Grinblatt & Liu 2008 - Debt Policy, Corporate Taxes, and Discount Rates (JET)

Journal of Economic Theory 141 (2008) 225254
www.elsevier.com/locate/jet
Debt policy, corporate taxes, and discount rates
Mark Grinblatt
a
, Jun Liu
b, c, d,,1
a
UCLA Anderson School of Management, USA
b
Cheung Kong Graduate School of Business, China
c
Southwestern University of Finance and Economics, China
d
Rady School at UCSD, USA
Received 19 October 2004; nal version received 7 November 2006; accepted 12 September 2007
Available online 22 October 2007
Abstract
This paper applies the standard risk-neutral valuation framework to tax shields generated by dynamic
debt policies. We derive a partial differential equation (PDE) for the value of the debt tax shield. For a
class of dynamic debt policies that depend on the assets free cash ows, value, and past performance, we
obtain closed-form solutions for the PDE. We also derive the tax-adjusted cost of capital for free cash ows
and analyze the conditions under which the weighted average cost of capital is an appropriate discount
rate. Finally, we derive closed-form solutions for equity betas, which differ from the formulas that have
traditionally been used to lever and unlever equity betas.
2007 Elsevier Inc. All rights reserved.
JEL classication: G1; G3
Keywords: WACC; Discount rate; Tax shields; Corporate tax
1. Introduction
An intrinsic difculty associated with the valuation of debt tax shields is identifying the risk
of the tax deductions arising from the stream of future debt interest expenses. As a consequence,
the rate at which one discounts the future stream of interest-related tax shields, and hence the
value of those tax shields, has eluded prior research, except for the simplest of cases. These cases
impose stringent restrictions on the cash ow process and debt policy to circumvent the complex
issue of risk and valuation. Among these are the models of [17,16].
Corresponding author. Rady School at UCSD, USA. Fax: +1 858 534 0745.
E-mail address: junliu@ucsd.edu (J. Liu).
1
On leave from the Rady School at UCSD, USA.
0022-0531/$ - see front matter 2007 Elsevier Inc. All rights reserved.
doi:10.1016/j.jet.2007.09.009
226 M. Grinblatt, J. Liu / Journal of Economic Theory 141 (2008) 225254
The Modigliani and Miller (M&M) debt policy is one where the debt level is constant, perpetual,
and not subject to default (although the formulas derived by M&M extend as is to the case where
default always leads to zero recovery). This debt policy implies that one can discount the streamof
future interest-based tax shields at the risk-free rate. If the tax rate is constant, as they assume, the
debt tax shields present value is necessarily proportional to the present value of the debt because
the cash ow stream from the tax shield is proportional to the cash ow stream from the debt.
Here, since the constant of proportionality is the corporate tax rate, the present value of the debt
tax shield is the product of the corporate tax rate and the present value of the debt. Modigliani and
Miller [17] also use this model to develop formulas for discount rates that account for the value
of the tax shield when cash ows have no tendency to grow.
The interesting case studied by Miles and Ezzell focuses on the dynamic issuance of perpetual
risk-free debt. This case assumes: (1) the free (or unlevered) cash ow (the after-tax cash ow
that stems directly fromthe real asset, which is unaffected by the nancing mix) follows a random
walk with no drift; (2) the cash ow stream is valued by applying a constant discount rate; (3)
the debt-to-asset ratio is constant; and (4) debt has no default. Under these assumptions, the cash
ow from each dates tax shield is of the same risk as the one period lagged free cash ow. As
Grinblatt and Titman [9], and Brealey and Myers [2] point out, in the continuous time limit of
this model, the cash ow stream from future debt-related tax deductions is of the same risk as
the free cash ow stream and thus can be discounted at the same rate. Miles and Ezzell, as well
as the standard textbooks, present formulas analogous to those in Modigliani and Miller for this
alternative debt policy.
The actual debt policies of rms tend to deviate from those specied by the ModiglianiMiller
and MilesEzzell models.
2
In these cases, the literature in nance offers little guidance on
valuation. For example, when debt interest is nonlinear in the free cash ow, the value of the debt
tax shield and the specication of the after-tax discount rate for valuing the asset are not known.
This paper proposes a no-arbitrage valuation of debt tax shields for a class of Markovian debt
policies. The value of the debt tax shield for dynamic debt policies can be derived as the solution
to a partial differential equation (PDE). Our approach allows for stochastic cash ow, stochastic
discount rates, and realistic recovery assumptions on default. For a large class of dynamic debt
policies, we can reduce the PDE to a system of ordinary differential equations, which are easily
solved numerically. With more restrictive assumptions, but still with more generality than exists
in the literature, we obtain the value of the tax shield as a closed-form expression.
The discount rate for free cash ows that accounts for the debt tax shield is also of critical
importance, both to practitioners and researchers. We showwhat adjustments are needed to convert
the weighted average cost of capital (WACC) to an appropriate discount rate. Such adjustments are
almost always needed as the WACC is an appropriate discount rate only in the ModiglianiMiller
and MilesEzzell cases, or in some linear hybrid of these two well-known cases.
Finally, for general debt policies, we derive a formula that characterizes the equity beta as a
function of the leverage ratio and the unlevered asset beta. This formulas special cases include
the standard textbook formulas of Hamada [10] and Miles and Ezzell [16], which are associated
with the ModiglianiMiller and MilesEzzell models, respectively.
Our approach differs from that found in prior research on debt tax shield valuation. In lieu of
strong restrictions on cash ows, discount rates, debt policy, or recovery on default, we use the
risk-neutral pricing framework in a continuous-time setting, following the option pricing approach
of BlackScholes [21] and Merton [22], and the insights of Cox and Ross [3], Ross [18], and
2
See, for example, Donaldson [4], Baker and Wurgler [1], Graham and Harvey [7], and Kayhan and Titman [13].
M. Grinblatt, J. Liu / Journal of Economic Theory 141 (2008) 225254 227
Harrison and Kreps [11]. The risk-neutral valuation methodology of asset pricing theory has been
applied before in a corporate setting, perhaps most elegantly by Ross [18]. Our approach, while
similar in spirit, uses this methodology to derive solutions that apply specically to the debt tax
shield. Moreover, our closed-form solutions yield economic insights and intuition about the value
of the debt tax shield. One example of this is a heuristic description of how the value of the debt
tax shield, as well as the appropriate discount rate, vary with debt policies that can be viewed as
weighted averages of the ModiglianiMiller and MilesEzzell debt policies.
The goal of our paper is to assess whether broad classes of exogenously specied debt poli-
cies lend themselves to tractable valuation solutions. The optimal debt policy, while important
for deciding which of our debt policies should be focussed on, is beyond the scope of this pa-
per. The optimal state-contingent debt path is a considerably different problem, and a generally
intractable one, once the economic framework becomes slightly less restrictive than that in the
extant literature.
3
Section 2 of the paper develops a general approach for valuing debt tax shields. It also analyzes
a large class of debt policies which have closed-form solutions for the tax shield and presents
two larger classes of cases for which numerical computation of the value of the debt tax shield is
trivial. Section 3 examines the weighted average cost of capital and relates it to valuation. It also
characterizes how the WACC is affected by dynamic debt policies and studies when the WACC
can be used to obtain valuations that properly account for the value of the debt tax shield. Finally,
this section derives closed-form solutions for tax-adjusted discount rates that generate the correct
tax-adjusted valuations of free cash ow streams. In most cases outside of the ModiglianiMiller
and MilesEzzell frameworks, we show that these discount rates differ from the WACC. Section 4
analyzes how to lever and unlever equity betas and equity risk premia for arbitrary debt policies.
Section 5 concludes the paper.
2. The valuation of debt tax shields
In a dynamically complete market, two assets with payoffs driven by the same source of
uncertainty are instantaneously perfectly correlated. This allows them to be valued in relation to
one another. Just as an option (whether listed or synthetically created from a dynamic strategy)
is valued in relation to its underlying security, so too can a debt tax shield be valued in relation to
the unlevered asset it is associated with.
In the continuous-time framework, it is irrelevant that the unlevered asset may not be directly
traded as a security. Just as we know the no-arbitrage value one should place on a synthetic option
that is created from a dynamic portfolio strategy in a stock and a bond, we know the value of what
effectively is a levered asset (and hence a tax shield) as a function of the value of an otherwise
identical unlevered asset, and vice versa. So long as the uncertainty behind the debt policy that
generates the tax shield is tied only to uncertainty in the free cash ows, debt tax shields are
simply derivatives. For this reason, most of the paper assumes that the free cash ow, the after-tax
cash ow that would be generated in the absence of debt nancing, satises a general Markov
diffusion process.
In our framework, it also makes no difference whether we treat the unlevered asset or the levered
asset or the tax shield as the underlying security. We are simply valuing these cash ow claims in
3
Optimal debt levels when the bankruptcy costs of debt are weighed against tax benets and costs are found in Leland
[15], Green and Hollied [8], and Ross [19].
relation to one another. For expositional convenience, and economic intuition that is consistent
with the extant literature, we use the unlevered asset as the more primitive underlying security.
An asset that is levered with risk-free debt has two sources of after-tax cash ow at date t: (1)
the free (or unlevered) cash ow, X
t
dt , which is unaffected by the assets nancing mix; and (2)
the ow from the debt interest tax shield, t
c
D
t
r
f
dt , which is the product of the tax rate t
c
and
the debt interest payment D
t
r
f
dt .
4
Because valuation is linear, the value of a levered asset is the
sum of the values of its two cash ow components.
For simplicity (and without loss of generality), we let one Brownian motion, B, drive the
uncertainty. That is, between dates 0 and T, with T possibly innite,
dX
t
= g(t, X)Xdt +o(t, X)XdB
t
.
With this assumption, the market is dynamically complete with two tradable assets.
5
This means
that a dynamic trading strategy can transform a levered asset into an unlevered asset, and vice
versa. Similarly, knowing the value of an asset for any given debt policy allows us to compute the
value of its tax shield for all debt policies. Solutions can be found with a variety of mathematically
equivalent approaches, but the most popular method involves the solution of a differential equation
generated by Itos Lemma and the principle of no arbitrage.
The continuous-time setting, described above, allows valuation of almost any derivative, in-
cluding tax shields, by applying the well-known no-arbitrage principle. For expositional clarity,
the model initially assumes no default on debt and that the debt is short-term. We also assume that
there are no personal taxes or other market frictions associated with debt beyond the corporate
tax.
6
One can easily extend the model to account for these kinds of frictions, as we will discuss
shortly.
2.1. Debt policy and levered asset valuation: the PDE
We begin by studying the valuation of a levered asset under a general class of debt policies.
The debt policy D can be any differentiable function of time t, free cash ow X, levered asset
value V
L
, and history H. That is,
D
t
= D(t, X
t
, V
L
t
, H
t
).
These general debt policies can depend, in a quite complicated manner, on the history of the asset,
such as past cash ows, past debt values, past asset values, in addition to current cash ow and
4
In order for the levered asset to have the same investment policy in the presence of debt, we assume, without loss of
generality, that the ow from the interest-based tax shield is paid out. It could be retained in a risk-free interest bearing
account and distributed later, but this has tax consequences for the rm. In essence, such retention amounts to negative
debt and it is the net debt policy for which we are computing the tax shield. Given this denition of how to account
for debt, and appropriate care taken to avoid double counting when this cash is eventually distributed, our results apply
irrespective of whether cash is retained or paid out. Similarly, our valuation equations appear to assume that free cash ow
is distributed to debt and equity holders. However, except for the tax consequences of the rms unnecessary retention of
liquid assets, as described above and which is easily adjusted for, the retention of future free cash ows does not affect an
assets present value.
5
This can be a risky and a risk-free security or two risky securities with uncertainty driven by B
t
.
6
Personal taxes are clearly important for asset valuation and debt policy, as Green and Hollield [8] prove theoretically
and document empirically. Their paper analyzes the optimal capital structure for a rm with a ModiglianiMiller debt
policy, bankruptcy costs, and a cash distribution policy to equity holders that is sensitive to the economic effects of the
corporate and capital gains taxes. Ross [19] studies the cost of capital and optimal capital structure with both corporate
and personal taxes in a setting with constant debt payments in no-default states.
current asset value.
7
We only require that the history dependence at a given date t be summarized
by additional date t state variables. This allows us to maintain the Markovian setting. Note, too,
that debt policy depends on the value of the levered asset, which in turn will depend on debt
policy. However, our valuation approach simultaneously determines the value of both the debt tax
shield and the value of the levered asset. (This is also true, for example, in Miles and Ezzell.)
Without loss of generality, we simplify notation by treating these state variables as the single
variable H
t
.
8
As long as the uncertainty associated with the path of X spans the relevant state
space for H, we will still be able to value the debt tax shield as a function of the value of the
unlevered asset. To maintain this desirable property, we assume that H
t
satises the diffusion
dH
t
=j
H
(t, X
t
, D
t
, V
L
t
) dt +o
H
(t, X
t
, D
t
, V
L
t
) dX
t
=j
h
(t, X
t
, D
t
, V
L
t
) dt +o
h
(t, X
t
, D
t
, V
L
t
) dB
t
,
where
j
h
(t, X
t
, D
t
, V
L
t
) = j
H
(t, X
t
, D
t
, V
L
t
) +o
H
(t, X
t
, D
t
, V
L
t
)g(t, X
t
)X
t
and
o
h
(t, X
t
, D
t
, V
L
t
) = o
H
(t, X
t
, D
t
, V
L
t
)o(t, X
t
)X
t
.
Recognize that the functional form of the exogenously specied j
h
and o
h
can be quite general,
which allows us to analyze many empirically relevant debt policies.
Given this description of debt policy and history, it follows that the date t value of the levered
asset, V
L
t
= V
L
(t, X
t
, H
t
), depends on the current date, cash ow, and history. Moreover, if the
free cash owstreamterminates at date T (essentially, becomes 0 at date T and forever thereafter),
the functional formof the valuation function will be inuenced by the proximity to the termination
date.
In the absence of arbitrage, dynamic completeness implies that this value necessarily satises
the partial differential equation (PDE)
*V
L
*t
+(g p)X
*V
L
*X
+
1
2
o
2
X
2
*
2
V
L
*X
2
+(j
h
po
h
)
*V
L
*H
+
1
2
o
2
h
*
2
V
L
*H
2
+oo
h
*
2
V
L
*H*X
X = r
f
V
L
(X +r
f
t
c
D), (1)
where we have dropped the arguments of g, o, etc., for notational simplicity. If the asset has a
nite life, the terminal condition is
V
L
(T, X
T
, H
T
) = 0.
This partial differential equation, a familiar extension of the well-known BlackScholes differ-
ential equation, is simply the no-arbitrage condition associated with an asset whose uncertainty is
spanned by the payoff to a dynamic trading strategy in the unlevered asset and a risk-free security.
The p(t, X) term in Eq. (1), (shortened to p for notational simplicity), is the premium per unit
of risk generated by changes in B. In a corporate setting, it would be traditional to think of this
7
Kayhan and Titman [13] have documented the importance of history on leverage ratios.
8
We can also regard H and the drift coefcients as vectors and the diffusion coefcients as a matrix with virtually no
change to any of our equations.
parameter as being determined by the instantaneous discount rate of the unlevered asset. However,
p also can be inferred from the levered assets value (or levered equity value) for any debt policy.
To derive the partial differential equation, note that Itos Lemma implies that the change in the
value of a levered asset plus all distributions of cash ow:
dV
L
t
+(X
t
+r
f
t
c
D
t
) dt
=
*V
L
*t
dt +
*V
L
*X
dX +
*V
L
*H
dH
+
_
1
2
*
2
V
L
*X
2
o
2
X
2
+
*
2
V
L
*H*X
oo
h
+
1
2
*
2
V
L
*H
2
o
2
h
_
dt +(X
t
+r
f
t
c
D
t
) dt
=
_
*V
L
*t
+
*V
L
*X
gX +
*V
L
*H
j
h
_
dt
+
_
1
2
*
2
V
L
*X
2
o
2
X
2
+
*
2
V
L
*H*X
oo
h
X +
1
2
*
2
V
L
*H
2
o
2
h
+X
t
+r
f
t
c
D
t
_
dt
+
_
*V
L
*X
+
*V
L
*H
o
H
_
oXdB.
The analogous equation for an otherwise identical unlevered asset with date t value V
U
t
=
V
U
(t, X) is
dV
U
t
+X
t
dt =
_
*V
U
*t
+
*V
U
*X
gX +
1
2
*
2
V
U
*X
2
o
2
X
2
+X
t
_
dt +
*V
U
*X
Xo dB
=
_
V
U
r
f
+
*V
U
*X
Xp
_
dt +
*V
U
*X
Xo dB,
where p, dened by its placement above, is simply a convenient symbol for a scaling of the risk
premium attached to dB. We can express p in terms of the instantaneous unlevered cost of capital,
r
U
t
as
p(t, X) =
r
U
t
r
f
* ln(V
U
t
)
* ln(X
t
)
. (2)
Clearly, instantaneous changes inV
L
andV
U
are perfectlycorrelated. Thus, toprevent arbitrage,
the ratios of the risk premia per dollar invested in the levered and unlevered assets must be
proportional to the risk born per dollar invested in each of the assets. This implies
*V
L
*t
+
*V
L
*X
gX +
*V
L
*H
j
h
+
1
2
*
2
V
L
*X
2
o
2
X
2
+
*
2
V
L
*H*X
oo
h
X +
1
2
*
2
V
L
*H
2
o
2
h
+X
t
+r
f
t
c
D
t
r
f
V
L
=
_
*V
L
*X
+
*V
L
*H
o
H
_
Xp,
which, when rearranged, gives us Eq. (1). In analogous fashion, the value of the tax-shield =
V
L
V
U
satises the PDE
*
*t
+(g p)X
*
*X
+
1
2
o
2
X
2
*
*X
2
+(j
h
po
H
)
*
*H
+
1
2
o
2
h
*
2
*H
2
+oo
h
*
2
*H*X
X = r
f
r
f
t
c
D. (3)
2.2. Discussion
The interpretation of Eq. (1) is straightforward. The left side represents the expected rate of
change in the assets value under the risk-neutral measure. On the right side, the termin parentheses
is the sumof the free cash owand the cash owfromthe tax-shield. Thus, if we divide both sides
by V
L
, our PDE is simply the well-known result that in the absence of arbitrage, the asset value
under the risk-neutral measure has an expected rate of growth equal to the difference between the
risk free return and the dividend yield (from payouts of the free cash ow and the tax-shield).
As long as the short-term risk-free rate and the discount rate (which can be any function of
the information set) for the cash ows of an otherwise identical unlevered asset are specied,
the standard continuous-time valuation methodology prices any future cash ow contingent on
the same source of uncertainty. This includes the cash ow from the unlevered asset, the levered
asset, or a tax-shield from a complex, yet realistic, debt policy.
When markets are complete, the risk-neutral drift of the cash ow X is uniquely inferred from
the price dynamics of any traded asset with uncertainty driven by B
t
. The useful economic intuition
that a portfolio of traded securities tracks the cash owproduced by the debt tax shield also applies
here. In such a complete markets setting, where we only try to capture relative values of claims
to levered and unlevered after-tax cash ows, there is no reason to specify what the traded assets
are. In principle, any nancial instruments that have exposure to B
t
can be used to determine the
unique risk-neutral drift. As a practical matter, it is convenient to think of p as a function of the
risk-premium on the unlevered asset, whether that asset is traded directly or created indirectly
from a dynamic strategic in a levered companys equity and risk-free debt.
Even in the continuous-time setting, markets can be incomplete because debt policy is affected
by stochastic interest rates, stochastic volatility, industry structure and competition, macroeco-
nomic conditions, and managerial discretion. When markets are incomplete, the valuation equa-
tions shown above are still valid, but the risk-neutral drift for X in Eq. (1) is non-unique.
9
Hence,
a violation of the pricing equation for a xed set of risk-neutral drifts does not necessarily imply
arbitrage. However, valuations that are not consistent with any choice of risk-neutral drifts would
imply arbitrage, as the fundamental theorem of asset pricing proves.
Our analysis does not account for personal taxes in the equilibrium cost of debt to rms. Ross
[19], in an elegant paper, considers both corporate and personal taxes. He uses a no-arbitrage
approach to value tax shields when rms can default. Furthermore, he derives the optimal capital
structure from balancing the trade-off between the tax shield and bankruptcy costs. The debt tax
shield conditional on no-default is a constant in Ross [19]. We have state-dependent debt tax
shields, but do not consider bankruptcy costs, personal taxes, or optimal capital structure.
While the issue of optimal capital structure is beyond the scope of this paper, minor changes in
notation on the right side of Eq. (1) easily address the bankruptcy cost and personal tax issues. In
these more general cases, we need to replace r
f
in the term r
f
V
L
with the (potentially stochastic)
default-adjusted short rate used in Dufe and Singleton [5] and the cash ow from the debt tax
shield, r
f
t
c
D, with the (potentially state-contingent) effect of both personal and corporate taxes
on the rms after-tax cost of debt in the absence of default. The derivation of the PDE, a non-
trivial extension of the Dufe and Singleton [5] default model, is available from the authors upon
request.
In principle, these partial differential equations (with or without extensions to bankruptcy and
more general tax issues) can be solved. However, without further restrictions, these differential
9
See Harrison and Kreps [11] for a discussion of this point in more general valuation settings.
equations are difcult to solve, even numerically. Hence, the remainder of this section explores
cases where solutions are insightful or, from a numerical perspective, quickly attainable. Essen-
tially, whenever we can transformthe PDEinto an ordinary differential equation (ODE), numerical
solutions are easily found. We explore two such classes of cases. In the rst, all of the models
parameters depend only on the contemporaneous level of the free cash ow, X. Here, because there
is no time dependence, Eq. (1) reduces to an ODE in X. In the second class of cases, which we
refer to as Additively Separable Assets, the levered assets value is additively separable in a set
of arguments, which consists of H and a nite collection of real powers of X: X
z
1
, X
z
2
, . . . , X
z
N
.
The coefcients of these arguments may be time dependent. The additively separable class of
cases is particularly interesting for its ability to generate closed-formsolutions for the value of the
debt tax shield. These apply to both nite-lived and perpetual assets. They arise whenever the N
cash ow coefcients of the value additive functions for debt and history are growing at constant
exponential rates and the remaining coefcients are constant. Given this level of generality, it
appears as if our closed-form solutions could generate fairly good approximations for the value
of a debt tax shield for many interesting debt policies.
The closed-formsolutions only require state-independent parameters for the cash owprocess,
the short rate, and the tax rate. For simplicity in notation, we present results for the case where
parameters of the cash ow process, the short rate, and the tax rate are constant.
2.3. Cash ows and debt with time-independent parameters
When the parameters of the dynamic process for X do not depend explicitly on time t (that is,
g(t, X) = g(X), o(t, X) = o(X), and p(t, X) = p(X)) and the debt policy D does not depend
on either time t or the history H
t
, the asset value V
L
depends only on the contemporaneous cash
ow level, X. In this case, the PDE for the value of the levered asset (1) becomes the second order
ODE:
(g(X) p(X))X
*V
L
*X
+
1
2
o
2
(X)X
2
*
2
V
L
*X
2
r
f
V
L
= (X +r
f
t
c
D(X)),
whichis trivial tosolve numericallyfor anyspecicationof g(X), o(X), andp(X).
10
Special cases
with closed-form solutions include the continuous-time versions of the ModiglianiMiller debt
policy (g(X) = 0, p(X) = p and D(X) = D implying V
U
= X/(r
f
+p) and V
L
= V
U
+t
c
D)
and the MilesEzzell debt policy (g(X) = 0, p(X) = p and D(X) = d
x
X, with d
x
constant,
implying V
U
= X/(r
f
+p) and V
L
= V
U
+t
c
Dr
f
/(r
f
+p)). We defer further discussion of this
as the class of debt policies analyzed next also includes the ModiglianiMiller and MilesEzzell
models as special cases.
2.4. Additively separable assets: numerical solutions
Additively separable assets have tax shields with values that are additively separable linear
functions of history, H, and any set of real powers of the cash ow, X
z
. Simple examples of
10
Any two boundary conditions, which implicitly determine the debt level in all states of the world, determine a unique
solution to the differential equation. Hence, specifying the debt policy is clearly sufcient for obtaining the levered assets
value.
additively separable assets include the constant coefcient quadratic case,
V
L
t
= V
U
t
+c
x
0
+c
x
1
X
t
+c
x
2
X
2
t
,
which is generated by the constant coefcient history-independent quadratic debt policy
D
t
= d
x
0
+d
x
1
X
t
+d
x
2
X
2
t
and the constant coefcient square root case,
V
L
t
= V
U
t
+c
x
0
+c
x
1/2
_
X
t
,
which is generated by the constant coefcient history-independent square root debt policy
D
t
= d
x
0
+d
x
1/2
_
X
t
.
In Leland [15] and Ross [19], the optimal capital structure has a debt level that is a fractional
power of the underlying cash ow.
A more complicated case arises when the debt level is history dependent with history given by
H
t
=H
0
e
m
h
t
1
2
(l
h
)
2
t l
h
B
t
+m
x
_
t
0
e
m
h
(t s)
l
2
(l
h
)
2
(t s)l
h
(B
s
B
t
)
X
s
ds
+m
d
_
t
0
e
m
h
(t s)
1
2
(l
h
)
2
(t s)l
h
(B
s
B
t
)
D
s
ds
+m
v
_
t
0
e
m
h
(t s)
1
2
(l
h
)
2
(t s)l
h
(B
s
B
t
)
V
L
s
ds
+l
x
_
t
0
e
m
h
(t s)
1
2
(l
h
)
2
(t s)l
h
(B
s
B
t
)
X
s
dB
s
+l
d
_
t
0
e
m
h
(t s)
1
2
(l
h
)
2
(t s)l
h
(B
s
B
t
)
D
s
dB
s
+l
v
_
t
0
e
m
h
(t s)
1
2
(l
h
)
2
(t s)l
h
(B
s
B
t
)
V
L
s
dB
s
.
In this special case, the diffusion process for H
t
satises
dH
t
= (m
x
X
t
+m
d
D
t
+m
v
V
L
t
m
h
H
t
) dt +(l
x
X
t
+l
d
D
t
+l
v
V
L
t
l
h
H
t
) dB
t
and thus has drift and volatility of
j
h
= m
x
X
t
+m
d
D
t
+m
v
V
L
t
m
h
H
t
, o
h
= l
x
X
t
+l
d
D
t
+l
v
V
L
t
l
h
H
t
.
This history process, with the ms and ls constant,
11
when combined with an analogous functional
form for the debt process, leads to a closed-form additively separable solution for the value of the
debt tax shield, as we show in the next subsection.
11
The special case, m
d
= 0 while m
x
= m
v
= l
d
= l
x
= l
v
= 0, says that the debt depends on past debt levels, as
empirical work seems to suggest.
The most general class of additively separable assets has history diffusion and debt policy of
the form:
dH
t
=
z
m
x
z
(t )X
z
t
+m
v
(t )V
L
t
+m
d
(t )D
t
m
h
(t )H
t
dt
+
z
l
x
z
(t )X
z
t
+l
v
(t )V
L
t
+l
d
(t )D
t
l
h
(t )H
t
dB
t
(4)
and
D
t
=
z
d
x
z
(t )X
z
t
+d
v
(t )V
L
t
+d
h
(t )H
t
, (5)
along with risk premia, p(t ), free cash ow growth rate, g(t ), and volatilities, o(t ) and o
H
(t ),
that depend only on time. An implication of g(t ) and p(t ) depending only on time is that the
price-earnings ratio for an unlevered asset, y
U
t
= V
U
t
/X
t
, depends only on time.
12
One can show that the value of a levered asset with debt policy and history satisfying these
properties is of the additively separable form
V
L
t
= V
U
t
+
z
c
x
z
(t )X
z
t
+c
h
(t )H
t
and the PDE, Eq. (1), is of the form
13
z
*c
x
z
*t
X
z
+
*c
h
*t
H +(g p)
z
zc
x
z
X
z
+
1
2
o
2
z
c
x
z
z(z 1)X
z
+
_
k
h
k
d
d
h
(k
v
+k
d
d
v
)c
h
_
H +
z
_
k
x
z
+k
d
d
x
z
+(k
v
+k
d
d
v
)c
x
z
_
X
z
+(k
v
+k
d
d
v
)V
U
c
h
r
f
z
c
x
z
X
z
+c
h
H
= r
f
t
c
z
d
x
z
X
z
+d
v
V
U
+
z
c
x
z
X
z
+c
h
H
+d
h
H
,
with k
q
= m
q
pl
q
for q {x, d, v, h}.
12
To prove this, note that X
t
drops out of the ratio
V
U
t
X
t
=
_
T
t
e
_
s
t
g(c)dc
e
_
s
t
r
U
c
dc
ds.
13
Note that many of terms involving V
U
cancel because of the no arbitrage PDE for V
U
.
Equating the coefcients of H and X
z
on each side produces N + 1 ordinary differential
equations with N being the number of powers of X that appear in the debt and history equations:
dc
x
z
dt
+(g p)zc
x
z
+
1
2
o
2
c
x
z
z(z 1)
+
_
k
x
z
+k
d
d
x
z
+(k
v
+k
d
d
v
)c
x
z
+(k
v
+k
d
d
v
)y
U
t
o
z,1
_
c
h
r
f
c
x
z
+r
f
t
c
_
d
x
z
+d
v
c
x
z
+d
v
y
U
t
o
z,1
_
= 0,
dc
h
dt
_
k
h
k
d
d
h
(k
v
+k
d
d
v
)c
h
_
c
h
r
f
c
h
+r
f
t
c
_
d
v
c
h
+d
h
_
= 0,
with o
z,1
a binary variable that takes on the value 1 if z = 1 and 0 otherwise,
14
and with the
terminal condition given by
c
x
z
(T ) = c
h
(T ) = 0.
This system of Riccati equations is easily solved numerically. These equations also are easily
altered to allow for default and more general tax shields. However, there are large classes of cases
that have closed-form solutions. We explore these below.
2.5. Additively separable assets with closed-form solutions
Suppose that each of the coefcients k
d
(t ), k
v
(t ), k
h
(t ), d
v
(t ), d
h
(t ) are constant and
k
x
z
(t ) = k
x
z
(0) e
g
k
z
t
,
d
x
z
(t ) = d
x
z
(0) e
g
d
z
t
,
with the constant growth parameters g
k
z
and g
d
z
possibly 0. For expositional clarity, we also assume
that the mean and volatility of the free cash ow growth rate, as well as the market price of risk,
are constant. That is, g(t, X) = g, o(t, X) = o, and p(t, X) = p. This allows us to express the
value of an unlevered asset with the growing annuity formula, as in the Gordon Growth Model:
V
U
t
=
X
t
r
U
g
_
1 e
(r
U
g)(T t )
_
. (6)
The Gordon growth assumptions imply that
* ln(V
U
t
)
* ln(X
t
)
= 1 and that the risk premium on the
unlevered asset r
U
r
f
= p.
We could allow g, o, and p to be deterministic functions of time and still achieve solutions
similar to those developed below but at the cost of expressions with confusing sets of integrals
in them. As this discussion is about the valuation of tax shields for complex debt policies, and
not about the complexities of valuation in a no-tax setting, we opt for an approach that makes the
latter valuation as uncomplicated as possible.
14
Without loss of generality, and only for notational simplicity, we assume that one of the powers of z is z = 1 if one
of d
v
, k
v
, d
x
1
, or k
x
1
is nonzero. Also, note that if z
i
= 0, we have an exponentially growing constant term. We explore a
special case with this feature later.
Under these assumptions, the system of Riccati equations is solved by
c
h
(t ) =
1 e
b(T t )
1
c
h
c
h
+b/a
e
b(T t )
c
h
, (7)
where
15
a = k
v
+k
d
d
v
,
b =
_
(k
h
k
d
d
h
+r
f
(1 t
c
d
v
))
2
4at
c
r
f
d
h
,
c
h
=
_
b
2
+4at
c
r
f
d
h
b
2a
and for z = z
1
, . . . , z
N
c
x
z
(t ) =r
f
t
c
d
x
z
(t )C
1
(g
d
z
) +
r
f
t
c
d
v
r
U
g
o
z,1
_
C
1
(0) e
(r
U
g)(T t )
C
1
(r
U
g)
_
+k
x
z
(t )C
2
(g
k
z
) +k
d
d
x
z
(t )C
2
(g
d
z
)
+
a
r
U
g
o
z,1
_
C
2
(0) e
(r
U
g)(T t )
C
2
(r
U
g)
_
, (8)
where
C
1
(z) =
1
1
c
h
c
h
+b/a
e
b(T t )
_
1 e
(g
x
z
z)(T t )
g
x
z
z
e
b(T t )
c
h
c
h
+b/a
_
1 e
(g
x
z
zb)(T t )
g
x
z
z b
__
,
C
2
(z) =
c
h
1
c
h
c
h
+b/a
e
b(T t )
_
1 e
(g
x
z
z)(T t )
g
x
z
z
e
b(T t )
_
1 e
(g
x
z
zb)(T t )
g
x
z
z b
__
,
with g
x
z
dened by
g
x
z
= r
f
(1 t
c
d
v
) (k
v
+k
d
d
v
)c
h
+(p g)z
1
2
o
2
z(z 1).
This closed-formsolution can be extended to the case of bankruptcy and more general tax shields.
For example, if the instantaneous ow from the tax reduction replaces D in Eqs. (4) and (5), the
solution for g
x
z
above requires that we replace the rst term, r
f
(1 t
c
d
v
), with the difference
between the default-adjusted short rate and d
v
.
Case 1: Perpetual assets. As T , Eq. (7) has
c
h
(t ) = c
h
15
For the tax shield of a nite-lived asset to have a nite value, b, given below, has to be a real number. Also, for history
to be stable, k
h
k
d
d
h
+r
f
(1 t
c
d
v
) has to be positive. Throughout the paper, we assume that parameters satisfy the
transversality conditions so that the debt tax shield is nite.
and the N cash ow coefcients given by Eq. (8) simplify to
c
x
z
(t ) =
k
x
z
(t )c
h
g
x
z
g
k
z
+
(r
f
t
c
+k
d
c
h
)d
x
z
(t )
g
x
z
g
d
z
+
ac
h
+r
f
t
c
d
v
(r
U
g)(r
f
(1 t
c
d
v
) +p g)
o
z,1
since, as T ,
C
1
(z) =
c
h
g
x
z
z
,
C
2
(z) =
1
g
x
z
z
.
This implies
V
L
t
=
_
1 +
ac
h
+r
f
t
c
d
v
r
f
(1 t
c
d
v
) +p g
_
V
U
t
+
z
_
k
x
z
(t )c
h
g
x
z
g
k
z
+
(r
f
t
c
+k
d
c
h
)d
x
z
(t )
g
x
z
g
d
z
_
X
z
t
+c
h
H
t
.
Case 2: Perpetual debt as a function of cash ows only. When T and d
v
= d
h
= 0,
c
h
= 0. In this case, the solution for Case 1 simplies to

c
x
z
(t ) =
r
f
t
c
d
x
z
(t )
g
x
z
g
d
z
implying
V
L
t
= V
U
t
+
z
r
f
t
c
r
f
+(p g)z
1
2
o
2
z(z 1) g
d
z
d
x
z
(t )X
z
t
.
Case 3: Debt that is a linear function of asset value plus constant growth. For this special case,
D
t
= d
x
0
(0) e
g
d
0
t
+d
v
V
L
t
,
with the sensitivity of debt to asset value constant; that is, d
v
(t ) = d
v
. The remaining coefcients
are 0. Note that the constant growth rate component in debt, g
d
0
, may differ from g, the expected
growth rate in cash ows. When d
v
is 0, debt grows at the constant geometric rate of g
d
0
(possibly
0). When d
x
0
(0) = 0, the debt to asset ratio is constant over the life of the asset. Hence, this policy,
as well as the stationary model described in the prior subsection, nests both the ModiglianiMiller
and MilesEzzell debt policies. (Note that if d
v
is nonzero, the expected growth rate in debt is
inuenced both by the expected growth rate in V
L
as well as g
d
0
.)
This is a case where c
h
(t ) = c
h
= C
2
(z) = 0 and the solution reduces to
V
L
t
= V
U
t
+c
x
0
(t ) +c
x
1
(t )X
t
,
with the values for c
x
z
from Eq. (8) becoming
c
x
0
(t ) = r
f
t
c
d
x
0
(t )
_
1 e
(r
f
(1t
c
d
v
)g
d
0
)(T t )
r
f
(1 t
c
d
v
) g
d
0
_
,
c
x
1
(t ) =
r
f
t
c
d
v
r
U
g
_
1 e
(r
f
(1t
c
d
v
)+pg)(T t )
r
f
(1 t
c
d
v
) +p g
_
.
Aparticularlysimple expressionexists for a perpetual leveredasset inCase 3. Here, the coefcients
above imply
16
V
L
t
= V
U
t
+t
c
_
r
f
r
f
(1 t
c
d
v
) g
d
0
d
x
0
(t ) +
r
f
r
f
(1 t
c
d
v
) +p g
d
v
V
U
t
_
. (9)
Note that when g = g
d
0
= d
v
= 0, Eq. (9) is the ModiglianiMiller value,
V
L
t
=V
U
t
+t
c
d
x
0
(t )
=V
U
t
+t
c
D
t
.
When d
v
is 0 but g
d
0
and g are nonzero, we have an extension of the ModiglianiMiller debt
policy that allows for growing debt and free cash ows that are expected to grow. As we will learn
in the next section of the paper on the WACC, the initial weighted average cost of capital, used
as a discount rate, does not generate this value as the value of the levered asset when g
d
0
= g.
However, there is a simple adjustment to the WACC that generates the correct levered asset value.
When d
x
0
(t ) = 0, debt policy is an extension of the continuous-time Miles and Ezzell debt
policy that allows for cash ows with nonzero expected growth. In this case, Eq. (9) indicates that
there is a proportional relationship between the value of a levered asset and its otherwise identical
unlevered counterpart:
V
L
t
=
r
f
+p g
r
f
(1 t
c
d
v
) +p g
V
U
t
implying
V
L
t
= V
U
t
+
r
f
r
f
+p g
t
c
D
t
.
If g
d
0
= g, the debt tax shield can be written as a simple weighted average of the tax shields
for the MilesEzzell constant leverage ratio debt policy and the extended ModiglianiMiller debt
policy (with possibly growing debt). In this case, Eq. (9) reduces to
V
L
t
= V
U
t
+w
t
r
f
r
f
g
t
c
D
t
+(1 w
t
)
r
f
r
f
+p g
t
c
D
t
,
where the weight
w
t
=
r
f
g
r
f
(1 t
c
d
v
) g
_
1
d
v
D
t
/V
L
t
_
.
16
One can also map D
t
into V
L
t
. A small amount of algebraic manipulation reveals
V
L
t
= V
U
t
+
r
f
r
f
g
d
0
t
c
_
D
t
+
g p g
d
0
r
f
(1 t
c
d
v
) +p g
d
v
V
U
t
_
.
Thus, the value of the debt tax shield is the sum of the ModiglianiMiller debt tax shield (with constantly growing debt)
and a term which may be positive or negative. The sign of the nal term in parentheses depends on whether g p, the
risk-adjusted growth rate of the free cash ows, is larger than g
d
0
, the growth rate for the nonstochastic debt component.
This weight is monotonically decreasing in d
v
, the sensitivity of debt to the value of the asset,
holding D
t
xed.
17
The ModiglianiMiller debt tax shield, which multiplies w
t
above, has a smaller value than
its MilesEzzell counterpart, which multiplies (1 w
t
) above.
18
Hence, for the same initial
debt level, increasing the debt sensitivity coefcient, d
v
, while holding the initial debt level xed,
reduces the value of the tax shield. This is because the value of the debt tax shield falls when its
risk increases, other things equal. If the debt sensitivity coefcient, d
v
, exceeds D
t
/V
L
t
, so that
w
t
is negative, the leverage ratio will rise as the assets value increases and fall when it decreases.
In this case, the value of the asset will be below that obtained with the constant leverage ratio
MilesEzzell debt policy. Conversely, if d
v
is negative, so that some of the existing debt is retired
when the assets value rises,
19
and debt is issued when the assets value declines, the value of the
debt tax shield will be above the t
c
D
t
value proposed by Modigliani and Miller. This conrms
the intuition in Grinblatt and Titman [9] and suggests that the appropriate discount rate for free
cash ows will be a weighted average of the WACC formulas proposed by Miles and Ezzell and
Modigliani and Miller. For d
v
> D
t
/V
L
t
, the weight on the ModiglianiMiller WACC must be
negative, for d
v
< 0, it is above 1, and otherwise, it lies between 0 and 1. As we will show later,
the weighting on the WACC formulas of ModiglianiMiller and MilesEzzell is identical to the
weighting of the respective tax shields given here.
The linear debt policy in Case 3 easily extends to discrete time. The linearity implies that
over any discrete interval, the levered asset is a xed-weight portfolio of an otherwise identical
unlevered asset and a risk-free security. In this case, the values of the levered and unlevered
assets are perfectly correlated, as both are linear functions of the cash ow. Solving the difference
equations that generate the no arbitrage value of the levered asset in an analogous manner yields
the discrete time analogue to Eq. (9)
20
:
V
L
t
=V
U
t
+t
c
_
r
f
r
f
(1 t
c
d
v
) g
d
0
d
x
0
(t )
+
r
f
(1 +r
f
+p)
(1 +r
f
)(r
f
(1 t
c
d
v
) +p g) pr
f
t
c
d
v
d
v
V
U
t
_
.
This valuation solution nests both the discrete-time ModiglianiMiller and MilesEzzell debt
policies as special cases.
The discrete case valuation formula, provided above, applies only to an innitely-lived asset.
A similar approach generates a discrete time closed-form solution for a nite-lived asset. It is
omitted for the sake of brevity.
17
To see this, it is necessary to obtain an equation for the weight without V
L
t
. This is accomplished by substituting the
former equation into the latter and solving for w
t
.
18
This and the statements that follow from it assume that p, the risk premium for the free cash ow, is positive. If the
free cash ow has a negative risk premium, the MilesEzzell value exceeds the ModiglianiMiller value.
19
This debt paydown pattern has been estimated in empirical work by Kaplan and Stein [23].
20
Simple algebraic manipulation indicates that the mapping from D
t
to V
L
associated with the equation below is
given by
V
L
t
= V
U
t
+
r
f
r
f
g
d
0
t
c
_
D
t
+
(1 +r
f
)(g p g
d
0
) +p(r
f
g
d
0
)
(1 +r
f
)(r
f
(1 t
c
d
v
) +p g) pr
f
t
c
d
v
d
v
V
U
t
_
.
3. The weighted average cost of capital and tax-adjusted discount rates
For nance practitioners, discounting expected free cash ows at a tax-adjusted discount rate
is the most popular way to value an asset. This section studies the relation between this discount
rate and the after-tax weighted average cost of capital. It develops formulas for these discount
rates for a variety of debt policies and shows when and why naive application of the weighted
average cost of capital as the appropriate discount rate can generate erroneous valuations.
We take the perspective of an investor at date 0. This investor would like to know the discount
rate, applied to expected future free cash ows, that generates V
L
0
. Our analysis will show that
this discount rate is rarely the WACC computed at date 0. Before we do this, it is important to
study the WACC and how it evolves through time.
3.1. Risk, expected return, and the WACC
In continuous time, the weighted average cost of capital of an asset is dened to be the assets
instantaneous expected return, r
L
, less the return component due to the debt tax shield
21
:
WACC
t
= r
L
t
r
f
t
c
D
t
V
L
t
. (10)
A levered assets date t instantaneous expected return, equivalent to its pre-tax weighted average
cost of capital, is dened by
22
r
L
t
dt =
E
t
(dV
L
t
) +X
t
dt +r
f
t
c
D
t
dt
V
L
t
. (11)
This expected return has three components: the expected capital gain, the free cash ow, and
the cash ow from the debt tax shield.
23
One can readily show from the no arbitrage condition
that
r
L
t
= r
f
+
_
* ln V
L
t
*X
t
+o
H
(t, X
t
, D
t
, V
L
t
)
* ln V
L
t
*H
t
_
p(t, X
t
)X
t
.
Combining this equation with Eq. (10) provides a direct formula for computing a WACC given
the value of the levered asset:
WACC
t
= r
f
_
1 t
c
D
t
V
L
t
_
+
_
* ln V
L
t
*X
t
+o
H
(t, X
t
, D
t
, V
L
t
)
* ln V
L
t
*H
t
_
p(t, X
t
)X
t
. (12)
Eq. (12) is a generalization of the ModiglianiMiller adjusted cost of capital formula. It is
convenient for computing the WACC given the extensive closed-form solutions derived in the last
21
This is the same formula used by practitioners and found in textbooks, as the levered assets expected return is a
portfolio-weighted average of the pre-tax expected returns of the (assumed risk-free) debt and equity nancing the asset.
22
This expected return is the appropriate discount rate for the capital cash ow stream, which is the net payout to all
cash ow claimants. See Ruback [20] for a lucid discussion of the advantages of this approach.
23
As mentioned earlier, the latter ow must be paid out to maintain the same investment policy and capital gains
appreciation as an otherwise identical unlevered asset.
section. For example, in the case of constant debt for a nite-lived asset with zero expected growth
and a constant risk premium for free cash ows,
* ln V
L
*H
= 0
and
* ln V
L
t
*X
t
=
V
U
t
V
L
t
1
X
t
.
In this case, the levered asset has a value of
V
L
t
= V
U
t
+t
c
D
t
(1 e
r
f
(T t )
),
and the formula for the WACC reduces to
WACC
t
= r
U
_
1 t
c
D
t
V
L
t
_
+t
c
p
D
t
V
L
t
e
r
f
(T t )
.
Note that the second term is decreasing in T and converges to 0 for perpetual assets. Hence, this
formula generates a larger WACC than that generated by the Modigliani and Miller formula for
perpetual assets.
It is also possible to write down a differential equation for the WACC. Substituting the expected
return formula, Eq. (11), into Eq. (10) implies
WACC
t
dt =
E
t
_
dV
L
t
_
+X
t
dt
V
L
t
. (13)
Applying Itos lemma to this equation, we nd that the WACC satises the PDE
WACC V
L
=
_
*V
L
*t
+gX
*V
L
*X
+
1
2
o
2
X
2
*
2
V
L
*X
2
+j
h
*V
L
*H
+
1
2
o
2
h
*
2
V
L
*H
2
+oo
h
*
2
V
L
*H*X
_
V
L
+X.
To understand the relation between the WACC and stochastic discount rates for free cash ows,
observe that j
t
, the date t stochastic instantaneous discount rate for X
t
that generates the value
of the levered asset, satises the stochastic integral equation
V
L
t
=
_

t
E
t
_
e
_
s
t
j
c
dc
X
s
_
ds.
By the FeynmanKac theorem, any j
t
that satises this stochastic integral equation also satises
the PDE
jV
L
=
_
*V
L
*t
+gX
*V
L
*X
+
1
2
o
2
X
2
*
2
V
L
*X
2
+j
h
*V
L
*H
+
1
2
o
2
h
*
2
V
L
*H
2
+oo
h
*
2
V
L
*H*X
_
V
L
+X.
Since the PDE that j has to solve is identical to the PDE that the WACC solves (by Itos Lemma),
the following proposition must hold
24
:
Proposition 1. TheWACCis identical to the stochastic instantaneous discount rate that generates
the levered asset value when applied to the free cash ows.
This insight is not very useful for capital budgeting practitioners. As Eq. (12) indicates, the
WACC at a future date depends on the state variables, X and H, at that date. Hence, future WACCs
are generally stochastic when viewed from date 0, the relevant date of the valuation.
3.2. The appropriate tax-adjusted discount rates
Even though the WACC is generally stochastic, it may be that some construct related to the
WACC can be used to discount expected future free cash ows to date 0 in a manner that accounts
for the tax shield. This subsection explores this issue. We rst begin by analyzing a discount rate,
known at date 0, that translates the expected free cash owat date s into its value an instant earlier.
We call this the tax-adjusted forward rate. Once having developed an understanding of what
this tax-adjusted forward rate is, and how to compute it, we prove that the initial WACC is an
appropriate tax-adjusted discount rate whenever the termstructure of tax-adjusted forward rates is
at. These forward rates, while computable, are fairly impractical for capital budgeting purposes.
However, they are consistent with a single tax-adjusted discount rate for free cash owsthe
tax-adjusted hurdle rate for the IRRwhich generates the same present value. Moreover, when
the unlevered assets are perpetual and the Gordon-Growth assumptions apply, this hurdle rate,
denoted j
, is easily obtained with a simple formula.

Dene date 0s tax-adjusted forward discount rate for cash ows at date s, f
s
, by
f
s
ds =
E
0
_
dV
L
s
+X
s
ds
_
E
0
_
V
L
s
_ . (14)
This is clearly an appropriate discount rate. It is known at date 0, and the recursive relationship
expressed in Eq. (14), applied iteratively, implies
V
L
0
=
_

0
e
_
t
0
f
s
ds
E
0
[X
t
] dt. (15)
How does this series of forward rates relate to the WACC? At most horizons, the comparison
is meaningless because the future WACCs are stochastic when viewed from date 0. As a general
matter, the ratio of date 0 expectations in Eq. (14), used to compute the date s forward rate, differs
from the date 0 expectation of WACC
s
. Moreover, the expectations in Eq. (14), while obtainable,
do not lend themselves to simple expressions.
Despite this, developing an understanding of tax-adjusted forward rates is useful for under-
standing when the initial WACC can be used for discounting. Trivially, f
s
converges to WACC
0
as
s approaches 0. Because f
0
= WACC
0
, and f
s
is an appropriate instantaneous discount rate for
date s cash ows, it follows that whenever f
s
= f
0
for all s, WACC
0
is an appropriate discount
rate for free cash ows.
24
Obviously, there is no reason for the boundary conditions to differ or for either of the partial differential equations to
be ill-behaved.
Proposition 2. Whenever the term structure of tax-adjusted forward rates is at (f
s
= f
0
, s),
the WACC is an appropriate tax-adjusted discount rate for free cash ows.
Two cases where this situation arises are the no-growth ModiglianiMiller model and the Miles
Ezzell model (with or without cash owgrowth). The only other situation where the forward term
structure is at is when debt policy is any weighted average of debt policies in these two models,
but only for g = g
d
0
. We explore this shortly.
The rarity of an equivalence between the initial WACC or expected future WACCs and the
corresponding forward rates should not be surprising. Although Eqs. (13) and (14) look similar,
Jensens inequality alone prevents the expectation of the former from equalling the latter when
V
L
t
, as well as the overall ratio in Eq. (13), are stochastic.
Date 0s constant tax-adjusted discount rate, j
, is dened by the following equation

V
L
0
=
_

0
e
j
t
E
0
[X
t
] dt.
In cases where the free cash ows have perpetual constant growth g, this tax-adjusted discount
(hurdle) rate is most easily computed as
j
= g +
X
0
V
L
0
. (16)
Eq. (16), an algebraic manipulation of the growing perpetuity formula, allows us to obtain j
from the formulas for V

L
0
developed in the prior section.
25
3.3. Classes of cases with easy numerical solutions
Recall from the last section that the values of assets with time-independent cash ows and
debt policy, as well as assets with additively separable history diffusion and debt equations could
easily be obtained numerically. In the former class of cases, the fundamental valuation equation
reduces to an ordinary differential equation in X. In the latter class, it reduces to a systemof Riccati
equations. In these cases, the WACCand appropriate discount rate, j
, are similarly solved. Just as

numerical solutions for V
L
are easily obtained, so too are solutions for the WACC using Eq. (12).
For j
, the formula in Eq. (16) generates the discount rate directly from the numerically solved
V
L
.
26
Similarly, the forward rates are the solutions to ordinary differential equations which can
be solved numerically.
27
Obviously, it is more illuminating to analyze closed-form solutions for these interesting vari-
ables. We turn our attention to this next.
25
With a nite-lived asset, j
is easily identied implicitly as the parameter that solves

V
L
0
=
X
0
j
g
_
1 e
(j
g)(T t )
_
.
26
For nite-lived assets, this discount rate can be easily solved implicitly, as described in the prior footnote.
27
The ODEs are available upon request.
3.4. Additively separable assets with closed-form solutions
Consider, as in the last section, the case where history follows the diffusion
dH
t
= (m
x
(t )X
t
+m
d
D
t
+m
v
V
L
t
m
h
H
t
)dt +(l
x
(t )X
t
+l
d
D
t
+l
v
V
L
t
l
h
H
t
)dB
t
and debt policy has the functional form
D
t
=
z
d
x
z
(t )X
z
t
+d
v
V
L
t
+d
h
H
t
.
Recall that g, p, and o are constant, d
x
z
(t ) = d
x
z
(0) e
g
d
z
t
, m
x
z
(t ) pl
x
z
(t ) = k
x
z
(t ) = k
x
z
(0) e
g
k
z
t
,
and T = .
Here, the partial derivatives in Eq. (12) have closed-form solutions, allowing us to express
the WACC as an explicit function of the free cash ow, history state variable, and leverage ratio,
D
t
/V
t
, as follows:
WACC
t
= r
f
_
1 t
c
D
t
V
L
t
_
+
1/(r
f
+p g) +
z
c
x
z
(t )zX
z1
t
+c
h
(t )o
H
V
U
t
+
z
c
x
z
(t )X
z
t
+c
h
(t )H
t
pX
t
, (17)
where the c coefcients are explicitly given in the prior section of the paper and with
o
H
=
z
l
x
z
(t )X
t
+l
d
D
t
+l
v
V
L
t
(l
h
+o)H
t
.
Case 1: Perpetual assets. Here, Eq. (17) simplies to the same expression, but with
c
h
(t ) = c
h
and
c
x
z
(t ) =
k
x
z
(t )c
h
g
x
z
g
k
z
+
(r
f
t
c
+k
d
c
h
)d
x
z
(t )
g
x
z
g
d
z
+
(k
v
+k
d
d
v
)c
h
+r
f
t
c
d
v
(r
U
g)(r
f
(1 t
c
d
v
) +p g)
o
z,1
,
where the constants g
x
z
and c
h
are given in the previous section of the paper. In contrast, the

constant tax-adjusted discount rate is
j
= g +
X
0
z
c
x
z
(0)X
z
0
+c
h
H
0
.
Case 2: Perpetual debt as a function of cash ows only. Consider, as in the last section, the case
where
D
t
=
z
d
x
z
(t )X
z
t
,
g, p, and o are constant, d
x
z
(t ) = d
x
z
(0) e
g
d
z
t
and T = . For this case, we showed that
V
L
t
= V
U
t
+
z
r
f
t
c
r
f
+(p g)z
1
2
o
2
z(z 1) g
d
z
d
x
z
(t )X
z
t
.
This valuation solution simplies Eq. (17) to
WACC
t
= r
f
_
1 t
c
D
t
V
L
t
_
+
1/(r
f
+p g) +
z
c
x
z
(t )zX
z1
t
V
U
t
+
z
c
x
z
(t )X
z
t
pX
t
,
where
c
x
z
(t ) =
r
f
t
c
d
x
z
(t )
r
f
+(p g)z
1
2
o
2
z(z 1) g
d
z
,
while the constant tax-adjusted discount rate
j
= g +
X
0
V
U
0
+
z
r
f
t
c
r
f
+(pg)z
1
2
o
2
z(z1)g
d
z
d
x
z
(0)X
z
0
.
This happens to be a case where the presentation of the term structure of tax-adjusted forward
rates computed from the date 0 valuation date will not signicantly lengthen the paper. To do this,
we need to take date 0 expectations of V
L
s
. Given the lognormal process for X
s
, the formula for
the conditional zth moment of X
s
is
E
0
_
X
z
s
_
= X
z
0
e
z(g
1
2
o
2
)s+
1
2
z
2
o
2
s
,
which can be substituted into the expected value of Eq. (14). Thus,
E
0
[V
L
s
] = V
U
0
e
gs
+
z
t
c
r
f
d
x
z
X
z
0
e
z(g
1
2
o
2
)s+
1
2
z
2
o
2
s
r
f
+(p g)z
1
2
o
2
z(z 1)
.
It follows that
dE
0
[V
L
s
]
ds
= V
U
0
g e
gs
+
z
t
c
r
f
d
x
z
(z(g
1
2
o
2
) +
1
2
z
2
o
2
)X
z
0
e
z(g
1
2
o
2
)s+
1
2
z
2
o
2
s
r
f
+(p g)z
1
2
o
2
z(z 1)
.
We now have all the ingredients to compute the date s tax-adjusted forward rate. It is given by the
formula:
f
s
=
V
U
0
(r
f
+p) e
gs
+
z
t
c
r
f
d
x
z
(z(g
1
2
o
2
) +
1
2
z
2
o
2
)X
z
0
e
z(g
1
2
o
2
)s+
1
2
z
2
o
2
s
r
f
+(p g)z
1
2
o
2
z(z 1)
V
U
0
e
gs
+
z
t
c
r
f
d
x
z
X
z
0
e
z(g
1
2
o
2
)s+
1
2
z
2
o
2
s
r
f
+(p g)z
1
2
o
2
z(z 1)
.
It is easily veried that WACC
0
= f
0
, but obviously, the date s WACC depends on X
s
and thus
cannot be identical to f
s
. Moreover, the date s forward rate is not the date 0 expectation of the
date s WACC, irrespective of whether expectations are taken with respect to the actual probability
density function or the probability density function generated by the risk-neutral measure.
Case 3: Perpetual debt that is a linear function of asset value plus constant growth. Recall that
for this special case, explored in the last section, debt policy is given by
D
t
= d
x
0
(0) e
g
d
0
t
+d
v
V
L
t
and a perpetual levered asset has a particularly simple functional form for its valuation:
V
L
t
= V
U
t
+t
c
_
r
f
r
f
(1 t
c
d
v
) g
d
0
d
x
0
(t ) +
r
f
r
f
(1 t
c
d
v
) +p g
d
v
V
U
t
_
.
This valuation solution simplies Eq. (17) to
WACC
t
= r
f
_
1 t
c
D
t
V
L
t
_
+
pX
t
(r
f
(1 t
c
d
v
) +p g)V
L
t
,
with V
L
t
given by the equation immediately above, while the appropriate tax-adjusted constant
discount rate, j
, is given by
j
= g +
X
0
V
U
0
+t
c
_
r
f
r
f
(1 t
c
d
v
) g
d
0
d
x
0
(t ) +
r
f
r
f
(1 t
c
d
v
) +p g
d
v
V
U
0
_.
Using Eq. (14), the simple valuation formula for this case can be used to show that the tax-
adjusted forward discount rate is given by the weighted average
f
s
=

0
(s)
0
(s) +
1
(s)
g
d
0
+

1
(s)
0
(s) +
1
(s)
(r
f
(1 t
c
d
v
) +p),
where the coefcients
0
(s) =
r
f
t
c
d
x
0
(0)
r
f
(1 t
c
d
v
) g
d
0
e
g
d
0
s
,
1
(s) =
X
0
r
f
(1 t
c
d
v
) +p g
e
gs
and
V
L
0
=
0
(0) +
1
(0).
It is possible to compare forward rates, WACCs, and appropriate constant tax-adjusted discount
rates here but the discussion is more illuminating if we focus on several special cases of this
example.
Case 3a: Constantly growing debt with no stochastic component and no expected cash ow
growth. This extensionof the ModiglianiMiller debt policyaccounts for the possibilityof growing
debt, g = d
v
= 0, and r
f
> g
d
0
> 0. The
0
(s) and
1
(s) coefcients above simplify to
0
(s) =
r
f
t
c
D
0
r
f
g
d
0
e
g
d
0
s
=
0
e
g
d
0
s
,
1
(s) =
X
0
r
f
+p
= V
U
0
implying
f
s
=

0
e
g
d
0
s
g
d
0
V
U
0
+
0
e
g
d
0
s
+
V
U
0
V
U
0
+
0
e
g
d
0
s
r
U
.
The initial WACC, which is identical to f
0
, is
WACC
0
=

0
V
L
0
g
d
0
+
V
U
0
V
L
0
r
U
=r
U
_
1 t
c
D
0
V
L
0
_
p
r
f
g
d
0
g
d
0
t
c
D
0
V
L
0
.
However, the tax-adjusted constant discount rate is
j
=
X
0
V
L
0
= r
U
V
U
0
V
L
0
= r
U
_
1
r
f
r
f
g
d
0
t
c
D
0
V
L
0
_
,
which the initial WACC exceeds by the amount
0
V
L
0
g
d
0
=
r
f
r
f
g
d
0
g
d
0
t
c
D
0
V
L
0
.
Case 3b: ModiglianiMiller static debt policy with constant expected growth in cash ows.
Here we assume that g > 0, but that d
v
= g
d
0
= 0. In this case
0
(s) = t
c
D
0
,
1
(s) =
X
0
r
f
+p g
e
gs
= V
U
0
e
gs
implying
f
s
=
V
U
0
V
L
0
e
gs
r
U
.
The initial WACC, which is identical to f
0
, is
WACC
0
=
V
U
0
V
L
0
r
U
= r
U
_
1 t
c
D
0
V
L
0
_
.
However, the tax-adjusted constant discount rate is
j
=
X
0
V
L
0
+g =
V
U
0
V
L
0
(r
U
g) +g
=
V
U
0
V
L
0
r
U
+gt
c
D
0
V
L
0
,
which exceeds the initial WACC by gt
c
D
0
V
L
0
. Like the other cases above, using the WACC as a
discount rate here can lead to serious valuation errors. For example, a company with a constant
amount of debt, a current free cash ow of $100 million, a perpetual cash ow growth rate of 5%,
a 30% corporate tax rate, and a 50% debt to asset ratio can plausibly have a true present value of
$2 billion. This present value would stem from a properly computed tax-adjusted discount rate
of 10%. The WACC in this case, 9.25%, leads to a valuation of more than $2.35 billion. What is
really striking here is the degree to which the tax shield is misvalued. The true value of the entire
tax shield is $300 million which is smaller than the valuation error arising from using the WACC
as the discount rate for free cash ows.
Case 3c: The ModiglianiMiller model with equal growth in debt and expected cash ows. The
last two cases illustrated that the WACC was not an appropriate tax-adjusted constant discount
rate. This is because it changed over time. In Case 3a, the initial WACC was too large because it
was expected to decrease. In Case 3b, the initial WACC was too small because it was expected to
increase. It is natural to think that if d
v
= 0 but g
d
0
= g > 0, so that debt growth keeps pace with
the expected growth of the cash ows, the initial WACC will still work. This indeed is the case.
To prove this, note that here
WACC
0
= f
0
=

0
V
L
0
g
d
0
+
V
U
0
V
L
0
r
U
= r
U
_
1 t
c
D
0
V
L
0
_
p
r
f
g
d
0
g
d
0
t
c
D
0
V
L
0
.
Thus, WACC
0
= j
whenever
0
V
L
0
g
d
0
+
V
U
0
V
L
0
r
U
= g +
(r
U
g)V
U
0
V
L
0
,
which only occurs when g
d
0
= g.
This happens to be a case in which the forward rate is independent of s. By Proposition 2, the
initial WACC works as an appropriate tax-adjusted discount rate under this condition.
Case 3d: Acombination of the extended ModiglianiMiller and MilesEzzell debt models with
equal growth in debt and expected cash ow. Here, we generalize Case 3c and once again show
that the initial WACC is an appropriate discount rate for free cash ows. In this example, g
d
0
= g
but d
v
may be nonzero. These assumptions imply:
0
(s)
0
(s) +
1
(s)
= 1
X
0
V
L
0
1
r
f
(1 t
c
d
v
) +p g
,
1
(s)
0
(s) +
1
(s)
=
X
0
V
L
0
1
r
f
(1 t
c
d
v
) +p g
.
To compute f
s
, the former ratio multiplies g, while the latter multiplies r
f
(1 t
c
d
v
) +p, and
then the two products are summed. Thus,
f
s
= f = j
= WACC
0
= g +
X
0
V
L
0
.
When g
d
0
= g, f
s
depends on s. Hence, within Case 3, it is only by imposing the condition g
d
0
= g
that one can use the initial WACCas an appropriate discount rate. This discount rate can be thought
of as a weighted average of the WACC from a Miles and Ezzell debt policy at an initial debt of
D
0
and the WACC from a Modigliani and Miller debt policy at an initial debt of D
0
. Specically,
denote the WACCs of ModiglianiMiller and MilesEzzell as
WACC
MM
0
= r
U
_
1 t
c
D
0
V
L
0
_
gp
r
f
g
t
c
D
0
V
L
0
,
WACC
ME
0
= r
U
t
c
r
f
D
0
V
L
0
,
respectively. Then the discount rate given above
WACC
0
= j
= f = w
0
WACC
MM
0
+(1 w
0
)WACC
ME
0
,
where
w
0
=
r
f
g
r
f
(1 t
c
d
v
) g
_
1
d
v
D
0
/V
L
0
_
.
Case 3e: The growth extension of the MilesEzzell model. Consider the case with d
v
= D
0
/V
0
,
d
x
0
(0) = 0. This is a continuous-time version of the MilesEzzell model, but with the extension
of expected growth in cash ows. Here,
V
U
t
=
X
t
r
f
+p g
and
V
L
t
= V
U
t
+
r
f
r
f
+p g
t
c
D
t
.
In this case, since
0
(s) is 0, the date s forward rate is independent of s, which has the value
f = j
= r
U
r
f
t
c
D
0
V
L
0
.
This tax-adjusted discount rate is identical to the weighted average cost of capital.
To understand this from another perspective, recognize that everything is stationary here, and
that the ow from the debt tax shield, being proportional to the free cash ow, shares its discount
rate, so that
r
L
= r
f
+p.
The WACC, from Eq. (10), is therefore the constant
WACC
t
= r
f
+p r
f
t
c
D
V
L
,
where by assumption, D/V
L
is the same at all points in time. Note that the growth rate of the
cash ows never appears when the WACC is written as a function of the leverage ratio, but that it
affects V
L
above.
3.5. Discussion
To develop an intuitive understanding of the results above, it is useful to rst review why
discounting free cash ows at the initial after-tax weighted average cost of capital sometimes
accounts for the value of the debt tax shield. Let us begin with the perpetual level debt model of
Modigliani and Miller. Because the market value of the debt nancing never changes in this model,
it is useful to think of the Modigliani and Miller analysis as the valuation of the debt tax shield
of a zero beta debt strategy. In such a model, the familiar equation V
L
= V
U
+t
c
D reects the
separate valuations of the two components of the asset: the free cash ow stream, with stochastic
value X
t
at date t, and the stream of cash ows from the interest-based tax shield, Dr
f
t
c
, which
is constant and identical at every date. Note that the gross (as opposed to net) pre-tax weighted
average cost of capital, which is also the gross expected return on the levered asset, is
1 +r
L
= 1 +(D/V
L
)r
f
+(E/V
L
)r
E
.
This can be viewed as the expected cash ow per dollar of assets to investors who buy an asset,
hold it for an instant, and then liquidate it. Both the asset value and the ow include the debt tax
shield. Hence, the gross after-tax WACC
1 +(D/V
L
)r
f
(1 t
c
) +(E/V
L
)r
E
is just the expected ow to investors per dollar of levered assets, less the expected ow per dollar
of assets from the debt tax shield (D/V
L
)r
f
t
c
. This net ow is identical to the free cash ow
per dollar of levered assets. Viewed with the arrow of time in reverse, this insight implies that
if we discount the free cash ow at the after-tax WACC, we end up back where we started, with
the one dollar value of the assets, including that component of value generated by the debt tax
shield.
It is easy to see that this insight about the zero beta debt strategy does not easily extend to the
case of an asset with a value that tends to grow. Here, the value of the unlevered asset is affected
by the growth rate, but the debt tax shield, having a perpetual value of t
c
D, is not affected by it.
Because the value of such a constant debt tax shield tends to decline over time as a proportion of
total asset value, the initial after-tax weighted average cost of capital is an inappropriate discount
rate for all future cash ows. Essentially, the weighted average cost of capital is changing over
time. In this simple case, we learned that there is a single discount rate that can be used to discount
all future free cash ows: the sum of the initial after-tax WACC and gt
c
D/V
L
, the latter being
the product of the growth rate of the free cash ows, the tax rate, and the debt to value ratio. For
a rm with a 4% growth rate, and a 50% tax rate and leverage, this alone represents a 100 basis
point error in the discount rate, even given the true unlevered cost of capital. We can see that
if the risk of the levered asset is not expected to change, as in the ModiglianiMiller model of
debt adjustment, which assumes no growth, discounting the expected free cash ow stream at the
initial WACC generates the value of the levered assets, including the debt tax shield.
The other case where the risk of the levered asset is not expected to change arises in an extension
of the Miles and Ezzell [16] model. Their case is one where the debt level is adjusted over time to
maintain a constant debt to value ratio. This is a positive beta debt strategy even though the debt,
at issue, is risk-free. The perpetually constant WACC is an appropriate discount rate in this case,
but the valuation is never out of peril if the actual debt policy associated with the asset differs from
the policy implicit in the formulas used for the valuation. For example, if the unlevered assets
risk premium is twice the risk-free rate, the MilesEzzell tax shield is a mere one-third that of
the ModiglianiMiller debt tax shield for the same debt level. If the risk-free rate is 4%, the bias
in the WACC from applying a MilesEzzell formula to a ModiglianiMiller debt policy becomes
200 basis points, even if the unlevered cost of capital for the asset is estimated perfectly.
These problems can become much more severe when the unlevered cost of capital is from
comparison assets with debt policies that are also mismatched to the formulas. For example,
an asset with a nonstochastic trend to grow its debt but pay it down if the cash ows turn out
to be surprisingly good, has a very low WACC other things equal. Unless this is recognized,
unlevering such an asset with a standard formula tends to produce too low an unlevered cost of
capital. Similarly, obtaining the WACC of such an asset at a target debt level using the standard
formulas tends to produce too large a WACC. In contrast, consider an asset for which debt has a
nonstochastic retirement trend, but for which debt tends to increase rapidly when cash ows are
surprisingly good. Applying the standard formulas to such an asset tends to generate unlevered
costs of capital that are too high given a known WACC and a WACC at a target debt level that is
too low given the assets unlevered cost of capital. If the former asset has a known WACC and is
used to generate the WACC for the latter asset (or vice versa), the bias can easily run to 500 basis
points or more.
28
4. Levering and unlevering equity betas
The relationship between equity betas and unlevered asset betas is critical for valuation. Equity
betas, unlike asset betas, are observed. Hence, to obtain the necessary discount rates, valuations
analyze the traded equity of assets that are deemed similar to the asset being valued. The problemis
that the risk of the traded equity of comparison assets is affected by the leverage policy associated
with the comparison asset. It is generally necessary to undo the leverage-induced distortion on the
equity beta of the comparison asset(s) in order to obtain the critical valuation inputs: the unlevered
asset beta or unlevered cost of capital. This section explores how to do this.
The return on an unlevered asset is given by
dV
U
+X
t
dt
V
U
=
_
r
f
+p
* ln V
U
* ln X
_
dt +
* ln V
U
* ln X
o dB,
while the return on a levered asset is given by
dV
L
+Xdt +r
f
t
c
D dt
V
L
=
_
r
f
+
_
* ln V
L
*X
+o
H
* ln V
L
*H
_
pX
_
dt
+
_
* ln V
L
*X
+o
H
* ln V
L
*H
_
oXdB.
The ratio of the terms that multiply dB in the two equations above represent the effect of leverage
on volatility. With risk-free debt, the proportional effect on beta is the same. Moreover, with risk-
free debt, the equity beta of a levered asset always is 1+D/E times the beta of the underlying asset
(including the asset component from the tax shield).
29
It follows that the formula for levering
and unlevering equity betas involves multiplying 1 +D/E by the ratio of the terms that multiply
dB above. That is,
[
E
t
=
_
1 +
D
t
E
L
t
_
* ln V
L
*X
+o
H
* ln V
L
*H
* ln V
U
*X
[
U
t
,
with the derivatives taken at date t.
The above equation suggests that whenever there is a closed-form solution for the levered asset
value, there is a closed-form solution for the equity beta formula, obtainable after rst taking
partial derivatives. For example, the extension of the ModiglianiMiller debt policy with g = 0
and g
d
0
= 0 has
* ln V
L
*H
= 0
and
* ln V
L
*X
=
V
U
V
L
* ln V
U
*X
,
28
Moreover, this analysis assumes that the WACC is an appropriate discount rate, which it rarely is.
29
This is just an inversion of the portfolio formula that generates the beta of the levered asset as a portfolio-weighted
average of the beta of its debt (0) and equity ([
E
).
which implies
[
E
t
=
_
1 +
_
1 t
c
r
f
r
f
g
d
0
_
D
t
E
L
t
_
[
U
t
.
When g
d
0
is nonzero, this formula differs from the leveragingunleveraging formula proposed by
Hamada [10] for the Modigliani and Miller debt policy. It also differs from this formula when
g = g
d
0
= 0 but the asset has a nite life. In the latter case,
* ln V
L
*H
= 0
and
* ln V
L
*X
=
V
U
V
L
* ln V
U
*X
.
However, because the nite horizon debt tax shield has a different value than the perpetual debt
tax shield, the formula reduces to
[
E
t
=
_
1 +
_
1 t
c
(1 e
r
f
(T t )
)
_
D
t
E
L
t
_
[
U
t
.
In contrast, the extension of the MilesEzzell policy with g = 0 has
* ln V
L
*H
= 0
and
* ln V
L
*X
=
* ln V
U
*X
= 1/X
implying
[
E
t
=
_
1 +
D
t
E
L
t
_
[
U
t
.
This is identical to the formula proposed by Miles and Ezzell and it applies to both an innitely-
lived and nite-lived asset (in contrast to the Hamada formula for the ModiglianiMiller debt
policy).
For a perpetual asset that is a hybrid of the ModiglianiMiller and MilesEzzell debt policies,
as given in Case 3 of the prior sections of the paper, the equity beta formula is
[
E
t
=
_
1 +
_
1 +t
c
_
d
v
D
t
/E
t
1
_
r
f
r
f
(1 t
c
d
v
) g
d
0
_
D
t
E
t
_
[
U
t
.
The formula above is a weighted average of the beta leveraging formulas of Hamada/Modigliani
Miller and MilesEzzell. That is
[
E
t
=
_
w
t
_
1 +
_
1 t
c
r
f
r
f
g
d
0
_
D
t
E
L
t
_
+(1 w
t
)
_
1 +
D
t
E
t
_
_
[
U
t
,
with the weight on the Hamada/ModiglianiMiller formula
w
t
=
r
f
g
d
0
r
f
(1 t
c
d
v
) g
d
0
_
1
d
v
D
t
/V
L
t
_
.
There are several additional cases with simple closed-form solutions for the equity beta. These
parallel cases 1 and 2 in the prior two sections of the paper. Since these involve mere substitutions
and elementary calculus, we omit them for the sake of brevity.
Finally, we note that since equity risk premia are proportional to equity betas, the formulas for
leveraging and unleveraging equity risk premia are the same as those above, with the levered and
unlevered equity risk premia substituting for their respective betas above.
5. Conclusion
This paper has undertaken a comprehensive valuation of debt tax shields. In as many cases
as possible, we offered closed-form solutions for the values of levered assets and the associated
debt tax shields. Our approach for obtaining these present values was the APV approach. The
tax-adjusted discount rates that generate the present values were reverse engineered, in that we
needed to use the present values to generate closed-form solutions for the discount rates.
The examples explored in this paper are particularly useful in that many dynamic debt policies
can be thought of as tting into one of these examples. The examples demonstrated that it is
possible to develop intuition from polar cases so that a manager can heuristically assess how his
discount rate and debt tax shield value will change, given the dynamic nature of the policy. For
instance, rms with debt policies that sluggishly but positively react to changes in the value of the
rms unlevered assets might be expected to have a debt tax shield with a value that lies somewhere
between the ModiglianiMiller and MilesEzzell values. The appropriate tax-adjusted constant
discount rate for the future cash ows of the unlevered assets, as well as the formulas for levering
and unlevering equity betas will also lie between the values given by the polar cases. Moreover,
when rms engage in what we term negative beta debt policies, paying down debt as the cash
ow prospects brighten, and vice versa, the formulas for the debt tax shield, discount rate, and
levered equity betas are again weighted averages of the two polar cases, with a negative weight
on the formula associated with the MilesEzzell positive-beta debt strategy and a weight above 1
on the ModiglianiMiller zero-beta debt strategy formula.
The comprehensive treatment of debt tax shields presented here is essential for practitioners.
Confusion has proliferated because the formulas that previously had been developed for the
simplest of cases are generally treated as black boxes without a clear understanding of where they
come from. It is rare when pedagogy appropriately links the formulas for levering and unlevering
betas to the value of the tax shield. The MilesEzzell valuation can easily be 1/3 the valuation
using the ModiglianiMiller approach. It is quite common, however, to observe both students and
nance professors mix the Miles and Ezzell formula for leveraging and unleveraging equity with
ModiglianiMiller inputs, thinking that they are getting a valuation of t
c
D for the debt tax shield.
In properly linking the values of tax shields for different debt policies to discount rates and laying
out the theory behind this linkage, we hope to remedy some of this confusion.
This paper is important, however, not just for those doing corporate valuations, but for those
doing research on capital structure and bankruptcy costs. In particular, it provides a set of bench-
marks for the impact of debt on asset values in a market that is frictionless, except for taxes.
Empirical research by Graham [6] and by Kemsley and Nissim [14] indicates that for the average
US rm, debt tax shields in the US are about the size of t
c
D. Clearly, more research that describes
the cross-sectional variation in this estimate is coming, and is warranted. To properly estimate
bankruptcy costs from extensions of this research, it is critical to know the value of the debt
tax shield in a market where the only friction is taxes. We believe that research in this area has
been hindered by a lack of understanding of benchmark valuations. By plugging this hole in the
literature, we hope to stimulate additional research.
Acknowledgments
The authors thank the UCLAAcademic Senate and the Harold Price Center for Entrepreneurial
Studies for nancial support and seminar participants at the University of Arizona, the University
of Kentucky, and the Financial Management Association meeting for comments on earlier drafts.
References
[1] M. Baker, J. Wurgler, Market timing and capital structure, J. Finance 57 (2002) 132.
[2] R. Brealey, S. Myers, F. Allen, Principles of Corporate Finance, eighth ed., McGraw-Hill/Irwin, Tokyo/Homewood,
IL, 2006.
[3] J. Cox, S. Ross, The valuation of options for alternative stochastic processes, J. Finan. Econ. 3 (1976) 145166.
[4] G. Donaldson, Corporate Debt Policy and the Determination of Corporate Debt Capacity, Harvard Graduate School
of Business Administration, 1961.
[5] D. Dufe, K. Singleton, Modeling term structures of defaultable bonds, Rev. Finan. Stud. 12 (1999) 687720.
[6] J. Graham, How big are the tax benets of debt?, J. Finance 55 (2000) 19011942.
[7] J. Graham, C. Harvey, The theory and practice of corporate nance: evidence from the eld, J. Finan. Econ. 60
(2000) 187243.
[8] R. Green, B. Hollield, The personal-tax advantages of equity, J. Finan. Econ. 67 (2003) 175216.
[9] M. Grinblatt, S. Titman, Financial Markets & Corporate Strategy, McGraw-Hill/Irwin, Tokyo/Homewood, IL, 2002.
[10] R. Hamada, The effect of the rms capital structure on the systematic risk of common stocks, J. Finance 27 (1972)
435452.
[11] J. Harrison, D. Kreps, Martingales and arbitrage in multiperiod securities markets, J. Econ. Theory 20 (1979)
381408.
[13] A. Kayhan, S. Titman, Firms histories and their capital structures, J. Finan. Econ. 83 (2007) 132.
[14] D. Kemsley, D. Nissim, Valuation of the debt-tax shield, J. Finance, 57 (2002) 20452074.
[15] H. Leland, Corporate debt value, bond covenants, and optimal capital structure, J. Finance 49 (1994) 12131252.
[16] J. Miles, R. Ezzell, Reformulating tax shield valuation, J. Finance 40 (1985) 14851492.
[17] F. Modigliani, M. Miller, The cost of capital, corporation nance and the theory of investment, Amer. Econ. Rev. 48
(1958) 261297.
[18] S. Ross, A simple approach to the valuation of risky streams, J. Bus. 51 (1978) 453475.
[19] S. Ross, Capital structure and the cost of capital, J. Appl. Finan. Spring/Summer (2005) 725.
[20] R. Ruback, Capital cash ows: a simple approach to valuing risky cash ows, Finan. Manage. 31 (2002) 85103.
[21] F. Black, M. Scholes, The pricing of options and corporate liabilities, J. Political Economy 81 (1973) 637654.
[22] R. Merton, Theory of rational option pricing, Bell Journal of Economics and Management Science 4 (1973)
141183.
[23] S. Kaplan, J. Stein, How risk is the debt of highly leveraged transactions? J. Financial Economics 27 (1990)
215246.

Grinblatt & Liu 2008 - Debt Policy, Corporate Taxes, and Discount Rates (JET)

Caricato da

Informazioni sul documento

Descrizione originale:

Titolo originale

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

Grinblatt & Liu 2008 - Debt Policy, Corporate Taxes, and Discount Rates (JET)

Caricato da

Copyright:

Formati disponibili

Journal of Economic Theory 141 (2008) 225254

= 0. In this case, the solution for Case 1 simplies to

, is easily obtained with a simple formula.

, is dened by the following equation

from the formulas for V

, are similarly solved. Just as

is easily identied implicitly as the parameter that solves

are given in the previous section of the paper. In contrast, the

Potrebbero piacerti anche