captured
by the second factor in equation (20), counteracts the Modigliani and Miller effect.
The net effect of leverage on expected stock returns depends on the parameterization of the
model, which we focus on next.
31
VI. Calibration and Simulation
Some parameters of the model have direct counterparts in the data. Accordingly, the tax rate is
taken to be 35 percent from Taylor (2003). The risk-free rate is taken to be 2 percent, using a time
series average of Fama's monthly T-bill returns from the CRSP database from July 1963 to June
2008. The yearly value of o
S
is set to 0.38 in order to match the average monthly Sharpe ratio of
the excess market return, using the excess market return series from Kenneth French's webpage,
again from July 1963 to June 2008. Because of the interaction between the resale price of capital
and book leverage, it is enough to preset only one of these parameters. Since there is no consensus
regarding the exact value of the resale price of capital, whereas we actually observe the book values
from Compustat, I preset book leverage to 0.5 and estimate the implied resale value of capital.
The remaining parameters for which we do not have direct observations are estimated via
maximum likelihood, using the long-run stationary distribution of the book-to-market values from
Compustat. It is possible to calibrate all the parameters using a collection of numbers from other
papers such as Zhang (2005), Cooper (2006), or Gomes and Schmid (2010), or to estimate the pa-
rameters that t the distribution of returns, as in Carlson, Fisher, and Giammarino (2004). Instead,
I make use of the distribution of book-to-market values, because explaining the cross-section of re-
turns consists of two important steps: getting the relationship between returns and book-to-market
values right and getting the distribution of book-to-market values right. If the model fails any of
these steps, it cannot produce the correct distribution of returns. Even worse, a model can claim to
explain the cross-section of returns correctly although it fails in both steps. Therefore, starting the
analysis with the distribution of book-to-market values provides a consistency check.
32
The appendix shows the derivation of the closed-form solution for the long-run stationary dis-
tribution of book-to-market values implied by the model. For estimation purposes, I make the
counterfactual assumption that book-to-market values are serially and cross-sectionally indepen-
dent and identically distributed, because the complex nature of the full information maximum
likelihood function would require resorting to simulated maximum likelihood, which would be
computationally expensive. Hayashi (2000, p. 465) shows that the resulting quasi-maximum like-
lihood estimator is indeed consistent. Therefore, it is a safe approach, given the high number of
rm-year observations in Compustat.
14
The resulting estimation values are presented in Table I, while Figure 4 gives the relationship
between conditional expected stock returns and the book-to-market values implied by the calibra-
tion. Indeed, we see that the Modigliani and Miller channel of leverage dominates the investment
channel, because the equity returns are increasing in book-to-market values.
Using this calibration, we can also immediately decompose the contribution of leverage to stock
returns through investment and the Modigliani-Miller channels. Figure 5 shows that introducing
debt has hardly any effect on business risk; hence the Modigliani-Miller channel easily dominates
the investment channel, as conrmed by Figure 4.
Using the parameter values in Table I, I simulate the model to obtain the statistics for various
book-to-market portfolios as in Fama and French (1992). Table II presents the simulation results
and the statistics obtained from the Compustat and CRSP data for the July 1963–June 2008 period.
This table shows that the simulated returns, book-to-market ratios, and market leverage are very
14
See Section 7.2 in Hayashi (2000), in particular the second remark on p.465. As in Fama and French (1992, Table
III), the smallest and largest 0.5% of the book-to-market ratios are winsorized.
33
3 2 1 1 2
logBM
10
20
30
40
Equity Returns in
Figure 4: Expected returns vs. book-to-market ratio using the estimated parameters.
34
Table I: Model parameters. The numbers in square brackets give 95 % condence intervals
calculated using 1000 bootstrap samples. Due to the large number of observations for book-to-
market value, the condence intervals are very tight. The productivity multiplier, /, affects the
levels of equity and debt but not the quantities we are interested in because these quantities depend
on the ratios, rather than levels. Hence, h is a free parameter and normalized to 1.
Model Parameters
Parameter Name Calibrated value Condence Interval
Interest Rate : 0.02
Price of Risk o
S
0.38
Corporate Tax t 0.35
Book Leverage / 0.50
Aggregate Volatility o
¹
0.14 [0.13,0.15]
Idiosyncratic Volatility o
i
0.51 [0.49,0.54]
Technology Drift j
A
-0.07 [-0.08,-0.06]
Maintenance Cost o 0.02 [0.0209,0.0249]
Capital Share in Production 1 ¸ 0.10 [0.1008,0.1035]
Resale/Purchase Value of Capital j 0.10 [0.1011,0.1022]
3 2 1 1 2
logBM
4
6
8
10
Unlevered Returns in
3.0 2.5 2.0 1.5 1.0 0.5 0.5
logBM
4
6
8
10
All Equity Returns in
Figure 5: Total yearly expected rm returns, debt and equity combined, for the levered rms
and all-equity nanced rms using estimated parameters.
35
Table II: Data versus simulation results with estimated parameters. Returns, BE/ME, ML, and
BL are time series averages of equally weighted cross-sectional means of equity returns, book-to-
market ratios, market leverage, and book leverage, respectively. These variables are dened in the
caption of Figure 1. The simulation results are the average of 200 simulations with 3000 rms and
1500 periods each. The rst 500 periods have been discarded to allow the system to converge to
its steady state. Using simulation results, the stocks are sorted according to their book-to-market
values in order to form 10 portfolios every month, as in Cooper (2006). Yearly sorting does not
change the results in a signicant way. The results are averages across simulations. Simulated
returns are adjusted upwards for ination. Columns 6 to 9 give benchmark model results; columns
10 and 11 give results when the maintenance cost, o, is changed from its benchmark value. Data
Source: CRSP and Compustat merged database, June 1963 to July 2008.
δ=0.01 δ=0.00
Portfolio Return BE/ME ML BL Return BE/ME ML BL Return Return
1 8.48 0.16 0.17 0.48 10.56 0.12 0.10 0.50 10.51 10.47
2 10.62 0.32 0.24 0.45 11.17 0.25 0.20 0.50 11.15 11.13
3 12.13 0.45 0.31 0.47 11.73 0.37 0.27 0.50 11.71 11.70
4 12.50 0.57 0.38 0.49 12.28 0.49 0.33 0.50 12.27 12.25
5 14.14 0.70 0.43 0.51 12.86 0.62 0.38 0.50 12.84 12.83
6 15.50 0.83 0.48 0.52 13.46 0.76 0.43 0.50 13.45 13.44
7 17.14 0.99 0.51 0.52 14.19 0.92 0.48 0.50 14.17 14.17
8 17.73 1.19 0.55 0.52 15.25 1.16 0.54 0.50 15.23 15.23
9 19.92 1.51 0.58 0.51 17.13 1.59 0.61 0.50 17.12 17.11
10 24.29 2.85 0.66 0.51 23.19 2.98 0.74 0.50 23.17 23.16
Data Model
36
close to the data, in accordance with the intuition presented in the paper: when book leverage is
relatively constant across different portfolios, value rms have higher leverage than growth rms;
hence, investing in the equity of value rms is riskier than investing in the equity of growth rms.
Moreover, the last two columns of Table II provide a sensitivity analysis for the maintenance
cost parameter, o, to support the claim that market leverage, and not operating leverage, is the main
driving force in the model because maintenance cost may cause an operating leverage channel as
in Carlson, Fisher, and Giammarino (2004). One might expect that stock returns will decrease as
the maintenance cost gets smaller because operating leverage increases the business risk. For ex-
ample, this is what we do observe in Zhang (2005). However, this intuition holds only if a change
in operating leverage does not affect the investment or nancing behavior of the rm. Indeed,
Zhang (2005) does not have any nancial leverage and investment decisions are independent of
operating leverage. In my model, given book leverage, a reduction in operating leverage decreases
the effective degree of investment irreversibility, the proof of which mirrors proposition 2, because
operating leverage increases the effective purchase and resale value of capital by the same amount.
Moreover, given investment behavior, a reduction in operating leverage also affects nancing be-
havior by increasing the debt capacity of the rms which in turn increases the equity risk. Table
II tells us that these effects counteract each other resulting in a small change in returns. Moreover,
the operating leverage parameter, which corresponds to my maintenance cost parameter, is equal
to 1.5 in Carlson, Fisher, and Giammarino (2004), whereas in my paper o = 0.02. which implies
that any effect of operating leverage should be much smaller in my paper, consistent with the result
in Table II.
37
Finally and not surprisingly, because this is a single factor model, the Capital Asset Pricing
Model (CAPM) ,s explain a signicant part of the variation in stock returns (not reported here).
This issue is addressed in the next section.
VII. Discussion
A. Failure of CAPM
One property of the model is that there is only a single systematic shock, and hence the conditional
CAPM holds. Although the unconditional version of the CAPM cannot explain perfectly the dif-
ferences in stock returns, it still explains a signicant fraction—more than what is predicted by the
data. This is a common property of the production-based models that try to explain cross-sectional
variation in stock returns with only one shock.
15
However, one reason the value premium is a puz-
zle is that it cannot be explained by the CAPM. Many investor-based models like the intertemporal
capital asset pricing model (ICAPM), as studied by Merton (1973), Campbell and Vuolteenaho
(2004), and Lettau and Wachter (2007), suggest that the CAPM fails because it does not price the
risk factors correctly when there are multiple aggregate shocks. The appendix shows that the ex-
tension of the model with shocks to the price of capital goods can easily generate a similar result
without changing any quantitative implications. The bottom line is that the leverage-based model
presented here does a very good job of capturing the distribution of returns and market leverage
across different book-to-market portfolios, whereas we can rely on additional macroeconomic fac-
15
Examples are Carlson, Fisher, and Giammarino (2004), Zhang (2005), and Berk, Green, and Naik (1999), among
others.
38
tors in order to generate the failure of the CAPM as in the studies above.
B. Fama-MacBeth Regressions
This section provides further empirical results in order to compare the implications of the model
with other previous studies that focus on the relationship between leverage and returns.
16
For
example, Johnson (2004) and Gomes and Schmid (2010) document a weak relationship between
market leverage and stock returns after accounting for book-to-market values.
17
Moreover, it is also
interesting to look at the relationship between stock returns and book leverage, although my model
is silent about this relationship since book leverage is constant in the model. In particular, Fama and
French (1992) document that book leverage is negatively related to stock returns after accounting
for market leverage, and Gomes and Schmid (2010) nd that the relationship between returns and
book leverage is weak before and after controlling for book-to-market values. In contrast to Gomes
and Schmid (2010), the results in George and Hwang (2010) seem to imply that book leverage
is negatively related to returns even after controlling for book-to-market values. This evidence
may seem contrary to the claim of the paper that market leverage is the main source of the value
premimum, at rst glance. In the following, I argue that this is not the case and, at the end, provide
an explanation that reconciles the claim with the evidence.
To study the relationship of returns with book-to-market values, market leverage, and book
16
I thank the anonymous referee for his comments that led to the development of this section.
17
A recent empirical study by Penman, Richardson, and Tuna (2006) nds that market leverage is negatively related
with equity returns after controlling for what they call the enterprise book-to-price ratio. However, their denition of
market leverage signicantly differs from Johnson (2004) and Gomes and Schmid (2010) because cash and short-term
investments are subtracted from nancial liabilities before calculating market leverage.
39
leverage in a unied framework I decompose book-to-market value into a market leverage and
a book leverage component. I stick to the denition of market and book leverage in Section I.,
which is similar to the denition in Gomes and Schmid (2010), to be consistent with the rest of the
analysis. Let ¹ be the book value of total assets, `1 the market value of equity, and 11 be the
book value of equity. Then, I can decompose the book-to-market ratio, 11,`1, as
11
`1
=
(¹ 11) , (`1 + ¹ 11)
`1, (`1 + ¹ 11)
11,¹
(¹ 11) ,¹
=
`1
1 `1
1 11
11
. (21)
where `1 and 11 are market and book leverage, respectively. Therefore, the value premium
should stem from either market leverage or book leverage. Note that the rst term on the right is
an increasing function of market leverage and the second term is a decreasing function of book
leverage.
This decomposition is somewhat different from Fama and French's (1992) decomposition,
11,`1 = (11,¹) (¹,`1), where ¹ is the book value of total assets. There is good rea-
son for this difference. Fama and French use ¹,11 as a measure of book leverage and ¹,`1 as
a measure of market leverage. While the ratio ¹,11 is clearly positively related with book lever-
age, the relationship between ¹,`1 and market leverage is not straightforward. In particular, we
can write ¹,`1 = (¹11),`1 +11,`1. The rst term, the ratio of debt to market value
of equity, is related to market leverage, as used in Gomes and Schmid (2010), for example. The
second term is the original book-to-market ratio. So, although ¹,`1 may be a good measure
40
Table III: Fama-MacBeth Regressions of stock returns on various variables. Book-to-market
value (BE/ME), market leverage (ML), and book leverage (BL) are dened in the caption of Figure
1. The coefcients are the time series average of regression coefcients for July 1963 to June 2008,
and the t-statistics (in parantheses) are the average regression coefcient divided by its time series
standard error. Note that some of the coefcients and t-statistics are similar to each other simply
because one variable is the sum of the other two. Each horizontal line is a separate regression.
log BE/ME log ML/(1-ML) log (1-BL)/BL
0.42 (6.38)
0.11 (2.61)
0.03 (0.69)
0.45 (8.12) -0.04 (1.14)
0.41 (6.39) 0.04 (1.14)
0.41 (6.39) 0.45 (8.13)
of market leverage, it includes more information than market leverage only. Instead, the terms
in my decomposition dissect the book-to-market ratio into its book leverage and market leverage
components completely.
Using this decomposition, I run all possible univariate and bivariate Fama-MacBeth regressions
of stock returns on the logarithms of 11,`1, `1, (1 `1) and (1 11) ,11. Table III
gives the results of these six regressions. The regression results are qualitatively the same when I
use Fama and French decomposition instead or when I run these regressions using log 11,`1,
log `1, and log 11, ignoring any decomposition. Adding rm size or the total returns over the
previous year do not change the regression results qualitatively, either.
Table III shows that while the sign of market leverage is positive and signicant in the univariate
41
regression, it becomes insignicant once we control for book-to-market value, as in Johnson (2004)
and Gomes and Schmid (2010). However, the lack of a statistically signicant coefcient on market
leverage after controlling for book-to-market value is not against the claim that market leverage is
the main source of the value premium. Suppose that the coefcient on market leverage remained
positive and statistically signicant after accounting for book-to-market value. This may only
imply that market leverage carries information about returns beyond what is captured in book-
to-market value. Moreover, the decomposition of book-to-market value into its market leverage
and book-leverage components suggests that the relationship between book-to-market value and
returns should stem either from market leverage or from book leverage. So, the question is which
of them drives the variation in stock returns (the value premium) and book-to-market values.
The unlevering exercise in Section I. already provides evidence for the importance of market
leverage in explaining the variation in stock returns across different book-to-market portfolios. For
the variation in book-to-market values, I look at the correlation of book-to-market value and its
components in the data. The time series average of the cross-sectional Spearman rank correlation
between book-to-market value and its market leverage component is 0.59 with standard deviation
0.07, whereas the rank correlation between book-to-market value and its book leverage component
is only 0.08 with standard deviation 0.09. Moreover, the time series average of the cross-sectional
Pearson correlation between the log of book-to-market value and its market leverage component
is 0.57 with standard deviation 0.06, whereas the Pearson correlation between the log of book-
to-market value and its book leverage component is only 0.02 with standard deviation 0.07. The
results are qualitatively the same when I use Fama and French's decomposition or log 11,`1,
42
log `1, and log 11, ignoring any decomposition. This evidence suggests that the market leverage
component is much more important than the book leverage component in explaining the variations
in book-to-market ratios. This analysis supports the idea that market leverage is the main driving
force behind the value premium.
The results for book leverage in Table III are in line with Gomes and Schmid (2010) and Fama
and French (1992). The coefcient for the book leverage component is small and insignicant be-
fore and after controlling for book-to-market value, which is supported by the analysis of Gomes
and Schmid (2010). Moreover, book leverage has a negative and signicant coefcient after con-
trolling for market leverage, as in Fama and French (1992). A striking result in the last regression is
that the coefcients of the market leverage and book leverage components are very similar in mag-
nitude, which seems to imply that the book-to-market effect wraps up any effect of market leverage
and book leverage. This result is also available in Fama and French (1992) with their decompo-
sition, presented in their Table III. So, the question becomes again, which of these components
accounts for more of the variation in book-to-market values and stock returns. As explained in
Section I., there is hardly any value premium left once the returns are unlevered and, as discussed
above, most of the variation in book-to-market values comes from the market leverage component.
Moreover, as illustrated below, the regression results presented so far can be consistent with the
book leverage component having no explanatory power for stock returns.
To see this last point more clearly, suppose that the correct model explaining returns is given
by :
1
= c
0
+ c
1
, + c, where c
0
and c
1
are coefcients, , is the fundamental factor, and c is the
error term with 1(c) = 0. Also, suppose that both book-to-market value and market leverage are
43
Table IV: Expected Values of Regression Coefcients
log 11,`1 log `1, (1 `1) log (1 11) ,11
,
·ov())
·ov())+·ov(&
1
)
,
·ov())
·ov())+·ov(&
2
)
0
,
·ov(&
2
)
·ov(&
1
)+·ov(&
2
)+
var(u
1
)var(u
2
)
var(f)
,
·ov(&
1
)
·ov(&
1
)+·ov(&
2
)+
var(u
1
)var(u
2
)
var(f)
,
·ov(&
1
)+·ov(&
2
)
·ov(&
1
)+·ov(&
2
)+
var(u
1
)var(u
2
)
var(f)
,
·ov(&1)
·ov(&
1
)+·ov(&
2
)+
var(u
1
)var(u
2
)
var(f)
,
·ov(&
1
)+·ov(&
2
)
·ov(&
1
)+·ov(&
2
)+
var(u
1
)var(u
2
)
var(f)
,
·ov(&
2
)
·ov(&
1
)+·ov(&
2
)+
var(u
1
)var(u
2
)
var(f)
imperfect proxies for this fundamental, so that log 11,`1 = , + n
1
and log `1, (1 `1) =
, +n
2
, where n
1
and n
2
are the measurement errors with respect to the original proxy, with 1(n
1
)
and 1(n
2
) constant, and 1(n
1
c) = 1(n
2
c) = co·(n
1
. n
2
) = 0. Finally, suppose that ·c: (n
1
) is
small relative to ·c:(n
2
), i.e., market leverage is subject to greater measurement error, which is
likely because, following the papers cited above, I approximate the market value of debt with the
book value of debt for the calculation of market leverage. Note that the book leverage component
would then be log (1 11) ,11 = log 11,`1 log `1, (1 `1) = n
1
n
2
and it would
be a random number completely unrelated to fundamentals. Table IV summarizes the expected
values of regression coefcients under this scenario. This scenario suggests that we should expect
the following results:
1. The book-to-market value and the market leverage component are highly correlated, whereas
the book-to-market value and the book leverage component have low correlation. Note that
the expected values of Pearson correlation coefcients are ·c: (,) ,
_
·c: (, + n
1
) ·c: (, + n
2
)
and ·c: (n
1
) ,
_
·c: (, + n
1
) ·c: (n
1
n
2
). respectively.
44
2. When we run two univariate regressions of returns on the book-to-market value and the mar-
ket leverage component separately, both regression coefcients should be expected to have
the same sign, but the coefcient of market leverage should be lower. (Regressions 1 and 2.)
We should also expect a small and insignicant coefcient when we run the regression of
returns on book leverage. (Regression 3.)
3. When we run the regression of returns on the book-to-market value and the market leverage
component, the coefcient on the book-to-market term should be large, whereas the coef-
cient on market leverage should be small. (Regression 4.) We should expect a similar pattern
when we replace market leverage with book leverage. (Regression 5.) The magnitude of the
coefcients on book leverage and market leverage should be smaller the smaller ·c: (n
1
) is
relative to ·c: (n
2
).
4. When we regress returns on the market leverage and book leverage components, both of
them should be large and have similar magnitude. (Regression 6.) The magnitude of the two
coefcients become closer the lower ·c:(n
1
) is relative to ·c: (n
2
).
All these items are in line with the regression results in Table III and the correlations discussed
before. Finally, this argument also sheds light on why we see constant average book leverage across
different book-to-market portfolios, as seen in Figure 1: As the sample size increases, the average
value of the book leverage component in each portfolio approaches the constant 1(n
1
n
2
) by the
law of large numbers.
So, the results presented in this section are compatible with an explanation where book-to-
45
market value and market leverage proxy for the same fundamental factor, despite that the book-
to-market proxy is subject to less error. Therefore, the regression results above do not refute the
hypothesis that market leverage is the main source of the value premium but actually are consistent
with the idea that market leverage and book-to-market value proxy for the same source of risk.
These results also suggest that book leverage is not necessarily related to book-to-market value
and stock returns systematically. Although book leverage is constant in the model, it is possible to
introduce cross-sectional variation in book leverage to the model, for example, by assuming that
some of the model parameters are rm or industry specic or by introducing an idiosyncratic shock
to a model parameter, such as the resale value of capital, j. This would not change the main results
of the paper, although it would make it harder to derive closed-form solutions and the resulting
clear intuition provided in this paper.
VIII. Conclusion
This paper presents a dynamic model of the rm with limited capital irreversibility and risk-free
debt contracts in order to analyze the effects of nancial leverage on investment and explain the
positive relationship between book-to-market values and stock returns. This model can capture sev-
eral regularities in the corporate nance and asset pricing literature in a parsimonious and tractable
way.
The model also suggests testable implications: The effect of nancial leverage on returns
through the investment channel implies that rms with a greater interest tax-shield invest as if
46
they faced greater investment irreversibility. Moreover, the model also suggests that industries
with a higher degree of investment irreversibility, corrected for the interest tax shield and leverage,
should have a weaker relationship between book-to-market values and stock returns. One can po-
tentially test these implications of the model by comparing the investment behavior and the value
premium in industries with different corrected investment irreversibility. Cohen and Polk (1998)
show that the value premium is a within-industry, rather than an inter-industry phenomenon, so
this would be a valid test.
Introducing debt into production-based asset pricing models opens several possibilities. For
example, the model presented here could be extended with time-varying interest rates in a frame-
work similar to Merton's (1973) intertemporal capital asset pricing model (ICAPM). This would
serve two purposes. First, it would decrease the explanatory power of the conditional market beta
for stock returns and get us one step closer to solving the value premium puzzle. Second, because
rms with a high book-to-market ratio also have higher leverage, they will have greater exposure
to interest rate shocks, further reinforcing the value premium. This is left for future research.
47
Appendix
A. Boundary Conditions
This section presents the boundary conditions for the HJB equation 9.
18
The boundary conditions
at the investment boundary, A
l
(1. /), are
J
1
(1. A
l
(1. /) . /) = 1 / (22)
J
11
(1. A
l
(1. /) . /) = 0 (23)
J
1A
(1. A
l
(1. /) . /) = 0 (24)
J
1b
(1. A
l
(1. /) . /) = 1. (25)
while the boundary conditions at the disinvestment boundary, A
1
(1. /), are given by
J
1
(1. A
1
(1. /) . /) = j / (26)
J
11
(1. A
1
(1. /) . /) = 0 (27)
J
1A
(1. A
1
(1. /) . /) = 0 (28)
J
1b
(1. A
1
(1. /) . /) = 1. (29)
18
Dixit (1993) is a good introduction to the derivation of these conditions.
48
Finally, if we denote the book leverage after adjustment as /
0
, the boundary conditions at the debt
adjustment boundaries
19
are given by
J (1. A
1
(1. /) . /) = J (1. A
1
(1. /) . /
0
) + (/
0
/) 1 c/
0
1 (30)
J
1
(1. A
1
(1. /) . /) c/ = J
1
(1. A
1
(1. /) . /
0
) + (/
0
/) c/
0
(31)
J
A
(1. A
1
(1. /) . /) = J
A
(1. A
1
(1. /) . /
0
) (32)
J
b
(1. A
1
(1. /) . /) = 1 (33)
(1 c)1 J
b
(1. A
1
(1. /) . /
0
) . (34)
The last of these conditions is the rst-order condition with respect to after-adjustment leverage, /
0
,
under the debt constraint. Hence, it holds as an equality if the new book leverage satises /
0
< j or
J (1. A. /
0
) 0, depending on the constraint imposed by the lender. The market value of equity,
J (A. 1. /), should be homogeneous of degree one in 1 and A, because the cash ows and the
adjustment costs on debt and investment are homogeneous in 1 and A.
20
Equation (30) says that, at the time of debt adjustment, the net shareholder value before and
after adjustment should be the same. If the shareholder value before adjustment were greater,
adjusting debt would harm the shareholders. If the shareholder value before adjustment were
lower, continuity of equity value in its arguments would imply that shareholders could gain from
19
There are potentially two adjustment boundaries for upward or downward adjustment. However, since the cost of
adjustment is the same in both cases, the boundary conditions are the same.
20
This argument is similar to the one in Abel and Eberly (1996) and Cooper (2006). To justify this homogeneity
property and hence that X=K is a sufcient statistic to describe the solution of the model, the reader can directly sub-
stitute in V (y; b) V (X=K; b) = J (X; K; b) =K and see that both the HJB equation and the boundary conditions
can be expressed in terms of V (y; b) and its derivatives.
49
adjustment prior to the specied adjustment time and hence the suggested adjustment time cannot
be optimal.
Equations (31) to (33) state that, at the time of adjustment, the shareholder value should be
differentiable with respect to the state variables. Intuitively, we can think of these smooth pasting
conditions as envelope conditions. To see this more clearly, note that /
0
is chosen such that the
right side of equation (30), i.e., the net shareholder value after adjustment, is maximized, and this
maximized shareholder value should be equal to before-adjustment shareholder value, that is,
J(1. A. /) = max
b
0
J(1. A. /
0
) + (/
0
/)1 c/
0
1 (35)
subject to /
0
j.
Using the envelope theorem, we can take the derivatives of both sides with respect to 1, A, and
/ and replace A = A
1
(1. /) to get equations (31) to (33). Therefore, according to the envelope
theorem, equations (31) to (33) yield the amount the shareholders should be compensated ex-ante
if capital K, productivity X, and before-adjustment book leverage, respectively, increase by a small
amount. Equation (34) is simply the rst-order condition of this last maximization problem and
states that the marginal gain from adjustment should be equal to or greater than the marginal loss
fromadjustment, because of the constraint /
0
j. Debt adjustment occurs if/when these optimality
conditions are satised.
50
B. Market Value and Stock Returns with Irreversibility and Financial Lever-
age
We can simplify our problem by dening ~ ¡ (¸. /) ¡ (¸. /) + :,:. Therefore, equations (11),
(13), and (12) can be rewritten as
:~ ¡ (¸. /) =


¸
+ j¸~ ¡
j
(¸. /) +
1
2
o
2
¸
2
~ ¡
jj
(¸. /) (36)
~ ¡ (¸
1
(/) . /) = j / + :(/) ,: | (/) and ~ ¡
j

1
(/) . /) = 0 (37)
~ ¡ (¸
l
(/) . /) = 1 / + :(/) ,: n(/) and ~ ¡
j

l
(/) . /) = 0. (38)
This makes the solution of the differential equation similar to the one in Abel and Eberly
(1996), which is a special case of my model that excludes leverage and the risk preferences of
investors. Following their analysis, I dene the following functions
j (r) =
1
2
o
2
r
2

_
j
1
2
o
2
_
r + : (39)
o (r) =
r
c
P
r
¸
r
c
P
r
c
N
(40)
c(r) =
1
j (¸)
_
1
¸
c
.
o (r)
¸
c
1
[1 o (r)]
_
. (41)
where c
1
and c
.
are the roots of the quadratic equation j (r) = 0 and satisfy c
1
1 ¸ 0
c
.
, given that we require : j for the convergence of solution to the rm's problem. Then, the
51
solution of this differential equation should be of the form
~ ¡ (¸. /) = H¸
¸
+ C
1
(/) ¸
c
P
+ C
.
(/) ¸
c
N
. (42)
The reason is that / appears only in the boundary conditions for ~ ¡ but not in the differential equa-
tion.
Dene H (¸)

/,j (¸) and G(/) ¸
l
(/) ,¸
1
(/). Then, the solution of the differential
equation for ~ ¡ (¸. /) is given by
21
~ ¡ (¸. /) = H (¸) ¸
1
(/)
¸

__
¸
¸
1
(/)
_
¸

¸
c
1
[1 o (G(/))]
_
¸
¸
1
(/)
_
c
P

¸
c
.
o (G(/))
_
¸
¸
1
(/)
_
c
N
_
. (43)
where G(/) is implicitly dened by
n(/)
| (/)
c(G(/)) G(/)
¸
c(1,G(/)) = 0 (44)
and the values of boundaries are given by


l
(/)
¸
=
n(/)
c(1,G(/))
and


1
(/)
¸
=
| (/)
c(G(/))
. (45)
Using this solution, the value function can be found by simply integrating ¡ (A,1) over 1 to
21
The solution is identical to that of Abel and Eberly (1996) once the purchase and resale prices of capital in their
model are substituted with u(b) and l (b), respectively. Therefore, I omit the lengthy details of the calculus that leads
to the solution, but they are available upon request.
52
get
22
J (1. A. /)
= H (¸) ¸
¸
1
_
1
1 ¸
A
¸
1

¸
1
(/)
¸

¸
c
1
1 o (G)
1 c
1
A
c
P
1
1c
P
¸
1
(/)
c
P

¸
c
.
o (G)
1 c
.
A
c
N
1
1c
N
¸
1
(/)
c
N
_

:(/)
:
1
=

HA
¸
1

+

1
1
(/) A
c
P
1
1c
P
+

1
.
(/) A
c
N
1
1c
N

:(/)
:
1
= J
¹1
+ J
G
+ J
1

:(/)
:
1. (46)
where J
¹1
is the value of assets in place, before costs, J
G
is the value of growth options, J
1
is the value of the disinvestment option, and :(/) ,:1 is the present value of maintenance and
nancing costs. Note that the derivation of the market value of equity does not make use of the
boundary conditions for debt restructuring. The book leverage, /, is constant in the inaction region
due to debt adjustment costs. Therefore, any effect of debt adjustment on market value of equity
should be captured by constants of integration

1
1
(/) and

1
.
(/) and hence we do not have a
separate term for debt adjustment option. The analysis here shows that the boundary conditions
for investment are enough to pin down these constants of integration.
Now we focus on stock returns. Apply Ito's Lemma to equity value in the inaction region,
J (1. A)
dJ =
_
j
A
AJ
A
+
1
2
o
2
A
2
J
AA
_
dt + oAJ
A
dn. (47)
22
Direct integration of q (y) would yield a constant of integration that should have the form D
X
(b) X, due to the
homogeneity property of the value function. However, direct substitution of J (X; K) into the HJB equation shows
immediately that D
X
(b) should be zero.
53
Use the relationship j = j
A
o
S
o
¹
and the HJB equation :J = : +jAJ
A
+
1
2
o
2
A
2
J
AA
to get
1
dt
1
_
dJ
J
_
=
:
J
+ : + o
S
J
A
A
J
. (48)
Plugging this expression into the return formula gives the excess returns
1
dt
1 (d1) : =
1
dt
1
_
:dt + dJ
J
_
: = o
S
o
¹
J
A
A
J
. (49)
To show that the returns are increasing in ¸, rst note that we can write the excess returns as
1
dt
1 (d1) : =
_
¸
J
¹1
J + :,:1
+ c
1
J
G
(1. A)
J + :,:1
+ c
.
J
1
(1. A)
J + :,:1
__
1 +
:,:1
J
_
o
S
o
¹
= (¸:
¹1
+ c
1
:
G
+ c
.
:
1
)
_
1 +
:,:1
J
_
o
S
o
¹
. (50)
where the rst term decomposes the total business risk into assets-in-place, the growth option and
the disinvestment option, while the second term captures the Modigliani-Miller effect.
C. Proof of Proposition 2
We can rewrite equation (44) as
n(/)
| (/)
=
G(/)
¸
c(1,G(/))
c(G(/))
. (51)
54
Here, n(/) ,| (/) captures the degree of irreversibility perceived by the shareholders and G(/)
¸
l
(/) ,¸
1
(/) captures the gap between the investment and disinvestment boundaries. It is easy to
show that the left side of this equation is increasing in /, and hence the right side of the equation
should be increasing in /. Abel and Eberly (1995) show in a lengthy and tedious proof that
G

ç(1¸G)
ç(G)
is increasing in G. Suppose G
0
(/) 0. Then, the left side would be weakly decreasing in /,
causing a contradiction.
D. Proof of Proposition 3
Debt is strictly preferable to equity nancing if the marginal value of debt net of the cost of nanc-
ing is positive, that is if J (1. A. /) +/1c/1 is increasing in / or J
b
(1. A. /) +(1 c) 1 0.
We should rst nd J
b
(1. A. /). Remember that the market value of equity has the form
J (1. A. /) =

HA
¸
1

+

1
1
(/) A
c
P
1
1c
P
+

1
.
(/) A
c
N
1
1c
N

:(/)
:
1. (52)
Because

1
1
(/) and

1
.
(/) are highly non-linear functions of / and because G(/) is an implicitly
dened function, a brute force approach would be too tedious. Instead, we focus on the value-
matching and smooth pasting conditions. Using the functional formand homogeneity of the market
value of equity in A and 1, the value-matching conditions for the marginal value of capital can be
55
expressed as


l
(/)
¸
+ (1 c
1
)

1
1
(/) ¸
l
(/)
c
P
+ (1 c
.
)

1
.
(/) ¸
l
(/)
c
N
= 1 t/ + (1 t)
o
:
(53)


1
(/)
¸
+ (1 c
1
)

1
1
(/) ¸
1
(/)
c
P
+ (1 c
.
)

1
.
(/) ¸
1
(/)
c
N
= j t/ + (1 t)
o
:
(54)
and the smooth-pasting conditions can be expressed as

H¸¸
l
(/)
¸
+ (1 c
1
) c
1

1
1
(/) ¸
l
(/)
c
P
+ (1 c
.
) c
.

1
.
(/) ¸
l
(/)
c
N
= 0 (55)

H¸¸
1
(/)
¸
+ (1 c
1
) c
1

1
1
(/) ¸
1
(/)
c
P
+ (1 c
1
) c
.

1
.
(/) ¸
1
(/)
c
N
= 0. (56)
If we take the derivatives of equations (53) and (54) with respect to / and plug equations (55)
and (56) in the resulting expressions we get
(1 c
1
)

1
0
1
(/) ¸
l
(/)
c
P
+ (1 c
.
)

1
0
.
(/) ¸
l
(/)
c
N
= t (57)
(1 c
1
)

1
0
1
(/) ¸
1
(/)
c
P
+ (1 c
.
)

1
0
.
(/) ¸
1
(/)
c
N
= t. (58)
56
which gives

1
0
1
(/) =
t
1 c
1
1 G(/)
c
N
G(/)
c
P
G(/)
c
N
1
¸
1
(/)
c
P
(59)

1
0
.
(/) =
t
1 c
.
G(/)
c
P
1
G(/)
c
P
G(/)
c
N
1
¸
1
(/)
c
N
. (60)
As a result, the derivative of market value of equity with respect to leverage is given by
J
b
(1. A. /) = t
_
1 G
c
N
G
c
P
G
c
N
1
1 c
1
_
¸
¸
1
_
c
P
+
G
c
P
1
G
c
P
G
c
N
1
1 c
.
_
¸
¸
1
_
c
N
+
1 t
t
_
1.
(61)
where / has been dropped in G(/) for the sake of brevity. Therefore the marginal value of debt is
given by
J
b
+ (1 c) 1 = t
_
1
1 G
c
N
G
c
P
G
c
N
1
1 c
1
_
¸
¸
1
_
c
P

G
c
P
1
G
c
P
G
c
N
1
1 c
.
_
¸
¸
1
_
c
N
_
1
c1. (62)
Since G 1 and c
1
1 0 c
.
, the term in square brackets is increasing in ¸. Therefore,
debt is the preferred form of nancing at every state if J
b
+ (1 c) 1 0 at ¸ = ¸
1
. Substituting
¸ = ¸
1
above reduces our condition to
c < c

(/) = t
_
1
1 G
c
N
G
c
P
G
c
N
1
1 c
1

G
c
P
1
G
c
P
G
c
N
1
1 c
.
_
. (63)
Since G 1 and c
1
1 0 c
.
, the second term is on the right side is negative and the third
57
term is less than 1. Moreover, the right side of this equation is decreasing in G, and we already
knowfromthe proof of the previous proposition that G
0
(/) 0. Therefore, c

(/) 0 and c
0
(/) <
0. The minimum for c

(/) is then attained when G !1, that is, c

= 1 1, (1 c
.
). So, if c <
1 1, (1 c
.
), then debt is always preferable regardless of state and degree of irreversibility.
E. Proof of Proposition 4
From equation (62) we have
J
b
+ 1 = t
_
1
1 G
c
N
G
c
P
G
c
N
1
1 c
1
_
¸
¸
1
_
c
P

G
c
P
1
G
c
P
G
c
N
1
1 c
.
_
¸
¸
1
_
c
N
_
1. (64)
which attains its minimum value for ¸ = ¸
1
. So, it is enough to show that this value is positive
once we substitute ¸ = ¸
1
, i.e. that
t
_
1
1 G
c
N
G
c
P
G
c
N
1
1 c
1

G
c
P
1
G
c
P
G
c
N
1
1 c
.
_
0. (65)
Again, since G 1 and c
1
1 0 c
.
, the second term on the right side is negative and
the third term is less than 1. Therefore, the term in square brackets should be positive. Since
J
b
+ 1 0, one of the smooth pasting conditions at debt adjustment is not satised, and hence it
is not optimal to readjust debt.
58
F. Market Value and Stock Returns with Investment Irreversibility Only
The market value of equity under this setting is the same, except that we should set / = o = 0. Let
H (¸) /,j (¸) and G ¸
l

1
. Then, the solution of the differential equation for ¡ (¸) where
¸ A,1 is given by
¡ (¸) = H (¸) ¸
¸
1
__
¸
¸
1
_
¸

¸
c
1
[1 o (G)]
_
¸
¸
1
_
c
P

¸
c
.
o (G)
_
¸
¸
1
_
c
N
_
. (66)
where G is the solution of
1
j
c(G) G
¸
c
_
G
1
_
= 0. (67)
Using this solution, the value function can be found by simply integrating ¡ (A,1) over 1 to get
23
J (1. A)
= H (¸) ¸
¸
1
_
1
1 ¸
A
¸
1

¸
¸
1

¸
c
1
1 o (G)
1 c
1
A
c
P
1
1c
P
¸
c
P
1

¸
c
.
o (G)
1 c
.
A
c
N
1
1c
N
¸
c
N
1
_
= J
¹1
+ J
G
+ J
1
. (68)
23
Direct integration of q (y) would yield a constant of integration that should depend linearly on X due to the
homogeneity property of the value function. However, direct substitution of J (X; K) into the HJB equation shows
immediately that this term should be zero.
59
To show that the returns are increasing in ¸, rst note that we can write the excess returns as
1
dt
1 (d1) : =
_
¸
J
¹1
(1. A)
J (1. A)
+ c
1
J
G
(1. A)
J (1. A)
+ c
.
J
1
(1. A)
J (1. A)
_
o
S
o
¹
=
_
¸
J
¹1
(1. ¸)
J (1. ¸)
+ c
1
J
G
(1. ¸)
J (1. ¸)
+ c
.
J
1
(1. ¸)
J (1. ¸)
_
o
S
o
¹
=
_
¸ + (c
1
¸)
J
G
(1. ¸)
J (1. ¸)
+ (c
.
¸)
J
1
(1. ¸)
J (1. ¸)
_
o
S
o
¹
. (69)
Now, it is easy to show that J
G
(1. ¸) ,J (1. ¸) is increasing in ¸ and J
1
(1. ¸) ,J (1. ¸) is decreas-
ing in ¸ by taking the derivatives. Since c
1
¸ 0 and c
.
¸ < 0, this last expression should be
increasing in ¸. Book-to-market value is equal to J(1. A),1 = J(1. ¸) with ¸ = A,1. More-
over, J(1. A) is increasing in A since a rm with higher productivity should have higher market
value, given capital. Hence J
j
(1. ¸) = J
A
(1. A) 0 and book-to-market is decreasing in ¸. As
a result, returns should be decreasing in book-to-market.
G. Proof of Proposition 6
In case of perfect irreversibility the rm does not have a disinvestment option. Therefore, the
market value of equity consists of the value of growth options and assets-in-place only. If we let
:
¹1
be the return on assets in place and :
G
be the return on the growth option, we can write the
expected returns to equity as
:
1
=
J
¹1
(1. A)
J (1. A)
:
¹1
+
J
G
(1. A)
J (1. A)
:
G
. (70)
60
where
J (1. A) = J
¹1
(1. A) + J
G
(1. A) (71)
:
1
=
J
A
J
¸
¸
¸
¸
co·
_
dA.
do
o

¸
¸
¸
(72)
:
¹1
=
J
¹1
A
J
¹1
¸
¸
¸
¸
co·
_
dA.
do
o

¸
¸
¸
(73)
:
G
=
J
G
A
J
G
¸
¸
¸
¸
co·
_
dA.
do
o

¸
¸
¸
. (74)
Moreover, given capital, rms with higher productivity have lower book-to-market values, and
hence are growth rms, for which the growth options constitute a greater share of market value.
Therefore, we should have
JJ
G
(1. A) ,J (1. A)
JA
0. (75)
With a little algebra, we can show that
JJ
G
(1. A) ,J (1. A)
JA
=
J
G
(1. A) ,J (1. A)
¸
¸
co·
_
dA.
oS
S

¸
(:
G
:
1
) . (76)
which together with (70) and (75) implies that :
G
:
1
:
¹1
. Therefore, growth options are
riskier than assets-in-place.
H. The Long-Run Distribution of Book-to-Market Values
In this section we calculate the stationary long-run distribution of book-to-market values. I assume
that a rm leaves the sample if ¸ = ¸

¸
1
. This assumption serves two purposes: First, rms
61
with very low market values will leave the sample, in accordance with the Security Exchange
Commission rule that requires delisting of companies whose share price falls below a certain value
or for performance-related reasons. Second, because this assumption caps book-to-market values,
it will improve the t of the average book-to-market values and stock returns at the highest decile.
In order to have a stationary distribution, I also assume that each rm that leaves the stock mar-
ket is replaced by another rm that enters the market after paying a xed cost linearly proportional
to its capital. This later assumption guarantees that the entry point for all rms is the same and is
characterized by ¸ = ¸, as a result of the homogeneity of the maximization problem in A and 1.
24
Using the model parameters, we can calculate the cross-section of returns in the long run by
looking at the stationary distribution of ¸ between two barriers, ¸

and ¸
l
. The exit-entry mecha-
nism discussed above implies that the long-run cross-sectional distribution of ¸ will be the same as
the long-run distribution of a process with a resetting barrier at ¸

. where the target after resetting
is ¸. Note that the case without exit is a special case of this mechanism, where ¸ = ¸
l
and ¸

= ¸
1
.
and the no-entry-cost is a special case with ¸ = ¸

= ¸
1
.
The law of motion for ¸ between barriers is given by d¸,¸ = j
A
dt + odn. Dene . log ¸,
.

log ¸

, and .
l
log ¸
l
, and let q (.) be the long-run distribution of .. Bertola and Cabellero
(1990) show that q (.) is given by the solution of the Kolmogorov forward equation
q
00
(.) = 2
_
j
A

1
2
o
2
_
o
2
q
0
(.) (77)
24
The same entry point is a simplifying assumption. Different entry points would not affect the functional form for
the market value of equity, since debt capacity is independent of the state variables.
62
separately for the regions [.

. .) and ( .. .
l
] with the following boundary conditions
q
0
_
.

_
= q
0
_
.
+
_
+ q
0
_
.
+
_
(78)
q
0
(.
l
) = 2
_
j
A

1
2
o
2
_
o
2
q (.
l
) (79)
q (.

) = 0. (80)
where q (.
+
) is the right limit and q (.

) is the left limit of the distribution function. We also have
the integral condition
_
:
U
:

q (.) d. = 1. (81)
Once we nd q (.), we can nd the distribution of ¸ using the transformation ,(¸) = q (ln ¸) ,¸.
A little algebra shows that the long-run distribution of ¸ is given by
,(¸) =
_
¸
¸
¸
¸
¸
_
¸
¸
¸
¸
¸
_
_
¹
1
¸
(2j
X
o
2
)¸o
2
+ 1
1
_
,¸ if ¸

< ¸ < ¸
¹
2
¸
(2j
X
o
2
)¸o
2
1
if ¸ ¸ < ¸
l
0 otherwise.
(82)
63
where ¹
1
. ¹
2
and 1
1
satisfy
0 = (¸

)
(2j
X
o
2
)¸o
2
¹
1
+ 1
1
(83)
0 =
_
¸
(2j
X
o
2
)¸o
2

)
(2j
X
o
2
)¸o
2
_
¹
1
¸
(2j
X
o
2
)¸o
2
¹
2
(84)
1 =
¸
(2j
X
o
2
)¸o
2

)
(2j
X
o
2
)¸o
2
(2j
A
o
2
) ,o
2
¹
1
+
¸
(2j
X
o
2
)¸o
2
l
¸
(2j
X
o
2
)¸o
2
(2j
A
o
2
) ,o
2
¹
2
+ln
_
¸
¸

_
1
1
. (85)
Then, we can write market-to-book values as J, (1 /) 1 = \ (¸) , (1 /). Once we dene
the function . (¸) = \ (¸) , (1 /), the long-run distribution of market-to-book values, :/, is
given by
, (:/) = ,
_
.
1
(:/)
_
¸
¸
¸
¸
d.
1
(:/)
d (:/)
¸
¸
¸
¸
(86)
for \ (¸
l
) , (1 /) :/ \ (¸
1
) , (1 /), and zero otherwise.
Once we have the long-run distribution of market-to-book values, I use the long-run distribution
derived from the data in order to estimate the model parameters using maximum likelihood.
I. Time-Varying Price of Capital
I assume that the price of capital follows a geometric Brownian motion,
25
that is,
d1
t
1
t
= j
1
dt + o
1
dn
1
. (87)
25
For example, this process may come from a perfectly competitive industry that produces capital goods, with a
linear production function subject to technology shocks that follow a geometric Brownian motion.
64
For this process, j
1
< 0 implies that increasing the capacity becomes cheaper over time, creating
a vintage capital effect. We can write the problem of the rm as
\(1
t
.
^
A
t
. 1
t
) = max
fol
t+s
,o1
t+s
,ob
t+s
g
1
t
_
1
0
o
t+c
o
t

_
1
t+c
.
^
A
t+c
. 1
t+c
_
d:

_
1
0
o
t+c
o
t
[(1 /)1
t+c
dl
t+c
(j /) 1
t+c
d1
t+c
]
+
1

c2fc:ob
t+s
6=0g
o
t+c
o
t
[d/
t+c
c(/
t+c
+ d/
t+c
)] 1
t+c
1
t+c
(88)
subject to
d1
t
= dl
t
d1
t
(89)
do
t
o
t
= :dt o

dn
¹,t
+ o
S1
dn
1,t
(90)
d
^
A
t
^
A
t
= j
¹
dt + o
¹
dn
¹,t
+ o
i
dn
i,t
(91)
d1
t
1
t
= j
1
dt + o
1
dn
1,t
. (92)
where

_
1
t
.
^
A
t
. 1
t
_
(1 t)
_
/
1 ¸
^
A
t
1

t
o1
t
1
t
:/1
t
1
t
_
+ ^ j
1
/1
t
1
t
. (93)
The new term ^ j
1
/1
t
1
t
appears because of the changes in the amount of debt as a result of changes
in the price of capital, where ^ j
1
= j
1
+ o
S1
o
1
is the risk-adjusted drift of the price process.
Intuitively, operating capital acts as an asset that provides an instantaneous return of d1,1 as a
65
result of the debt agreement and price movements and ^ j
1
is the risk-adjusted value of this return.
I assume that the risk prices of
^
A and 1 have opposite signs, because good times are characterized
by higher productivity and lower input prices.
Note that this problem is linearly homogeneous in
^
A
t
and 1
t
, since both variables follow a
geometric Brownian motion and enter linearly into the problem. Therefore, I can divide everything
by 1
t
and dene A
¸
t

^
A
t
,1
t
, : (1. A) (1.
^
A. 1),1 and J (1. A) \(1.
^
A. 1),1. This
will give us the following Hamilton-Jacobi-Bellman equation in the inaction region
(: ^ j
1
) J = : (1. A) + jAJ
A
+
1
2
o
2
A
2
J
AA
. (94)
where
j =
1
¸
(^ j
¹
^ j
1
) +
1
2
_
1
¸
1
_
1
¸
_
o
2
1
+ o
2
¹
+ o
2
i
_
(95)
o
2
=
1
¸
2
_
o
2
1
+ o
2
¹
+ o
2
i
_
(96)
with ^ j
¹
= j
¹
o

o
¹
. This Bellman equation is very similar to the Bellman equation in the orig-
inal model, and the boundary conditions are identical. As a result of this close relationship with
the original model, all the analysis for the original model holds for this extended version. In partic-
ular, any investment and nancing policy, distribution of book-to-market values, and stock returns
under the original model can be replicated under the extended version. However, the behavior of
the conditional CAPM will change signicantly. We focus on this in the next sections.
66
I..1 Stock Returns
In this section I provide the expressions of individual stock returns, denoted by i, and of the value-
weighted market return in order to analyze the relationship of actual conditional expected returns
with those implied by the CAPM. It can be shown that the equity returns in this extended version
are given by
d1
i
=
_
: + o

o
¹
\
i
^
A
^
A
i
\
i
o
S1
o
1
\
i1
1
\
i
_
dt
+
\
i
^
A
^
A
i
\
i
(o
¹
dn
¹
+ o
i
dn
i
) +
\
i1
1
\
i
o
1
dn
1
. (97)
Due the homogeneity of \(1.
^
A. 1) in
^
A and 1, we have
W
iP
1
W
i
=
_
1
W
i
^
X
^
A
i
W
i
_
. Using this
and the denition of J, we can write the last equation as
d1
i
=
_
: + o

o
¹
J
iA
A
i
¸J
i
o
S1
o
1
_
1
J
iA
A
i
¸J
i
__
dt
+
J
iA
A
i
¸J
i
(o
¹
dn
¹
+ o
i
dn
i
) +
_
1
J
iA
A
i
¸J
i
_
o
1
dn
1
. (98)
Note that, similar to the original model, the effect of book-to-market value is captured by the
elasticity of the market value of equity with respect to productivity shocks J
iA
A
i
,J
i
.
Using individual stock returns, we can derive the market return as
d1
n
=
_
: + o

o
¹
_
i
A
i
J
iA
(1
i
. A
i
) di
¸
_
i
J
i
(1
i
. A
i
) di
o
S1
o
1
_
1
_
i
A
i
J
iA
(1
i
. A
i
) di
¸
_
i
J
i
(1
i
. A
i
) di
__
dt
+
_
i
A
i
J
iA
(1
i
. A
i
) di
¸
_
i
J
i
(1
i
. A
i
) di
o
¹
dn
¹
+
_
1
_
i
A
i
J
iA
(1
i
. A
i
) di
¸
_
i
J
i
(1
i
. A
i
) di
_
o
1
dn
1
. (99)
67
I..2 Market Beta and Failure of the Conditional CAPM
Using the individual rm and value-weighted market returns presented above, I will show the
failure of the conditional CAPM. For the sake of simplifying the notation, dene

n
=
_
i
A
i
J
iA
(1
i
. A
i
) di
¸
_
i
J
i
(1
i
. A
i
) di
(100)

i
=
J
iA
A
i
¸J
i
. (101)
Using the formulas above, we can calculate the conditional market beta as
,
i
=
co· (d1
n
. d1
i
)
·c: (d1
n
)
=

i

n
o
2
¹
+ (1
i
) (1
n
) o
2
1

2
n
o
2
¹
+ (1
n
)
2
o
2
1
. (102)
Then, the expected instantaneous return implied by the conditional CAPM is not equal to condi-
tional expected returns, that is,
,
i
1 [d1
n
:dt] =

i

n
o
2
¹
+ (1
i
) (1
n
) o
2
1

2
n
o
2
¹
+ (1
n
)
2
o
2
1
[
n
(o

o
¹
+ o
S1
o
1
) o
S1
o
1
] dt
6= [
i
(o

o
¹
+ o
S1
o
1
) o
S1
o
1
] dt = 1 [d1
i
:dt] . (103)
which implies the failure of the conditional CAPM.
More interestingly, it is possible that a rm with higher returns has a lower beta. To see this,
68
rewrite beta as
,
i
=

i
[
n
(o
2
¹
+ o
2
1
) o
2
1
] + (1
n
) o
2
1

2
n
o
2
¹
+ (1
n
)
2
o
2
1
. (104)
While the expected return is increasing in
i
(which leads to a value premium) we have J,
i
,J
i
<
0 if o
2
¹
,o
2
1
< (1
n
) ,
n
, which holds when
n
is sufciently small. Note that
R
i
A
i
J
iX
(1
i
,A
i
)oi
R
i
J
i
(1
i
,A
i
)oi
=
1
R
i
1
i
J
iK
(1
i
,A
i
)oi
R
i
J
i
(1
i
,A
i
)oi
, due to homogeneity, and hence
n
is small when J
1
is particularly high, that
is, during good times. This also implies that the excess market return is countercyclical, because
market return is positively related to
n
.
In this model, the CAPMfails because there is a systematic factor that the CAPMdoes not price
correctly; that is, the shocks to the price of capital goods. The bottom line is that we can generate
the failure of the CAPM with relative ease compared with the effort involved in generating the
correct cross-sectional distribution of stock returns and book-to-market values simultaneously.
References
Abel, Andrew B., and Janice C. Eberly, 1994, A Unied Model of Investment Under Uncertainty,
The American Economic Review 84(5), pp. 1369-1384.
Abel, Andrew B., and Janice C. Eberly, 1995, Optimal Investment with Costly Reversibility,
Working Paper, National Bureau of Economic Research.
Abel, Andrew B., and Janice C. Eberly, 1996, Optimal Investment with Costly Reversibility.
Review of Economic Studies 63(4), 581–593.
Berk, Jonathan B., Richard C. Green, and Vasant Naik, 1999, Optimal investment, growth
options and security returns, Journal of Finance 54, 1153–1607.
Bernardo, Antonio E., Bhangwan Chowdry, and Amit Goyal, 2007, Growth Options, Beta, and
the Cost of Capital, Financial Management 36(2), 5-17.
Bertola, Giuseppe and Ricardo J. Caballero, 1990, Kinked Adjustment Costs and Aggregate
Dynamics, NBER Macroeconomics Annual 1990 5, 237–288.
Black, Fischer, 1972, Capital Market Equilibrium with Restricted Borrowing, Journal of Busi-
69
ness 45(3), 444–455.
Campbell, John Y. and Tuomo Vuolteenaho, 2004, Bad Beta, Good Beta, American Economic
Review 94(5), 1249–1275.
Carlson, Murray D., Adlai J. Fisher, and Ron Giammarino, 2004, Corporate Investment and
Asset Price Dynamics: Implications for the Cross Section of Returns, Journal of Finance 59(6),
2577–2603.
Choi, Jaewon, 2010, What Drives the Value Premium: The Role of Asset Risk and Leverage,
University of Illinois Working Paper
Cohen and Polk, 1998, The Impact of Industry Factors in Asset-Pricing Tests, HBS Working
Paper.
Cooper, Ilan, 2006, Asset Pricing Implications of Nonconvex Adjustment Costs and Irre-
versibility of Investment, Journal of Finance 61(1), 139–170.
Dixit, Avinash K., 1993. The Art of Smooth Pasting. (Harwood Academic Publishers: Chur,
Switzerland).
Fama, Eugene F., and Kenneth R. French, 1992, The Cross-Section of Expected Stock Returns,
Journal of Finance 47, 427–465.
Fama, Eugene F., and Kenneth R. French, 1993, Common Risk Factors in the Returns on Stocks
and Bonds, Journal of Financial Economics 33, 3–56.
Fisher, Edwin O., Robert Heinkel, and Josef Zechner, 1989, Dynamic Capital Structure Choice:
Theory and Tests, The Journal of Finance 44(1), 19–40.
George, Thomas J. and Chuan-Yang Hwang, 2010, A Resolution of the Distress Risk and
Leverage Puzzles in the Cross Section of Stock Returns, Journal of Financial Economics, forth-
coming.
Gomes, Joao F., and Lukas Schmid, 2010, Levered Returns, Journal of Finance 65(2), 467-494.
Grinblatt, Mark, and Sheridan Titman, 2001. Financial Markets and Corporate Strategy, 2nd
Edition (McGraw-Hill).
Hall, Robert E., 2004, Measuring Factor Adjustment Costs, Quarterly Journal of Economics
119(3), 899–927.
Hayashi, Fumio, 2000. Econometrics (Princeton University Press, Princeton, NJ.).
Hecht, Peter, 2000, The Cross Section of Expected Firm (not equity) Returns, HBS Finance
Working Paper No. 03-044.
Johnson, Timothy, 2004, Forecast Dispersion and the Cross Section of Expected Returns, Jour-
nal of Finance 59(5), 1957-1978.
Lintner, John, 1965, The Valuation of Risk Assets and the Selection of Risky Investments in
Stock Portfolios and Capital Budgets, Review of Economics and Statistics 47(1), 221–245.
Livdan, Dmitry, Horacio Sapriza, and Lu Zhang, 2009, Financially Constrained Stock Returns,
Journal of Finance 64(4), 1827–1862.
Lettau, Martin, and Jessica A. Wachter, 2007, Why is Long-Horizon Equity Less Risky? A
Duration-Based Explanation of the Value Premium, Journal of Finance 62, 55–92.
70
Merton, Robert C., 1973, An Intertemporal Capital Asset Pricing Model, Econometrica 41(5),
867–887.
Miao, Jianjun, 2005, Optimal Capital Structure and Industry Dynamics, Journal of Finance
60(6), 2621-2659.
Modigliani, Franco, and Merton H. Miller, 1958, The Cost of Capital, Corporation Finance and
the Theory of Investment, The American Economic Review 48(3), 261–297.
Penman, Stephen H., Scott A. Richardson, and Irem Tuna, 2007, The Book-to-Price Effect in
Stock Returns: Accounting for Leverage, Journal of Accounting Research 45(2), 427–467.
Sharpe, William F., 1964, Capital Asset Prices: A Theory of Market Equilibrium under Condi-
tions of Risk, Journal of Finance 19(3), 425–442.
Taylor, Jack, 2003, Corporation Income Tax Brackets and Rates,1909–2002, Statistics Income
Bulletin, Fall.(http://www.irs.gov/pub/irs-soi//99corate.pdf).
Welch, Ivo, 2004, Capital Structure and Stock Returns, Journal of Political Economy 112(1),
106–131.
Zhang, Lu, 2005, The Value Premium, Journal of Finance 60(1), 67–103.
71
Nontechnical Summary
Motivation and Approach
Firms with a high ratio of book value of equity to market value of equity (value rms) earn higher
expected stock returns than rms with a low book-to-market equity ratio (growth rms). How-
ever, conventional wisdom tells us that growth options should be riskier than assets-in-place and
should therefore have higher expected returns than value rms, which derive their value from
assets-in-place. Additionally, Fama and French (1992) have shown that portfolios of stocks with
different book-to-market ratios have similar riskiness as measured by the standard capital asset
pricing model (CAPM) of Sharpe (1964), Lintner (1965), and Black (1972). This phenomenon,
known as the "value premium puzzle," helped the Fama and French model replace the CAPM as
the benchmark in the asset pricing literature.
This paper presents a dynamic model of the rm with risk-free debt contracts, investment irre-
versibility, and debt restructuring costs, in order to analyze the effects of nancial leverage on in-
vestment and to explain the cross-sectional differences in equity returns. Investment irreversibility
is modeled as in Abel and Eberly (1996), i.e., the resale value of capital is less than the purchase
price of capital. The nancing decisions are similar to those in Fischer, Heinkel, and Zechner
(1989) and Gomes and Schmid (2010), who add capital restructuring costs to the standard tradeoff
theory of capital structure, whereby a rm chooses its nancing policy by balancing the costs of
bankruptcy against the benets of debt, such as tax shields due to interest payments. The model
developed in this paper assumes that rms benet from the tax shield of debt, as in the tradeoff
72
theory, and that they face additional costs at the time of debt restructuring. However, in this paper
debt has two properties distinct from its properties in previous papers: it is risk free and endoge-
nously limited by the lenders to a certain fraction of capital to ensure that the rm honors its debt
payment.
The paper then calibrates the model, drawing on data from the literature and estimating the
remaining parameters using maximum likelihood, based on the long-run stationary distribution
of book-to-market values from the Compustat database. The model captures several facts in the
corporate nance and asset pricing literature in a parsimonious and tractable way and provides
testable implications.
Main Results
1. An important property of the model is that book leverage, that is, the fraction of total capital
supplied by lenders, is state independent. Book leverage is endogenously determined in a manner
that ensures that the rm's value is non-negative even in the worst-case scenario so that the rm
avoids bankruptcy and debt remains risk-free. This worst-case scenario is independent of the state
variables, and hence a revision of the debt agreement at a later date would lead to the same level
of leverage. Thus, it is not optimal for a rm to change its book leverage once it is set. As a result,
book leverage remains the same across rms with different book-to-market equity ratios, whereas
market leverage differs signicantly. This nding is supported by the data. Moreover, because the
level of debt is constant when the rm does not invest, the rm's market debt-to-equity ratio varies
closely with uctuations in its own stock price. This implication of the model is in line with the
73
results of Welch (2004), who nds that U.S. corporations do little to counteract the inuence of
stock price changes on their capital structures.
2. The paper also compares the debt capacity implied by the debt agreement that induces the
no-bankruptcy condition with the standard collateralized debt agreement that limits debt to the
resale value of capital and nds that the former allows greater debt capacity than the latter and
hence is preferred by the rm.
3. The paper shows that investment irreversibility without leverage causes a growth premium
rather than a value premium, i.e., it causes a negative value premium. The rm's investment
opportunity is a call option, because the rm has the right, but not the obligation, to buy a unit of
capital at a predetermined price. As is known from the nancial options literature, when the price
of the underlying security rises and falls, the price of the call option rises and falls at a greater
rate. This suggests that the value of a growth option, that is, the call option to invest, should be
more responsive to economic shocks than the assets-in-place. Therefore, growth options increase
the riskiness of the rm. Similarly, the disinvestment opportunity is a put option, because the rm
has the right, but not the obligation, to sell a unit of capital at a predetermined price. The value
of this put option is negatively related to the value of the underlying asset, because the gain from
exercising it is higher for less productive rms. Therefore, the disinvestment option provides the
value rms that have low productivity with insurance against downside risk and hence reduces
their riskiness. This proposition is different from the recent literature, for example, Zhang (2005)
and Cooper (2006), which presents investment irreversibility as the source of the value premium.
4. The paper nds that nancial leverage affects stock returns directly, through its effect on
74
equity risk, as in Modigliani and Miller (1958), and indirectly, through its effect on business risk,
by inuencing investment decisions. These two channels have opposing effects on the relationship
between book-to-market ratios and stock returns: The Modigliani and Miller channel increases
the equity risk and stock returns of value rms because they have higher market leverage. The
business risk channel increases the effective degree of irreversibility faced by the shareholders and
hence decreases the value premium. However, the Modigliani-Miller effect strongly dominates
the investment channel and explains the major share of the value premium. The paper also pro-
vides empirical analysis consistent with the claim that market leverage is the main source of value
premium.
5. The model is calibrated using maximum likelihood estimation and the long-run distribution
of book-to-market values from the Compustat database. The model is then simulated using the
calibrated values in order to calculate the stock returns and market leverage implied by the model.
The simulation results show that the model ts the data well.
6. Although nancial leverage can explain the major share of the value premium, while in-
vestment irreversibility alone generates instead a growth premium, investment irreversibility still
contributes importantly to improving the model's t with the data, by generating a wide range of
book-to-market values.
7. The model provides two new testable implications: Because nancial leverage increases
effective investment irreversibility through the interest tax-shield, rms with a greater interest tax-
shield invest as if they faced greater investment irreversibility. Moreover, the relationship between
investment irreversibility and stock returns in the model suggests that industries with a higher
75
degree of investment irreversibility, after taking leverage and the interest tax shield into account,
should have a weaker relationship between book-to-market and stock returns.
To summarize, this paper makes a number of contributions to the growing literature that tries
to link corporate decisions to asset returns. First, the debt capacity of the rm is endogenously
determined. Second, the closed-form solution of the model identies explicitly how investment
irreversibility, nancial leverage, and their interaction affect the cross-section of stock returns.
Third, the paper provides a novel calibration exercise that ts the model to the distribution of
book-to-market values, and the calibrated model captures the distribution of book leverage, market
leverage, and stock returns reasonably well. Fourth, the paper provides both theoretical and em-
pirical analysis consistent with the claim that market leverage can explain the major portion of the
value premium. Finally, the model also provides two new testable implications.
76