You are on page 1of 65

Equity Valuation: More than Thirty Years

Later
Peter O. Christensen
School of Economics and Management
University of Aarhus, Denmark
Gerald A. Feltham
Sauder School of Business
University of British Columbia, Canada
Preliminary and Incomplete
First version: December 5, 2006
This version: September 11, 2008
Contents
1 Introduction 1
2 Risk-adjusted Discount Rates 7
2.1 Enterprise Valuation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.2 Estimating the Cost of Capital . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2.1 The WACC-method . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2.2 The APV-method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2.3 Estimating the equity cost of capital . . . . . . . . . . . . . . . . . . 17
2.3 Value Additivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3 Multi-period Asset Pricing Theory and Accounting Relations 21
3.1 Fundamental Theorem of Asset pricing . . . . . . . . . . . . . . . . . . . . . 22
3.2 Accounting Relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.3 Alternative Formulations of Accounting Value Relations . . . . . . . . . . . . 28
3.4 Equilibrium Valuation Indices . . . . . . . . . . . . . . . . . . . . . . . . . . 30
4 An Accounting-based Multi-period Equity Valuation Model 33
5 An Accounting-based Multi-period Equity Valuation Model with Expo-
nential Utility 36
6 A VAR Model with Exponential Utility 39
7 Comparison to the Standard Textbook Approach 41
References 47
Appendix A: Proof of Theorems 49
Appendix B: Power Utility and Log-normal Distributions 55
Appendix C: Habit Formation 57
Appendix D: Risk-adjusted Expected Cash Flows and Certainty Equivalents 59
1 Introduction
The valuation of uncertain income streams is at the heart of …nancial and accounting research
as well as in almost all areas of business. Financial statement analysis and equity valuation
challenge market prices of traded stocks, and the valuation of a …rm’s equity is the key in any
equity deal be it an acquisition, a merger, or a private equity deal. The signi…cant increase in
the capitalization of private equity funds in recent years also calls for good models of valuing
equity when there is no market price. Basic capital budgeting is yet another prominent
example in which the valuation of uncertain investment returns is at the heart of the issue.
Fundamentally, the question in all of these settings is how to calculate the value today of a
stream of uncertain future cash ‡ows or earnings. The key issues involved are how to account
for taxes, in‡ation, payment patterns, and not least for risk.
In this paper, we …rst provide a short review of standard valuation models as they are
presented in textbooks and used in practice. These models are based on estimates of expected
future cash ‡ows or earnings and a risk-adjusted cost of capital estimated from time-series of
stock returns using some form of capital asset pricing model. We argue that this approach
is based on single-period results which do not extend to multi-period settings. For example,
the concept of the weighted average cost of capital is based on the well-known result from
single-period portfolio theory that the expected return of a portfolio of two assets (such as
equity and debt) is equal to the weighted average of the expected returns on the individual
assets. On the other hand, in a multi-period setting, the cost of capital is like an internal
rate of return on a sequence of expected cash ‡ows. However, the internal rate of return on
a portfolio of two assets is not equal to the weighted average of the internal rates of returns
on the individual assets except under very restrictive assumptions such as the same constant
growth in perpetuity of the expected cash ‡ows on the individual assets. If the latter is
indeed descriptive, then a simple valuation model like the Constant Growth model may be
used, i.e., we may use a simple multiples-based model. Moreover, if the cost of capital is
estimated based on short-term stock returns, then leverage, interest rates and risk premia
must all be constants. Recall that interest rates change continuously, and that leverage and,
thus, equity risk premia change whenever the stock price changes.
We develop a comprehensive model based on contemporaneous multi-period asset pricing
theory. The purpose is twofold. Firstly, we take the initial steps in developing better valua-
tion models which can explicitly recognize key economic phenomena like stochastic interest
rates, di¤erences between short- and long-term interest rates, time-varying risk premia, and
the fact that the time-series properties of cash ‡ows and earnings a¤ect the proper risk-
adjustments. Equities are long-term claims. Why is it that the information in a readily
1
observable term structure of interest rates is not used in practical equity valuation? In his
seminal paper, “The Valuation of Uncertain Income Streams and the Pricing of Options,”
Mark Rubinstein was the …rst to explicitly show more than thirty years ago how closely eq-
uity prices and the prices of …xed income claims are related in a multi-period setting. In this
paper, we reintroduce this link for the purpose of practical equity valuation. The basic idea
is instead of discounting expected cash ‡ows or earnings with risk-adjusted discount rates,
we discount risk-adjusted expected cash ‡ows or earnings using zero-coupon interest rates
from a readily observable term structure of interest rates (which re‡ects expected future spot
interest rates and their risk). A key contribution of this paper is to explicitly identify the
risk-adjustments in a general multi-period equilibrium model. However, new empirical work
is needed in order to …nd the best empirical speci…cation of the model.
Secondly, we use the comprehensive multi-period valuation model for a set stylized ex-
amples to shed light on the types and the magnitudes of errors which may be made using
standard valuation models based on risk-adjusted discount rates. As it will become clear,
these errors are far from being trivial. In particular, we show that growth, persistence and
mean reversion of cash ‡ows or earnings are key determinants of not only expected future
cash ‡ows or earnings but also of proper risk-adjustments, whereas these factors do not di-
rectly a¤ect the risk-adjusted discounts rates in standard valuation models. Of course, these
factors may indirectly a¤ect the estimates of the cost of capital from time-series of stock
returns. We also show that in certain cases, a risk-adjusted discount rate can simply not be
used to …nd the current value of an uncertain future cash ‡ow (or accounting number). This
approach requires that the expected cash ‡ow and its current value have the same sign and,
obviously, this need not be the case. In order to illustrate this fact, a simple example is the
valuation of a forward contract. The forward price is determined such that the current value
of the contract is equal to zero, and the payo¤ is the di¤erence between future spot price and
the forward price. Unless the expected spot price is equal to the forward price (which would
imply risk-neutral pricing), the only risk-adjusted discount rate, which would give a current
price of zero, would be plus or minus in…nity. Instead, risk-adjusting the expected payo¤ on
the contract yields that the forward price is equal to the risk-adjusted expected spot price.
More generally, this is indeed the unifying principle used in the pricing of derivatives.
We frame our valuation question in terms of valuing a …rm’s common equity. However, it
should be noted that our approach is applicable for the general question of how to …nd the
value today of a stream of future cash ‡ows or earnings in other settings, such as in basic
capital budgeting. We teach our students in their …rst …nance class that the value today of
an investment is the net present value of the incremental future cash ‡ows it generates. We
also teach them that they should be very careful distinguishing cash receipts from revenues
2
and expenditures from costs—the di¤erence being the non-cash accruals. The key, of course,
is the impact of the time value of money. Nevertheless, in recent years there has been an
increased interest in accounting-based valuation models which instead of discounting future
cash ‡ows determine the net present value by discounting future residual income or abnormal
earnings growth anchored by current book value or capitalized one-period ahead earnings,
respectively. These accounting-based models provide exactly the same net present values
as the discounted cash ‡ows models. The reason for this equivalence is the articulation of
…nancial statements within and across periods. For example, a conservative assessment of
the book value of an asset today will eventually show up as higher residual income numbers
in future periods—the “error” in current book value is perfectly balanced against the increase
in the net present value of future residual income numbers.
The cash ‡ow is a very easy concept to understand for …rst year students, whereas
understanding the implications of …nancial statement articulation requires more training.
An advantage of accounting-based models over the discounted cash ‡ow models is that
value creation is re‡ected earlier in the accounting-based models. For example, in high
growth …rms, the free cash ‡ows are typically negative in the short run due to high capital
investments in new property, plant and equipment. Accrual accounting recognizes these
investments as assets on the balance sheet and, thus, the …rm may still have positive earnings
and even positive residual income (which measures earnings in excess of the capital cost of
using the book value of invested capital). In the …rst place, …rms have for several centuries
used accrual accounting instead of cash ‡ow accounting (which would fully expense capital
investment expenditures like a reduction in free cash ‡ows), to provide information about
value creation and performance of its employees—in a timely manner and for each period
separately.
It is important to recognize that value creation (or performance) as re‡ected in …nan-
cial statements does not perfectly re‡ect value creation in market value terms due to the
particular accounting rules applied. For example, research and development expenditures
are typically fully expensed, and …rms do not recognize the net present value of a capital
investment as an asset on the balance sheet at the investment date. Instead, these bene…ts
will be recognized as increases in future earnings as they are realized. The existence of these
accounting rules can only be understood by considering the stewardship role of accounting
and the incentives of managers and …rms to misrepresent information. The determination
of the particular accounting rules must strike a balance between timeliness, accuracy and
reliability. The beauty of accounting-based valuation models is that they recognize value
creation in the short run although imperfectly, and that any “errors” made will be perfectly
o¤set in the net present value of subsequent residual income numbers.
3
As a practical matter, …nancial analysts and others evaluating equities do typically not
forecast free cash ‡ows directly even if they are using the discounted free cash ‡ow model.
A likely reason is that past free cash ‡ow numbers contain a lot of transitory noise due,
for example, to the particular timing of past capital investment expenditures. Instead, a
common approach is to forecast value drivers like sales, pro…t margins, and asset turnover.
From these forecasts, future operating earnings and book values of operating assets can be
calculated. These are the key inputs to the accounting-based valuation models. Hence, these
models can be applied directly by calculating future residual operating income or abnormal
operating income growth from forecasts of operating income, book values of operating as-
sets, and the capital charge on opening book values or last period’s free cash ‡ows. If the
discounted free cash ‡ow models are used, the forecasts of free cash ‡ows are obtained from
the operating asset relation, i.e., ending book value of operating assets equals opening book
value of operating assets plus operating income minus the free cash ‡ow from operations
(which is the amount that can be used to pay the debt and the shareholders). Obviously,
these forecasts of free operating cash ‡ows are no more reliable than the forecasts of the ac-
counting numbers on which they are based. Hence, the choice between discounted free cash
‡ow models versus accounting-based models is merely a choice between how the results and
the analysis are presented, i.e., given the same set of forecasts the models are mathematically
equivalent (if the growth assumptions and the discounting are not messed up).
Although important issues, the focus of this paper is not to discuss the most e¢cient ways
of forecasting future accounting numbers or free cash ‡ows or to advocate accounting-based
valuation models over discounted free cash ‡ow models. Instead, our focus is on discounting
cash ‡ows or accounting numbers in a theoretically sound manner in a multi-period setting
under uncertainty. As mentioned above, our approach is signi…cantly in‡uenced by Rubin-
stein (1976). In that sense, the paper is on the foundations of valuation and discounting
under uncertainty and, hence, should be of relevance to both an accounting and a …nance
audience.
Our starting point is the implications of no-arbitrage in perfect multi-period …nancial
markets. A fundamental property of no-arbitrage pricing is that the valuation operator is a
linear functional, or just the value additivity principle. That is, the value of a portfolio of two
uncertain income streams must be equal to the sum of the values of the individual income
streams. An immediate consequence of the value additivity principle is that the value of a
sequence of future uncertain cash ‡ows must be equal to the sum of the values of each of
the future uncertain cash ‡ows. Similarly, the value of a given future uncertain cash ‡ow
must be equal to the value of the expected cash ‡ow plus the value of di¤erence between the
uncertain cash ‡ow and the expected cash ‡ow. If calculated at the future date, the latter
4
is referred to as (minus) the cash ‡ow risk premium. The di¤erence between the expected
cash ‡ow and the cash ‡ow risk premium is the risk-adjusted expected cash ‡ow. Since the
risk-adjusted expected cash ‡ow is a certain number, its current value must be equal to its
discounted value using the current zero-coupon interest rate for the particular future date as
the discount rate. Hence, no-arbitrage implies that the current value of a sequence of future
uncertain cash ‡ows is equal to the sum of the sequence of future risk-adjusted expected
cash ‡ows discounted by the associated zero-coupon interest rates.
In standard …nance textbooks, such as Grinblatt and Titman (2002), this method is re-
ferred to as the certainty equivalent method—with the certainty equivalent being equal to
the risk-adjusted expected cash ‡ow. However, one should be careful about the language
when it comes to actually determining the certainty equivalents. In the economics of uncer-
tainty, the certainty equivalent of uncertain wealth is de…ned as the level of wealth which
has the same utility as the expected utility of the uncertain wealth, i.e., the wealth level
which makes the decision maker indi¤erent between receiving this level of wealth with cer-
tainty and retaining the uncertain wealth. This concept is not equal to the risk-adjusted
expected cash ‡ow in equilibrium valuation models (as it might appear from reading at least
some …nance textbooks). One way to see this is to note that certainty equivalents do not
satisfy the value additivity principle and, thus, violate the implications of no-arbitrage. The
certainty equivalent of a portfolio of two uncertain cash ‡ows is only equal to the sum of
their individual certainty equivalents if these cash ‡ows are stochastically independent and
the decision maker’s preferences have no wealth e¤ects (i.e., exponential utility). Hence, it
is impossible to determine the certainty equivalents of individual securities in a meaningful
way even if the investors’ utility functions are known—the certainty equivalent of an asset
depends on the other risks the decision maker has in his portfolio (and in his other income).
Moreover, the certainty equivalent of optimal invested wealth is the value the decision maker
attaches to total wealth (like in a take-it-or-leave-it o¤er), whereas the equilibrium prices of
risky securities are determined such that no investor has any incentive to change his optimal
portfolio of securities (on the margin). In a general equilibrium model with HARA utili-
ties, we show that this implies that the risk-adjusted expected cash ‡ows are lower than the
certainty equivalents.
If the language of certainty equivalents is retained for risk-adjusted expected cash ‡ows,
the meaning should be clear: the certainty equivalent of an uncertain cash ‡ow is the certain
cash ‡ow which has the same equilibrium value as the uncertain cash ‡ow. Of course, this
de…nition of certainty equivalents is moot until it is speci…ed how it can be calculated. Unfor-
tunately, no-arbitrage alone provides no guide to actually determining the risk-adjustments
to expected cash ‡ows. No-arbitrage pricing is very useful to …nd the relative prices between
5
redundant derivative securities and the primary securities. However, an equilibrium model
is required to determine the prices of the primitive assets spanning the market risks.
We assume that equilibrium prices are determined in an e¤ectively dynamically complete
market (such that equilibrium allocations are Pareto optimal), and that the investors have
homogeneous beliefs and time-additive preferences. In that case, it is well-known (see, for
example, Christensen and Feltham, 2003, Chapter 6) that the risk-adjustment to expected
cash ‡ows can be determined as the covariance between the cash ‡ow and a valuation in-
dex measurable with respect to aggregate consumption at that date. Furthermore, if it is
assumed that aggregate (or log-aggregate) consumption and the cash ‡ow are jointly nor-
mally distributed, then Stein’s Lemma can be used to separate the valuation index out of
the covariance as a factor equal to the expected value of the derivative of the valuation in-
dex. Hence, the risk-adjustment is determined as the covariance between the cash ‡ow and
(log-)aggregate consumption times this valuation-index factor. The latter can be determined
from the market prices of aggregate consumption claims (like in the standard CAPM), or
by using a general equilibrium approach, i.e., by assuming a particular set of investor pref-
erences. In the latter approach, normally distributed aggregate consumption goes well with
exponential utilities, whereas log-aggregate consumption goes well with power utilities. The
former combination is probably the best known in accounting research, whereas the latter is
the preferred combination in …nance. Hence, we will show both.
The remainder of the paper is organized as follows. In the …rst section we give a short
review of standard valuation models based on risk-adjusted discount rates with an emphasis
on the many implicit assumptions made in this approach. In Section 3 we review key
results frommulti-period asset pricing theory in discrete-time, and show how equity valuation
models can equivalently be based on free cash ‡ows or accrual accounting numbers. Based on
these results, we derive an accounting-based multi-period equity valuation model in Section
4 with equilibrium risk-adjustments determined by the prices of aggregate consumption
claims. Given the analysis in the preceding section, a similar model can be derived using
free cash ‡ows. Section 5 includes a general equilibrium analysis of a setting in which
the investors have exponential utility, and aggregate consumption and residual operating
income are jointly normally distributed. Appendix B shows how the analysis is adjusted if
aggregate consumption is lognormally distributed and investors have power utility (average
risk tolerance is simply replaced by the common risk cautiousness), Appendix C extends the
setting to preferences with external habit formation (which recently has gained popularity
in asset pricing theory), and Appendix D discuss the relationship between risk-adjusted
expected cash ‡ows (or accounting numbers) and certainty equivalents. For purpose of
illustration, Section 6 examines a simple setting in which residual operating income and
6
aggregate consumption are given by a simple …rst-order vector-autoregressive model with
mean-reversion to a deterministic exponential trend. The analysis stresses the importance
of both the contemporaneous correlation between residual operating income and aggregate
consumption and the time-series properties of these processes. This analysis is the basis for
the comparison of the general equilibrium analysis and the standard approach of using a
time-independent risk-adjustment to the required rates of returns in Section 7. We skip the
concluding remarks. The reader is probably already su¢ciently confused at this point—but
hopefully at a higher level.
2 Risk-adjusted Discount Rates
In this section, we review the standard approaches to (equity) valuation using risk-adjusted
discount rates. Emphasis will be on the many implicit assumptions made in this approach.
If you at any point reading through this section have had enough of outrageous assumptions,
you should immediately stop reading and go to the next section. Indeed, this section is
written in order to motivate you to be open-minded for alternative approaches to equity
valuation. The following sections aim at developing such alternative approaches.
The common practice in equity valuation is to use some kind of dividend, free cash ‡ow,
residual operating income, or abnormal operating income growth discount model. In the
latter three models, the discount rate used is the …rm’s cost of capital, while an equity cost
of capital is used in the former model. For example, using free cash ‡ows, ,c,
t
, the value of
the …rm (also denoted the enterprise value) at the valuation date t, \
t
, is determined as the
discounted expected free cash ‡ows conditional on information available at date t, i.e.,
1
\
t
=
1

t=t+1
Et
[,c,
t
]
11
tt
t
. (1)
where 11
t
is the …rm’s gross cost of capital at the date t. Of course, taking the value of the
…rm and the expected future free cash ‡ows as given, the …rm’s cost of capital is merely an
internal rate of return, i.e., the implied cost of capital. This construct (or plug number) may
be useful for strategy evaluation within the …rm, but it is certainly not useful for security
valuation or net present value calculations within the …rm—in these cases we are trying to
determine the current value of the sequence of future uncertain cash ‡ows. Instead, we must
determine the …rm’s cost of capital directly.
1
In this section, we will be using the free cash ‡ow discount model to illustrate the standard valuation
approaches, but we could equally well have used the accounting-based models. These latter models will take
center stage in subsequent sections.
7
2.1 Enterprise Valuation
Suppose, for the moment, we are able the estimate the expected one-period ahead gross
returns on …rm value,
Et
[11
t+1,t
], given the available information at date t. t = t. t + 1. ...
(see below). By the de…nition of conditional expected returns, it follows that
\
t
=
Et
[,c,
t+1
+ \
t+1
]
Et
[11
t+1,t
]
. t = t. t + 1. ...
Using this relation iteratively starting at date t, we get that
\
t
=
Et
[,c,
t+1
]
Et
[11
t+1,t
]
+
Et
[
Et+1
[,c,
t+2
+ \
t+2
]
Et+1
[11
t+2,t+1
]
1
]
Et
[11
t+1,t
]
The one-period ahead conditional expected return at t + 1 is, in general, a random variable
given date t information. Hence, by the formula for covariances and iterated expectations,
it follows that
\
t
=
Et
[,c,
t+1
]
Et
[11
t+1,t
]
+
Et
[,c,
t+2
+ \
t+2
]
Et
[
Et+1
[11
t+2,t+1
]
1
]
Et
[11
t+1,t
]
+
Cov
t
[
Et+1
[,c,
t+2
+ \
t+2
].
Et+1
[11
t+2,t+1
]
1
]]
Et
[11
t+1,t
]
=
Et
[,c,
t+1
]
Et
[11
t+1,t
]
+
Et
[,c,
t+2
+ \
t+2
]
Et
[
Et+1
[11
t+2,t+1
]
1
]
Et
[11
t+1,t
]
+ Cov
t
_
Et+1
[,c,
t+2
+ \
t+2
].
Et+1
[11
t+2,t+1
]
1
Et
[
Et+1
[11
t+2,t+1
]
1
]
_
Et
[
Et+1
[11
t+2,t+1
]
1
]
Et
[11
t+1,t
]
Clearly, using one-period ahead expected returns iteratively for valuation of multi-period
cash ‡ow streams is problematic for several reasons. First, as it appears from the last term,
this procedure requires a model for how the future cum-dividend value of free cash ‡ows,
i.e., ,c,
t+2
+ \
t+2
, are correlated with future conditional expected returns. If the expected
returns are de…ned as the spot riskless interest rate plus a risk premium, this correlation
does not appear to be trivial. For example, in a single-factor interest rate model, like the
Cox, Ingersoll, and Ross (1985) model, a lower spot interest rate would lead to a higher
ex-dividend value of future cash ‡ows even if these are non-stochastic (due to lower discount
rates). Secondly, even if we assume that cum-dividend values are uncorrelated with expected
8
returns, the discount factor in the second term is higher than if the discount factor is based
on the expected returns based on date t information (due to Jensen’s inequality), i.e.,
Et
[
Et+1
[11
t+2,t+1
]
1
]
Et
[11
t+1,t
]

1
Et
[11
t+1,t
]
Et
[
Et+1
[11
t+2,t+1
]]
=
1
Et
[11
t+1,t
]
Et
[11
t+2,t+1
]
.
with the di¤erence being determined by the uncertainty in next period’s conditional expected
return. Hence, in this setting, “average” expected future returns tend to overstate the implied
cost of capital in (1). Hughes, Liu and Liu (2008) develop these ideas more fully and show
how it a¤ects the interpretation of empirical …ndings for the relationship between …rms’ cost
of capital and the informational environment.
Valuation based on estimates of risk-adjusted short-term expected returns is immensely
simpli…ed if it is assumed that future expected returns are deterministic. In that case, we
can determine the current value of the …rm as
2
\
t
=
1

t=t+1
Et
[,c,
t
]

t
c=t
Ec
[11
c+1,c
]
.
Deterministic one-period ahead expected returns require that future spot interest rates and
risk premia are deterministic. Suppose the risk premia are the same in all future periods.
We may then write these expected returns as
Ec
[11
c+1,c
] = (1 +t
c
) (1 + `)
where t
c
is the spot one-period interest rate at date : and ` is the constant risk premium.
Hence, the discount factors may be written as

t
c=t
Ec
[11
c+1,c
] = (1 +`)
tt

t
c=t
(1 +t
c
)
Note that with deterministic interest rates, one-period forward rates are equal to the asso-
ciated future spot interest rates and, thus, the zero-coupon interest rates, t
tt
, at date t for
maturity dates t, are given by
(1 +t
tt
)
tt
=

t
c=t
(1 +t
c
) .
2
A transversality condition is also at work here in an in…nite horizon setting.
9
Hence, the value of the …rm may be written as
\
t
=
1

t=t+1
Et
[,c,
t
]
[(1 +t
tt
) (1 + `)]
tt
.
Equating this expression for the value of the …rm with the …rm value in the de…nition of
the …rm’s cost of capital (1), how is the cost of capital 11
t
a¤ected by, for example, growth
in free cash ‡ows? In order to illustrate this as simply as possible, consider an example in
which the …rm’s future free cash ‡ows are deterministic (like a default-free bond). In that
case, there are no risk premia, i.e., ` = 0. Table 1 shows the prices of two such two-period
cash ‡ow sequences using a given set of zero-coupon interest rates.
Date t + 1 Date t + 2 Price, \
t
11
t
÷1 WACC
Interest rates, t
tt
4% 6%
Cash ‡ow sequence ¹ 50 100 137,08 5,58%
Cash ‡ow sequence 1 100 50 140,65 4,96%
Portfolio of ¹ and 1 150 150 277,73 5,30% 5,27%
Table 1: Cost of capital for a non-‡at term structure of interest rates.
There are several important things to note from this very simple example. Note …rst that
the term structure of interest rates is assumed to be increasing (which is the most common
empirically). This implies that the increasing sequence ¹ has a higher cost of capital, i.e.,
internal rate of return, than the decreasing sequence 1. Of course, this is due to the fact
that the cash ‡ows in the increasing sequence ¹ “load” relatively more on the high two-
period zero-coupon interest rate than on the low one-period zero-coupon interest rate. This
illustrates the fact that with non-‡at term structures of interest rates the cost of capital
used to discount multi-period sequences of cash ‡ows must re‡ect not only risk but also the
expected growth in cash ‡ows.
3
In implementations of the standard approaches to equity
valuation using risk-adjusted discount rates, it is well appreciated that the term structure
of interest rates is not ‡at, and the recommendation in almost all textbooks is to use a
10-year treasury yield as a proxy for the riskless interest rates. It is less appreciated that
the cost of capital must also re‡ect the growth in the sequences of cash ‡ows which are to
be discounted—one can only guess the size of the error introduced.
The last line in Table 1 shows the price and the cost of capital of a portfolio of the two
cash ‡ow sequences ¹ and 1. Of course, the cost of capital on the portfolio is somewhere
3
This is well known in the …xed income literature: the yield-to-maturity on a bond does not only depend
on its maturity, it also depends on its payment pro…le (see, e.g., Caks, 1977, and Christensen and Nielsen,
1987, for the so-called coupon- and bond-type e¤ects on yield-to-maturity, respectively). Hence, in the …xed
income literature discounting is always done using the term structure of zero-coupon interest rates.
10
between the cost of capital on the individual cash ‡ow sequences. The common approach
in equity valuation is to calculate the cost of capital for portfolios of, for example, debt and
equity, as a weighted average cost of capital (WACC) using the relative values as weights.
Note that, in the example, the WACC underestimates the cost of capital on the portfolio.
This re‡ects that, in general, for multi-period sequences of cash ‡ows, the cost of capital on
a portfolio of these sequences is a highly non-linear weighted average of the cost of capital on
the individual sequences. The WACC calculation assumes it is a linear weighted average, and
this only holds in single-period settings and in settings in which the sequences of expected
cash ‡ows are assumed to grow at the same rate in perpetuity.
2.2 Estimating the Cost of Capital
In practice, the …rm’s cost of capital 11
t
is not estimated directly. Instead, it is estimated
indirectly by …rst estimating the the equity cost of capital 11
t
and the debt cost of capital
11
t
and then from these calculate the …rm’s cost of capital using accounting identities and
value additivity principles. A common approach is to determine it as the weighted average
cost of capital, i.e.,
11
t
- \¹CC
t
=
o
t
o
t
+ 1
t
11
t
+
1
t
o
t
+ 1
t
11
t
, (2)
where o
t
and 1
t
are the market values of the …rm’s common equity and net debt, respectively,
with o
t
+ 1
t
= \
t
. Like for the …rm’s cost of capital, we may estimate the cost of capital
for debt and equity as the implied cost of debt and equity, respectively, taking the market
values of debt and equity and their expected net cash ‡ows as given. That is, the implied
cost of capital for the two sources of capital are determined by
o
t
=
1

t=t+1
Et
[n
t
]
11
tt
t
.
1
t
=
1

t=t+1
Et
[.
t
]
11
tt
t
.
where n
t
and .
t
are the net payments to equity- and debtholders at date t, respectively, with
the sum being equal to the free cash ‡ow by the …rm’s budget constraint, ,c,
t
= n
t
+ .
t
.
If the expected payments are all in…nite sequences growing at the same constant rate, then
it is easy to show that the …rm’s cost of capital is indeed equal to the weighted average cost
of capital but, otherwise, it does not hold—the internal rate of return on a portfolio is not
11
equal to the weighted average of the individual assets’ internal rates of return.
4
If debt is
close to being riskless debt, it might be a reasonable approach to estimate the cost of debt as
its implied cost of capital. However, we are trying to determine the value of equity, so this
approach clearly does not work for equity—we must determine the cost of equity directly.
5
Before we go into the details of how the cost of equity is determined empirically, it is
useful to look at a simple T-account (in market value terms) in order to keep the language
straight.

t
o
t
1o
t
1
t
1C
t
:
\
t
\
t
The debits are the value of an otherwise identical unleveraged …rm, l¹
t
, the value of the
future corporate tax savings on net debt, 1o
t
, the value of future bankruptcy costs, 1C
t
,
plus any other debt related items. Hence, we may determine the value of the …rm both as
the sum of debt and equity values, i.e.,
\
t
= o
t
+ 1
t
. (3)
and as the sum of the values of the comprehensive operating assets, i.e.,
\
t
= l¹
t
+ 1o
t
+ 1C
t
+ ... (4)
The latter is the so-called adjusted present value (APV) method. Combining the two rela-
tions yields the value of equity as
o
t
= (l¹
t
+ 1o
t
+ 1C
t
+ ...) ÷1
t
. (5)
The APV-method seems to be the recommended method in standard …nance textbooks
(see, for example, Grinblatt and Titman, 2002), whereas most accounting textbooks seem to
4
Let both E
t
[n

] and E
t
[.

] grow at a constant rate q = G÷1. Then
o
t
= E
t
[n
t+1
],(11
t
÷G), 1
t
= E
t
[.
t+1
],(11
t
÷G), and \
t
= E
t
[)c)
t+1
],(11
t
÷G).
Using the …rm’s budget constraint, we get
11
t
= [E
t
[n
t+1
] + E
t
[.
t+1
]] ,\
t
+ G = [o
t
(11
t
÷G) + 1
t
(11
t
÷G)] ,\
t
+ G = (o
t
,\
t
)11
t
+ (1
t
,\
t
)11
t
which is (2).
5
Did you notice the circularity in the \¹CC calculation?
12
recommend the WACC-method to discount free cash ‡ows from operations with an after-tax
weighted average cost of capital (see, for example, Penman, 2007, and Lundholm and Sloan,
2004). The advantage of the APV-method is that it directly follows the value additivity
principle such that the value of the …rm is given as the sum of the values of the di¤erent
components. This gives much more ‡exibility in how, for example, one wants to model the
tax savings from debt and how to include an estimated value of future bankruptcy costs.
The latter, of course, is most relevant for highly leveraged …rms. On the other hand, the
WACC-method has problems in including bankruptcy costs consistently, and it requires a
particular and strong assumption about the future debt policy (see below).
6
The conceptual di¤erence between the two approaches is what is included in the operating
assets, the free cash ‡ows, and in the operating income in accounting-based models, in
particular, how the corporate tax savings on net interest payments are dealt with. The
APV-method has a comprehensive de…nition of operating assets, operating income, and free
cash ‡ows, whereas the WACC-method uses a narrowde…nition of operating assets, operating
income, and free cash ‡ows, i.e., excluding bankruptcy costs and the corporate tax savings
on net interest (and any associated deferred tax bene…ts). In the following discussion we
will, for simplicity, ignore any bankruptcy costs, but it should be clear from the discussion
above how to include these in the APV-method.
2.2.1 The WACC-method
As above, de…ne comprehensive free cash ‡ows, ,c,
t
, as the net cash ‡ow paid to debt and
equity, and free cash ‡ows from operations, ,c,o
t
, as everything in the comprehensive free
cash ‡ows which has nothing to do with the …rm’s …nancing, i.e., the “unleveraged free cash
‡ows.” If there are no bankruptcy costs, the di¤erence between the two is the tax savings from
net interest. Assume, for simplicity, that these tax savings are equal to the corporate tax
rate t
c
times one minus the expected one-period ahead return on debt, i.e.,
Et
[11
t+1,t
] ÷1,
times the opening value of net debt, i.e.,
,c,
t+1
= ,c,o
t+1
+ t
c
[
Et
[11
t+1,t
] ÷1] 1
t
.
By the de…nition of the …rm’s expected one-period ahead return, it then follows that
\
t
=
Et
[,c,o
t+1
+ \
t+1
] + t
c
[
Et
[11
t+1,t
] ÷1] 1
t
Et
[11
t+1,t
]
.
6
Here, bankruptcy cost must not be taken to mean higher yields on debt due to default risk. Bankruptcy
costs are deadweight losses associated with default.
13
Starting at the valuation date t and rewriting, we get that
\
t
=
Et
[,c,o
t+1
+ \
t+1
]
Et
[11
t+1,t
] ÷t
c
[
Et
[11
t+1,t
] ÷1]
1t
\t
.
Note from (3) that we can view the value of the …rm as the value of a portfolio of the net debt
and the equity. Hence, it follows (from basic portfolio theory) that the expected one-period
ahead return on the …rm is equal to the weighted average of the expected one-period ahead
returns on the individual assets in the portfolio, i.e.,
Et
[11
t+1,t
] =
Et
[11
t+1,t
]
o
t
\
t
+
Et
[11
t+1,t
]
1
t
\
t
.
Substitution then yields that
\
t
=
Et
[,c,o
t+1
+ \
t+1
]
Et
[11
t+1,t
]
St
\t
+
Et
[11
t+1,t
]
1t
\t
÷t
c
[
Et
[11
t+1,t
] ÷1]
1t
\t
or, equivalently,
\
t
=
Et
[,c,o
t+1
+ \
t+1
]
Et
[\¹CC
¹T
t
]
. (6)
where
Et
[\¹CC
¹T
t+1,t
] =
Et
[11
t+1,t
]
o
t
\
t
+ [
Et
[11
t+1,t
] ÷t
c
[
Et
[11
t+1,t
] ÷1]]
1
t
\
t
. (7)
That is, the WACC-approach takes the free cash ‡ows from operations, ,c,o
t+1
, and recog-
nizes the tax shield from net debt in the after-tax discount factor, i.e., the weighted average
of expected one-period ahead after-tax returns,
Et
[\¹CC
¹T
t+1,t
].
We can now try to use (6) iteratively as we did in Section 2.1, and we will encounter
exactly the same problems with stochastic expected one-period ahead returns as well as some
additional problems. In order to use (6) to derive the familiar WACC-valuation formula,
\
t
=
1

t=t+1
Et
[,c,o
t
]
Et
[\¹CC
¹T
t+1,t
]
tt
.
we must assume that the weighted average of expected one-period ahead after-tax returns
are constant over time, i.e.,
Et
[\¹CC
¹T
t+1,t
] =
Et
[\¹CC
¹T
t+1,t
]. t = t + 1. ... This means that
not only are the expected one-period ahead expected returns on debt and equity constants,
but the weights must also be constants, i.e., leverage, 1
t
,\
t
, must be constant (which also
means that o
t
,\
t
is a constant, since \
t
= o
t
+\
t
). Hence, at each instant the stock price
14
changes, new debt has to be issued or existing debt must be retired to keep the leverage
constant (since debt has a stock price sensitivity di¤erent from one). Of course, changes in
leverage is closely related to changes in expected one-period ahead expected returns on debt
and equity. Would it be possible that these changes are o¤setting such that the expected
discount factors
Et
[\¹CC
¹T
t+1,t
] are independent of leverage? Unfortunately, the answer is
no. This is because the discount factors include the tax shields from net debt.
2.2.2 The APV-method
The APV-method calculates the values of the free cash ‡ows from operations and the tax
shields of interest on net debt separately, i.e.,
\
t
= l¹
t
+ 1o
t
The value of the free cash ‡ows from operations at date t is

t
=
Et
[,c,o
t+1
+ l¹
t+1
]
Et
[1l¹
t+1,t
]
.
where
Et
[1l¹
t+1,t
] is the expected one-period ahead return on the unleveraged assets. Note
that the expected one-period ahead returns on the unleveraged assets must be independent
of leverage—this is Modigliani/Miller at work. Hence, compared to the WACC-method, it
is less restrictive to assume that the expected one-period ahead returns on the unleveraged
assets are constant, so we may determine the value of the unleveraged assets as

t
=
1

t=t+1
Et
[,c,o
t
]
Et
[1l¹
t+1,t
]
tt
.
The value of the tax shields of interest on net debt is similarly determined as
1o
t
=
t
c
[
Et
[11
t+1,t
] ÷1] 1
t
+
Et
[1o
t+1
]
Et
[11o
t+1,t
]
.
where
Et
[11o
t+1,t
] is the expected one-period ahead return on the tax savings. As noted
above, an advantage of the APV-method over the WACC-method is that the assumptions on
future tax shields can be made independently. Two common assumptions in the literature
are that (A) the expected amounts of future tax shields are equal to the expected current tax
shield, and (B) the amount of future debt is continuously adjusted so that the tax shields are
proportional to the free cash ‡ows from operations—but, of course, any other assumption is
also feasible.
15
(A) Constant expected tax shields
In this setting, the expected one-period ahead return on the tax savings is the same as the
expected one-period ahead return on debt, i.e.,
Et
[11o
t+1,t
] =
Et
[11
t+1,t
].
and assuming this is a constant in perpetuity, we get
1o
t
=
t
c
[
Et
[11
t+1,t
] ÷1] 1
t
Et
[11
t+1,t
] ÷1
= t
c
1
t
.
Using (5) and the explicit value of the tax savings, we can express the expected one-period
ahead return on equity as follows,
Et
[11
t+1,t
] =
Et
[1l¹
t+1,t
]
_
1 +
1
t
÷1o
t
o
t
_
+
Et
[11o
t+1,t
]
1o
t
o
t
÷
Et
[11
t+1,t
]
1
t
o
t
=
Et
[1l¹
t+1,t
] + (
Et
[1l¹
t+1,t
] ÷
Et
[11
t+1,t
])
(1 ÷t
c
) 1
t
o
t
. (8)
or in “beta” terms
7
,
1
t
= ,

t
+
_
,

t
÷,
1
t
_
(1 ÷t
c
) 1
t
o
t
. (9)
These are the standard relations to unlever expected equity returns and equity betas, but
recall that we have made a very strong assumption about constant expected one-period
ahead returns on debt in order to establish these relations.
(B) Proportional tax shields
In this setting, the risks associated with future tax shields are the same as the risks associated
with future free cash ‡ows from operations and, hence,
Et
[11o
t+1,t
] =
Et
[1l¹
t+1,t
].
The determination of the …rm value can therefore be based on the comprehensive free cash
‡ows discounted at the expected one-period ahead returns on the unleveraged assets, i.e.,
\
t
=
Et
[,c,
t+1
+ \
t+1
]
Et
[1l¹
t+1,t
]
. (10)
Recall that the expected one-period ahead returns on the unleveraged assets are independent
7
The beta of an asset is de…ned as the ratio of the conditional covariance between the asset’s return and
some factor, and the conditional variance of the factor. Note that the factor does not have to be the return
on the market portfolio. Hence, there is no implicit assumption of the CAPM being true at this point.
16
of leverage. Note that, contrary to WACC-method, the tax shields are re‡ected in the
comprehensive free cash ‡ows, and not in the discount factor. This is also re‡ected in how
the expected one-period ahead returns on equity and equity betas are unleveraged in this
setting. Again, using (5) we get that
Et
[11
t+1,t
] =
Et
[1l¹
t+1,t
]
_
1 +
1
t
÷1o
t
o
t
_
+
Et
[11o
t+1,t
]
1o
t
o
t
÷
Et
[11
t+1,t
]
1
t
o
t
=
Et
[1l¹
t+1,t
] + (
Et
[1l¹
t+1,t
] ÷
Et
[11
t+1,t
])
1
t
o
t
. (11)
and that
,
1
t
= ,

t
+
_
,

t
÷,
1
t
_
1
t
o
t
. (12)
That is, these relations are precisely the same as in a before tax setting.
Of course, we can also rewrite (11) to obtain a before-tax weighted average of expected
one-period ahead returns, i.e.,
Et
[\¹CC
1T
t+1,t
] =
Et
[1l¹
t+1,t
] =
Et
[11
t+1,t
]
o
t
\
t
+
Et
[11
t+1,t
]
1
t
\
t
. (13)
Note that the before-tax
Et
[\¹CC
1T
t+1,t
] is independent of leverage (as opposed to the after-
tax
Et
[\¹CC
¹T
t+1,t
] in (7)), since it is determined as the expected returns in an otherwise
identical unleveraged …rm. Hence, if the expected equity returns are estimated using a
time-series of stock returns over a period with relatively constant leverage, and the expected
equity returns are unleveraged by (11), then the estimate of the before-tax
Et
[\¹CC
1T
t+1,t
] is
also relatively precise. Assuming that these expected unleveraged returns are constant, and
using the one-period value relation (10) iteratively, we get
\
t
=
1

t=t+1
Et
[,c,
t
]
Et
[\¹CC
1T
t+1,t
]
tt
.
and comparing to (2) it follows that the …rm’s cost of capital is 11
t
=
Et
[\¹CC
1T
t+1,t
].
But again, why should the expected one-period ahead returns on the unleveraged assets be
constant in a multi-period setting under uncertainty?
2.2.3 Estimating the equity cost of capital
The common approach to determine the equity cost of capital is to use one of the standard
asset pricing models such as the CAPM or the APT model. For example, the CAPM says
that the expected one-period ahead return on a stock is equal to the riskless return plus the
17
expected one-period ahead excess return on the market portfolio times the stock’s beta, i.e.,
Et
[11
t+1,t
] = 1
t+1,t
+ [
Et
[1`
t+1,t
] ÷1
t+1,t
] ,
1
t
. (14)
where 1
t+1,t
= 1 +t
t
is one plus the riskless spot interest rate, 1`
t+1,t
is the gross return
on the market portfolio, and
,
1
t
=
Cov
t
[11
t+1,t
. 1`
t+1,t
]
Var
t
[1`
t+1,t
]
.
The …rst step is to estimate the equity beta as the slope of a time-series regression of equity
returns on the returns of some stock index (as a proxy for the market portfolio). Of course,
this assumes that the equity beta is constant over the estimation period. As we noted above,
this requires that leverage is relatively constant over the estimation period (and that the
unleveraged equity and debt betas are also constant). Two types of regression equations are
typically used. The …rst uses a time-series of raw returns over the estimation period, i.e.,
11
t+1,t
= c
1
t
+ ,
1
t
1`
t+1,t
+
t+1,t
. t = :. : + 1. .... t ÷1. (15)
where 11
t+1,t
and 1`
t+1,t
are the observed returns in period t on the stock and the stock
index, respectively.
8
The second recognizes that the riskless spot interest rate varies over
the estimation period and, thus, uses excess returns, i.e.,
11
t+1,t
÷1
t+1,t
= c
1
t
+ ,
1
t
[1`
t+1,t
÷1
t+1,t
] +
t+1,t
. t = :. : + 1. .... t ÷1. (16)
Under the assumptions of the CAPM with varying interest rates, there is an omitted variable
in (15), while (16) provides a direct test of the validity of the CAPM, i.e., whether the
estimated alpha in (16) is di¤erent from zero. On the other hand, it is well known that the
CAPMis only valid in a multi-period setting if the investment opportunity set is deterministic
or investors are myopic, i.e., have log utilities. A deterministic investment opportunity set
means that the so-called capital market line, i.e.,
Et
[11
t+1,t
] = 1
t+1,t
+
[
Et
[1`
t+1,t
] ÷1
t+1,t
]
Var
t
[1`
t+1,t
]
1
2
Var
t
[11
t+1,t
]
1
2
.
where 11
t+1,t
is the return on an optimally invested portfolio, can only shift in a determin-
istic manner. That is, the intercept, which is the riskless return, and the slope, which is the
8
We are not going to discuss the econometric problems in the estimation such as the fact that using a
stock index as proxy for the market portfolio induces an errors-in-variables problem resulting in a downward
biased beta estimat.
18
“market price of risk,” must both be non-stochastic. Hence, there is an implicit assumption
in the test for a zero alpha in (16) that spot interest rates are deterministic.
The next step is to estimate the expected one-period ahead excess return on the market
portfolio. It might be tempting to estimate this expected return simply by calculating the
average excess return on the stock index over the estimation period. However, if continuously
compounded returns are used in the estimation (which is generally recommended), then only
the starting and ending cum-dividend values of the index over the estimation period a¤ect
the estimate. Hence, such estimates are very unreliable. Instead, relying on the CAPM, a
Fama-MacBeth type cross-section regression of average returns on beta-sorted portfolios on
their estimated betas is typically used to estimate the expected excess returns on the market
portfolio, i.e.,
11
i
= ¸
0
+ ¸
A
´
,
i
+ n
i
.
Among the empirical stylized facts for these regressions are that the estimates of the market
risk premium ¸
A
are typically too low compared to average excess returns on broad stock
indices, and that there is a lot of variation in the estimates across di¤erent empirical studies.
The following quote is from Penman (2007), page 691:
“Compound the error in beta and the error in the risk premium and you have a
considerable problem. The CAPM, even if true, is quite imprecise when applied.
No one knows what the market premium is. And adopting multifactor pricing
models adds more risk premiums and betas to estimate. These models contain a
strong element of smoke and mirrors.”
In the recent asset pricing literature, the game in town is time-varying risk premia.
There is signi…cant empirical evidence that excess returns are time-varying and to some
extent predictable (see, for example, Campbell and Shiller, 1988a and 1988b, and Lettau
and Ludvigson, 2001). This may be a useful way to proceed, but stochastic expected one-
period ahead returns remain a problem for iteratively using single-period value relations to
value a multi-period sequence of future cash ‡ows or accounting numbers, i.e., throwing away
potentially signi…cant covariances, violating Jensen’s inequality, non-‡at term structures of
interest rates, etc.
2.3 Value Additivity
As noted above a fundamental property of valuation in perfect markets is the value additivity
principle, i.e., the value of a portfolio of two sequences of cash ‡ows must be equal to the sum
of the values of the individual sequences of cash ‡ows. The standard valuation approaches
19
reviewed above do not satisfy this principle. Even if we are successful in estimating the
expected one-period ahead return on equity (and debt), unlevering it to get a risk-adjusted
discount rate for the operations, and we are reasonably certain that it is constant through
time, then the risk-adjusted discount rate re‡ects the risk in the operations from now un-
til perpetuity. Hence, the risk-adjusted discount rate used to discount a particular future
expected cash ‡ow does not re‡ect the risk in that cash ‡ow but, rather, the risk in the
whole sequence of future cash ‡ows. Hence, the value of the sequence of future cash ‡ows is
determined as the value of the portfolio of future cash ‡ows and not, necessarily, as the sum
of the value of the individual future cash ‡ows.
Ang and Liu (2004) recognize that one should not discount future expected cash ‡ows
with the same discount rate. They propose a model in which a term structure of required
rates of return is used to discount future expected payments (see also Brennan, 1997, and
Brennan and Xia, 2006). That is, the value of a sequence of uncertain future cash ‡ows, say
free cash ‡ows, is determined as
\
t
=
1

t=t+1
Et
[,c,
t
]
11
tt
tt
. (17)
where the risk-adjusted discount rates, i.e., 11
tt
÷ 1, are date-speci…c (like zero-coupon
rates). They parameterize their model with speci…c stochastic processes for asset betas and
market risk premia (in return form) and …nd that signi…cant pricing errors can occur if
a constant discount rate is used. Hence, it appears as if they determine the value of the
sequence of future cash ‡ows as the sum of the values of the individual future cash ‡ows
using speci…c risk-adjusted discount rates for each future expected cash ‡ow. However, this
might in fact not be the case, because the starting assumption is a model for how market
risk premia and how the return betas for the whole sequence of cash ‡ows evolve over time.
More direct evidence that “there is something rotten in the State of Denmark” is that the
value additivity principle and (17) imply that the current value of ,c,
t
for some particular
future date t must be positive if
Et
[,c,
t
] 0 (at least if t ÷t is even). Obviously, this needs
not to be the case.
Besides not satisfying the value additivity principle, the trouble with this kind of model is
that everything is hidden in the date-speci…c risk-adjusted discount rates—and in a very con-
voluted way. For example, the information in the current directly observable term structure
of interest rates is not used—the date-speci…c risk-adjusted discount rate is some compli-
cated mixture of the zero-coupon interest rate for that date, the date-speci…c market price of
risk, and the date-speci…c systematic risk of the cash ‡ow, where the latter is determined not
only by the contemporaneous covariance with the market return but also by the time-series
20
properties of cash ‡ows and market returns.
The approach described in the following sections aims at separating out these various
e¤ects on the current value of some future uncertain cash ‡ow (or residual operating in-
come/abnormal operating income growth), and such that the value of a sequence of future
cash ‡ows can be determined as a sum of the values of the individual cash ‡ows, i.e.,
\
t
(¦,c,
t
¦
tt
) =
1

t=t+1
\
t
(,c,
t
).
This requires that we go back to the fundamentals of asset pricing in multi-period settings
in which we determine the risk and the values of the individual future cash ‡ows. In order
to sketch the approach, we can use the well known result that no-arbitrage in the …nancial
market implies that the current value of a future cash ‡ow ,c,
t
is given as
\
t
(,c,
t
) =
Et
[,c,
t
] + Cov
t
[,c,
t
. ¡
t
]
(1 +t
tt
)
tt
.
where t
tt
is the zero-coupon interest rate at date t for maturity t, and ¡
t
is the ratio
between the risk-neutral probabilities based on zero-coupon bonds (the so-called forward
probabilities) and the true conditional probabilities. This latter object is the Radon-Nikodym
derivative for the two probability measures, or the so-called valuation index. Hence, the
discounting for time and the adjustment for risk are separated—the risk-adjustment is made
in the numerator, and the discounting for time is done through the zero-coupon interest rates
in the denominator. Obviously, the task at hand then is to determine the risk-adjustment in
the numerator—the zero-coupon interest rates are almost directly observable. This requires
that we specify the stochastic process for the future cash ‡ows (but not the returns which
depend on future prices), and how fundamental risks associated with future cash ‡ows are
priced in the market through the valuation index.
3 Multi-period Asset Pricing Theory and Accounting
Relations
In this section we …rst provide a short review of multi-period asset pricing theory starting
with the implications of assuming no-arbitrage in the …nancial markets. We then intro-
duce the basic accounting relations for …nancial statement articulation within and across
periods. These relations are the basis for the equivalence of the discounted free cash ‡ow
and accounting-based valuation models. The section concludes with the determination of
21
equilibrium valuation indices in e¤ectively dynamically complete markets with time-additive
utility and homogenous expectations.
3.1 Fundamental Theorem of Asset pricing
Our model closely follows the discrete-time model in Christensen and Feltham, 2003, Chap-
ters 6 and 9, except that we assume a continuous state space (in order to allow for continuous
distributions such as normal and log-normal distributions). Let the standard probability
space (. . ) with homogeneous investor beliefs be given, and denote probabilizable
events observable at date t by ¸
t
¸
t
. Trading and consumption take place at dates
t = t. t + 1. .... 1 conditional on public information
t
.
9
Date t will throughout denote
the current date at which the valuation is performed, and dates t = t + 1. .... 1 denote fu-
ture dates. Dividends and ex-dividend prices of marketed securities are stochastic processes
adapted to the …ltration = ¦
t
¦
t=t,...,T
. That is, the dividend d
)t
and the ex-dividend
price \
)t
of security , = 1. .... J at date t can be written as functions of the observable events
at that date, i.e., d
)t
= d
)t

t
) and \
)t
= \
)t

t
), respectively. The fundamental theorem
of asset pricing can now be stated as follows.
Theorem 1 No-arbitrage in the securities market (and mild regularity conditions) implies
that there exists a strictly positive event-price measure 1(¸
t

t
) such that the ex-dividend
price at date t of any marketed security can be written as
\
)t

t
) =
T

t=t+1
_
d
)t

t
) d1(¸
t

t
). , = 1. .... J. (18)
The linearity of the pricing functional is evident at this point. The event-price measure
applies to any stream of marketed future dividends, and summations over future dates and
integrals over the future events are linear operators. Hence, the prices of securities and
portfolios of securities are determined as a “summation” of its date-event speci…c dividends
times the date-event speci…c event prices. This is the value additivity principle in perfect
(although not necessarily complete) …nancial markets. The following analysis re-expresses
this fundamental asset pricing result in several di¤erent ways, but in all cases such that value
additivity principle is preserved.
As is well known, the fundamental no-arbitrage asset pricing relation in (18) can be
given alternative representations. One set of representations normalizes the event-prices,
and another set of representations re-expresses the dividends in terms of free cash ‡ows or
9
We assume a …nite horizon setting here in order to avoid additional, mainly technical, issues of asset
pricing in in…nite horizon settings.
22
accounting numbers using accounting relations. The following theorem states alternative
representations of the …rst type.
10
Theorem 2 The fundamental asset pricing relation (18) is equivalent to each of the follow-
ing:
(a) There exist forward probability measures Q
tt
for dates t = t + 1. .... 1 such that
\
)t
=
T

t=t+1
1
tt E
Q
tt
[d
)t
]. (19)
where 1
tt
is the price at date t of a zero-coupon bond paying one unit of account at
date t, and
E
Q
tt
[] is the conditional expectations operator under Q
tt
.
(b) There exist event-price de‡ators :
t
for dates t = t + 1. .... 1 such that
\
)t
=
T

t=t+1
Et
[:
t
d
)t
]. (20)
(c) There exist valuation indices ¡
tt
for dates t = t + 1. .... 1 such that
\
)t
=
T

t=t+1
1
tt
¦
Et
[d
)t
] + Cov
t
[d
)t
. ¡
tt
]¦ . (21)
where the valuation indices are given by the Radon-Nikodym derivative between the
probability measures Q
tt
and , i.e.,
¡
tt
=
dQ
tt
d

t

t
). (22)
In (a) the event-prices are normalized by the riskless discount factor, 1
tt
, for the period
from t to t (in order to transform the event-prices into probability densities), while in (b)
the event-prices are normalized by the conditional densities, d(¸
t

t
). In (c) the event-
prices are normalized by both the discount factor and the probabilities and, furthermore, it
follows that
Et

tt
] = 1. In the following, we use these equivalent formulations of asset prices
interchangeably.
10
For notational simplicity, we will in the following suppress the dependence of, for example, prices on
the speci…c events j
t
whenever no confusion should be possible.
23
3.2 Accounting Relations
As noted above, the set of alternative representations of no-arbitrage prices of equity claims
of the second type builds on a set of accounting relations. For notational simplicity, we focus
on a particular …rm and, therefore, we drop the subscript , on asset prices in the remainder
of the paper.
Following Feltham and Ohlson (1999) we assume the following accounting articulation
relations of future income statements and balance sheets t = t+1. .... 1 (see also Christensen
and Feltham, 2003, Chapter 9, for further discussion of these relations).
(A) Clean Surplus Relation (CSR):
All changes in the book value of equity except transactions with common equityholders
go through the income statement, i.e.,

t
= /·
t1
+ :i
t
÷d
t
+ o
t
. /·
T
= 0.
where /·
t
is the book value of equity at date t, :i
t
is net income in period t, d
t
is
the net-dividend paid to equityholders prior to closing the books at date t, and o
t
is
the market value of issued equity at date t through anticipated equity transactions.
Furthermore, the book value of equity and net income are separated into …nancial and
operating activities, i.e.,

t
= ,c
t
+ oc
t
.
:i
t
= ,i
t
+ oi
t
.
(B) Financial Asset Relation (FAR):
All transfers to common equityholders are made through the …nancial assets, and these
assets are increased by …nancial income and free cash ‡ows denoted ,c,
t
, i.e.,
,c
t
= ,c
t1
+ ,i
t
+ ,c,
t
÷d
t
+ o
t
.
(C) Financial Assets Marked-to-Market (FAM):
The risk-adjusted expected …nancial income equals the riskless spot interest rate (de-
noted t
t1
) times the opening book value of the …nancial assets, i.e.,
E
Q
t,t1
[,i
t
] = t
t1
,c
t1
.
24
(D) Operating Asset Relation (OAR):
The operating assets are increased by operating income and reduced by the free cash
‡ows transferred to the …nancial assets, i.e.,
oc
t
= oc
t1
+ oi
t
÷,c,
t
.
(A) is the well known clean surplus relation, which states that the net income concept is
comprehensive income. Note that it must hold for all future dates, while it is not important
that past …nancial statements satisfy this relation. Of course, if past reformulated …nancial
statements are used to forecast future statements, then consistency would imply that these
past statements are also on a comprehensive income basis. If there are anticipated equity
transactions (such as outstanding executive stock options), then the accounting must be
“super-clean,” since our objective is to value common equity, i.e., the equity concept is
“pure-equity” (see Christensen and Feltham, 2003, Section 9.4).
The assets are divided into operating assets and …nancial assets. Financial assets (and
the associated …nancial income) are those assets, which are marked-to-market on the balance
sheet and earn a competitive risk-adjusted expected return as implied by (C). Assets not
satisfying these conditions are classi…ed as operating assets.
We are not going to invoke the Modigliani-Miller assumptions of capital structure irrel-
evance. The issue of how to consistently account for corporate tax savings of …nancial debt
is central to practical equity valuation. Note that we are using the same set of risk-adjusted
probabilities Q
tt
to price the before-investor-tax dividends on all claims including both …nan-
cial and equity claims. Hence, there is an implicit assumption here that all claims are taxed
equally in the hands of investors, so that the relative prices can be determined on a before
tax basis (as it is common in standard approaches to equity valuation).
11
Therefore, …nancial
income is income from the …nancial assets before corporate taxes, implying that corporate
taxes (or corporate tax savings) on …nancial income (or net …nancial expenses) are part of
operating income. Furthermore, any recognition of deferred taxes (or deferred tax bene…ts)
is classi…ed as an operating liability (asset). This is di¤erent from the WACC-approach in
which tax shields of interest payments on net debt are part of …nancial income. In that
approach, the tax bene…ts of debt are re‡ected in a lower required rate of return used to
discount the free cash ‡ows from operations or residual operating income from operations—
a somewhat strange idea. Our approach is similar to the so-called adjusted-present-value
11
If di¤erential personal taxation on equities and bonds is viewed as an important institutional feature, a
quick …x is to mark down the zero-coupon interest rates used in (21) by the relative personal tax disadvantage
on equities compared to bonds.
25
(APV) technique, in which the tax advantage to debt is valued separately as an operating
asset potentially generating value (see Section 2.2).
The free cash ‡ow ,c,
t
is like a dividend from the operating assets to the …nancial assets,
cf. (D) and (B). Note that with our classi…cation of all taxes as part of the operations, the
free cash ‡ow is equal to operating cash minus cash investments in operating assets (as usual)
minus total paid taxes, and it is equal to the cash ‡ow distributed to the …nancial assets net
of payments to common equity holders.
In addition to these accounting relations, we de…ne spot residual income for the …nancial
and operating activities as income minus the riskless spot interest rate times the opening
book value, i.e.,
:i
c
t
= :i
t
÷t
t1

t1
.
:,i
c
t
= ,i
t
÷t
t1
,c
t1
.
:oi
c
t
= oi
t
÷t
t1
oc
t1
.
Note the di¤erence here compared to the standard approach of using risk-adjusted capital
charges for calculating residual income: (a) an equity cost of capital is used for residual
income; (b) an after-tax cost of capital for residual …nancial income, and (c) a weighted
average cost of capital for residual operating income. Using these approaches, additivity
of these concepts in the sense that :i
t
= :,i
t
+ :oi
t
requires that all assets including the
operating assets are marked-to-market (which is obviously not the case). In our de…nition of
the three residual income measures, the capital charge on opening book values of all assets
is the beginning-of-period before-tax riskless spot interest rate. Furthermore, note that by
the FAM relation, the risk-adjusted expected residual …nancial income is equal to zero.
The abnormal operating income growth model is the latest addition to the arsenal of
accounting-based equity valuation models. While the residual income models essentially
work with the price-to-book ratio, the abnormal income growth models work with another
common multiple, namely the price-earnings ratio. Let the spot yield-spread between the
spot riskless interest rate for some future period t and the current spot riskless interest rate,
o
t1,t
, be de…ned as

c
t1,t
= 1 +o
c
t1,t
=
1 +t
t1
1 +t
t
- 1 +t
t1
÷t
t
.
De…ne spot abnormal operating income growth in period t as
coiq
c
tt
= :oi
c
t
÷
c
t1,t
:oi
c
t1
.
26
This is equivalent to the usual de…nition of abnormal operating income growth except for
the spot yield-spread (and the fact that residual operating income is based on spot interest
rates). Hence, the spot abnormal operating income growth in some future period t is positive
if, and only if, the growth in spot residual operating income exceeds the yield spread (for
:oi
c
t1
0):
coiq
c
t

t
) 0 =
:oi
c
t
:oi
c
t1

c
t1,t
.
The following theorem establishes the relations between the current market value of common
equity o
t
= \
t
+,c
t
, future free cash ‡ows, and future accounting numbers (see also Feltham
and Ohlson (1999) and Christensen and Feltham, 2003, Propositions 9.3 and 9.4, and Ohlson
and Juettner-Nauroth, 2005).
12
Theorem 3 No-arbitrage, CSR, FAR, FAM, and OAR are su¢cient for the following free
operating cash ‡ow and spot accounting-value relations
\
t
=
T

t=t+1
1
tt E
Q
tt
[,c,
t
] =
T

t=t+1
1
tt
[
Et
[,c,
t
] + Cov
t
[,c,
t
. ¡
tt
]] (23)
= oc
t
+
T

t=t+1
1
tt E
Q
tt
[:oi
c
t
] = oc
t
+
T

t=t+1
1
tt
[
Et
[:oi
c
t
] + Cov
t
[:oi
c
t
. ¡
tt
]] (24)
=
1
t
t
_
E
Q
t+1,t
[oi
t+1
] +
T

t=t+2
1
t
1
tt E
Q
tt
[coiq
tt
] ÷1
Tt

t
)
E
Q
Tt
[:oi
T
]
_
. (25)
where the risk-adjusted probabilities and the valuation indices are the same as in the dividend-
value relations in Theorem 2.
Given mark-to-market accounting of the …nancial assets, the task in equity valuation is the
determination of the value of operations, which is the left-hand side of the value relations in
Theorem 3. These value relations appear like the similar relations in the standard approaches
calculation net present values by discounting expected cash ‡ows or accounting numbers by
risk-adjusted discount rates. The key di¤erences are that the calculation of capital charges
and discounting (and the capitalization of the net present value in (25)) are performed using
before-tax spot and zero-coupon interest rates, and that expectations are calculated using
the forward risk-adjusted probabilities. Using the de…nition of the valuation indices, the
risk-adjusted expectations can also be represented as the non-adjusted expectations plus the
covariances with the valuation indices. That is, risk-adjustments are made in the numerator
as opposed to in the discount rates in the denominator.
12
All proofs are given in Appendix A.
27
3.3 Alternative Formulations of Accounting Value Relations
In Theorem 3, the discount factors 1
tt
, t = t + 1. .... 1, at the valuation date t are the
prices of zero-coupon bonds at the valuation date. These are easily estimated using current
prices of traded treasury bonds and your favorite term structure model.
13
The book values
of the …nancial and operating assets are also readily observable from (reformulated) …nancial
statements. Expected future free cash ‡ows are outcomes of a strategic and …nancial state-
ment analysis of the …rm. However, in the calculation of expected residual operating income
or expected abnormal operating income growth, we need estimates of both the expected
future spot interest rates, and the covariance between spot interest rates and book values of
operating assets or operating income.
14
These terms are not usually estimated in standard
security analyses and, moreover, the covariances with the future spot interest rates may be
a¤ected by the choice of accounting policy, for example, moving from historical cost towards
fair value accounting for some operating assets. Hence, these covariances may interact with
the risk-adjustment terms, such as Cov
t
[:oi
c
t
. ¡
tt
] . in subtle ways, which may “hide” the full
consequences of changes in accounting policy on risk-adjusted expected future accounting
numbers. Note, however, that future free cash ‡ows are independent of accounting poli-
cies. In this section, we rede…ne residual operating income and abnormal operating income
growth using current forward interest rates as opposed to future spot interest rates, and
derive the associated accounting value relations. These relations also highlight the fact that
it is the underlying free cash ‡ows and their time-series properties which are determining
the risk-adjustments.
The current forward interest rate ,:
t1,t
for period t is de…ned as the interest rate on
a contract for a single-period deposit from date t ÷1 to t, which implies that the contract
has a value of zero at the current date t, i.e.,
0 = 1
t1,t
÷11
t1,t
1
tt
. 11
t1,t
= 1 +,:
t1,t
.
Hence, the current forward interest rates are readily observable from the current zero-coupon
prices, i.e.,
11
t1,t
=
1
t1,t
1
tt
. t = t + 1. .... 1. (26)
13
Note that, in principle, we do not need a fully dynamic model of the stochastic process for interest
rates—we only need to calibrate to the current term structure of interest rates.
14
For example,
Et
[roi
s

] = Et
[oi

] ÷Et
[i
1
oa
1
]
= Et
[oi

] ÷[Et
[i
1
] Et
[oa
1
] + Cov
t
[i
1
, oa
1
]] .
28
Note that forward interest rates are, in general, not equal to expected future spot interest
rates (unless, of course, the latter are deterministic). Now, rede…ne forward residual operating
income and forward abnormal operating income growth to be
:oi
)
tt
= oi
t
÷,:
t1,t
oc
t1
.
and
coiq
)
tt
= :oi
)
tt
÷
)
t1,t
:oi
)
t1,t
.
respectively, where the current forward yield-spread is de…ned as

)
t1,t
= 1 +o
)
t1,t
=
1 +,:
t1,t
1 +t
t
- 1 +,:
t1,t
÷t
t
.
We can now restate the accounting value relations in Theorem 3 as follows.
Theorem 4 No-arbitrage, CSR, FAR, FAM, and OAR are su¢cient for the following for-
ward accounting-value relations
\
t
= oc
t
+
T

t=t+1
1
tt
_
Et
[:oi
)
t
] + Cov
t
[,c,
t
. ¡
t
]
¸
(27)
=
1
t
t
_
Et
[oi
t+1
] +
T

t=t+2
1
t+1,t
1
tt Et
[coiq
)
tt
] ÷1
Tt Et
[:oi
)
Tt
]
_
+
T

t=t+1
1
tt
Cov
t
[,c,
t
. ¡
t
]. (28)
One advantage of these forward accounting-value relations from an implementability point of
view, is that they only require an estimation of the current term structure of interest rates,
i.e., current zero-coupon bond prices from which the forward rates can be derived using
(26), forecasts of operating income and operating assets, and then, of course, the net present
value of the risk-adjustments to future free cash ‡ows. That is, there is no need to forecast
future spot interest rates and how they are correlated with future operating income and
operating assets. From a theoretical point of view, the forward accounting-value relations
also highlight the fact that accounting policy choice has no impact on risk premia (given that
the valuation indices are una¤ected), since the future free cash ‡ows are independent of the
accounting policy choice. Hence, a future change in accounting policy to earlier recognition
of information about future cash ‡ows (with, for example, fair value accounting) has no
impact on risk premia (nor on the ex-ante value of the …rm),
15
see also Christensen, de la
15
Of course, this is a somewhat trivial result, since the ex-ante …rm value can also be determined as the
29
Rosa, and Feltham (2008).
Adisadvantage of the forward accounting-value relations is that they “mix” future accrual
accounting numbers with free cash ‡ows. Of course, the operating asset relation (OAR) can
be used to substitute out the free cash ‡ows. For example, the forward residual operating
income relation (27) can be restated as follows
\
t
= oc
t
+
T

t=t+1
1
tt
_
Et
[:oi
)
t
] + Cov
t
[oi
t
÷(oc
t
÷oc
t1
) . ¡
t
]
¸
= oc
t
+
T

t=t+1
1
tt
_
Et
[:oi
)
t
] + Cov
t
[oi
t
÷,:
t1t
oc
t1
÷oc
t
+ 11
t1,t
oc
t1
. ¡
t
]
¸
= oc
t
+
T

t=t+1
1
tt
_
Et
[:oi
)
t
] + Cov
t
[:oi
)
t
. ¡
t
] ÷Cov
t
[cocq
)
tt
. ¡
t
]
_
. (29)
where forward abnormal operating asset growth is de…ned as
cocq
)
tt
=
_
oc
t
oc
t1
÷11
t1,t
_
oc
t1
.
Hence, compared to the spot residual operating income relation (24), there is an additional
risk-adjustment component in (29) if forward residual operating income is also used for risk
adjustments.
Since we (in the current version of the paper) do not investigate the impact of changes in
accounting policies, we will in the following (also for notational ease) perform our analysis
on the basis of the spot accounting-value relations and, in particular, on the spot residual
operating income relation (24). Similar analyses can be performed using any of the other
dividend-, free cash ‡ow, and abnormal income growth relations.
3.4 Equilibrium Valuation Indices
The remaining item in the value relations is the risk-adjustment due to the covariance with
the valuation index. Unfortunately, no-arbitrage alone does not tell us much about the
valuation indices. All we know is that they are positive and have expected values of one—at
this point they are merely mathematical constructs without economic content (except they
are derived from an assumption of no-arbitrage).
There are various ways of getting more information about the valuation indices. We will
assume that equilibrium prices are formed in an e¤ectively dynamically complete market.
net present value of future free cash ‡ows.
30
By this we mean that there are su¢cient trading opportunities for the investors to trade to
a Pareto e¢cient risk sharing. Moreover, we will assume investors have time-additive utility
functions and homogeneous beliefs. This leads to the following result (see, for example,
Christensen and Feltham, 2003, Chapter 6, and the references therein).
Theorem 5 Let an equilibrium in an e¤ectively dynamically complete market be given. As-
sume investors have homogeneous beliefs, and di¤erentiable, strictly increasing, and con-
cave, time-additive utility functions of event-contingent consumption, n
it
(c
it
), de…ned on
C
i
= [c
i
. ·), where n
0
it
(c
it
) ÷·, for c
it
| c
i
.
(a) Individual equilibrium consumption plans are measurable with respect to aggregate con-
sumption, r
t
, at each date, i.e., c
it
= c
it
(r
t
), and c
it
(r
t
) is an increasing function of
r
t
, i = 1. .... 1.
(b) The prices of zero-coupon bonds paying one unit of account at date t are given by
1
tt

t
) =
E
[n
0
it
(c
it
)[¸
t
]
n
0
it
(c
it

t
))
. t = t + 1. .... 1; i = 1. .... 1. (30)
(c) The valuation indices based on zero-coupon bonds are measurable with respect to aggre-
gate consumption at each date and given by the“scarcity” of aggregate consumption as
measured by the investors’ marginal utility of consumption, i.e.,
¡
tt
(r
t

t
) =
n
0
it
(c
it
(r
t
))
E
[n
0
it
(c
it
)[¸
t
]
. t = t + 1. .... 1; i = 1. .... 1. (31)
This theorem is the key result in consumption-based asset pricing models. The …rst-order
condition for the investors’ decision problems (with time-additive utility) implies that each
investor’s valuation index is determined by the investor’s marginal utility of consumption.
The additional assumptions of homogeneous beliefs and e¢cient risk sharing imply that all
investors’ valuation indices are perfectly aligned and measurable with respect to aggregate
consumption. Hence, the risk-adjustments in the preceding value relations relate to the
covariance of dividends, free cash ‡ows, residual operating income, or abnormal operating
income growth with the “scarcity” of aggregate consumption. But how can we empirically
measure the “scarcity” of aggregate consumption? One approach is to make an ad hoc
assumption that the valuation index is given by a linear factor structure using, for example,
the Fama-French portfolios as factors (see Nekrasov and Shro¤, 2006). We are not going to
take that route, but rather use a general equilibrium approach.
31
Before closing this section, we should note that the zero-coupon bond prices in (30) and
the valuation indices in (31) are given in “real terms,” i.e., the units of accounts are units
of the consumption good. If in‡ation is stochastic, this is a problem empirically—…rstly,
because the set of traded real bonds is very limited and, thus, it may not be possible to
estimate the real term structure of interest rates reliably from traded bonds and, secondly,
because, in practice, forecasts of future accounting numbers and free cash ‡ows are in nominal
terms. Moreover, some stocks may be a better hedge against in‡ation than others. In the
current version of the paper, we do not develop a full analysis of the impact of uncertain
in‡ation. However, the following corollary indicates the key ingredients of an analysis in
which the prices of nominal zero-coupon bonds are used as the numeraire (instead of the
real prices of real bonds). Let j
t
denote the price level of the consumption good at date
t. t = t. t + 1. .... 1, such that the real prices of securities and the real dividends are given
as \
.
t
,j
t
and d
.
t
,j
t
, respectively.
16
Corollary 6 Let an equilibrium in an e¤ectively dynamically complete market be given (as
in Theorem 5).
(a) Individual equilibrium consumption plans are measurable with respect to real aggregate
consumption, r
t
, at each date, i.e., c
it
= c
it
(r
t
), and c
it
(r
t
) is an increasing function
of r
t
, i = 1. .... 1.
(b) The nominal prices of zero-coupon bonds paying one unit currency at date t are given
by
1
.
tt

t
) =
E
[n
0
it
(c
it
),j
t

t
]
n
0
it
(c
it

t
)),j
t
. t = t + 1. .... 1; i = 1. .... 1.
(c) The nominal valuation indices based on nominal zero-coupon bonds are measurable with
respect to the pair of real aggregate consumption and the price level at each date, i.e.,
¡
.
tt
(r
t
. j
t

t
) =
n
0
it
(c
it
(r
t
)),j
t
E
[n
0
it
(c
it
),j
t

t
]
. t = t + 1. .... 1; i = 1. .... 1.
(d) The nominal prices of securities are given as
\
.
t
=
T

t=t+1
1
.
tt
_
Et
[d
.
t
] + Cov
t
[d
.
t
. ¡
.
tt
]
_
.
16
We present this result without proof—it essentially follows from the investors’ …rst-order conditions
recognizing that their preferences are de…ned over units of real consumption.
32
Of course, the presence of uncertain in‡ation does not a¤ect the allocation of real aggregate
consumption in an e¤ectively dynamically complete market, which is (a). However, more
securities may be needed to ensure that investors can achieve a Pareto e¢cient allocation—
they need to be able to insure in‡ation risk. If the price level is deterministic, then
1
.
tt

t
) = 1
tt
j
t
j
t
.
which is the standard “Fisher relation” for the relationship between nominal and real interest
rates. One the other hand, if in‡ation is stochastic, then the prices of nominal bonds will
not only re‡ect the expected in‡ation but also include an in‡ation risk premium due to
the covariance between the marginal utility of real consumption and the purchasing power
of one unit of nominal currency, i.e., 1,j
t
. Similarly, the nominal valuation index used to
calculate the nominal risk premia for risky nominal securities not only re‡ects the scarcity
of real consumption but also the purchasing power of nominal dividends. Hence, there are
two sources of the risk-adjustment to expected nominal dividends: the covariance with the
scarcity of real consumption, and a term which re‡ects the nominal dividends’ ability to
hedge in‡ation risk.
Of course, using the accounting relations, the nominal dividend value relation can be
re-expressed to obtain the accounting value relations as above. Note that the no-arbitrage
value relations shown in the preceding sections do not rely on which unit of account we are
using—it is only when we wish to identify the equilibrium valuation index in terms of real
consumption that we need to be speci…c about the units of account. As noted above, we will
not pursue these issues much further in this version of the paper. However, we will include
a few footnotes describing the necessary adjustments to the analysis in order to account for
stochastic in‡ation.
4 An Accounting-based Multi-period Equity Valuation
Model
Identi…cation of the valuation index requires more assumptions on distributions and (or)
preferences. Standard distributions in asset pricing models are normal distributions and
lognormal distributions. First, we need a mathematical result for normal distributions known
as Stein’s Lemma (see Rubinstein, 1976, for a proof).
Lemma 7 If A and 1 are jointly normally distributed random variables and 1() is some
33
di¤erentiable real valued function with
E
[[1
0
(1 )[] < ·, then
Cov(A. 1(1 )) =
E
[1
0
(1 )] Cov[A. 1 ].
If we assume that future spot residual operating income and aggregate consumption are
jointly normally distributed, then Stein’s Lemma and the accounting-value relation (24)
imply that
17,18
\
t
= oc
t
+
T

t=t+1
1
tt
_
Et
[:oi
c
t
] +
Et

0
tt
(r
t
)] Cov
t
[:oi
c
t
. r
t
]
_
(32)
Recognizing that the valuation indices in the dividend-value relation (21) and the accounting-
value relation (24) are the same, and that a claim on aggregate date t consumption can be
priced using (21), we get that the price of this claim is
\
a
tt
= 1
tt
_
Et
[r
t
] +
Et

0
tt
(r
t
)] Cov
t
[r
t
. r
t
]
_
=
Et

0
tt
(r
t
)] =
\
a
tt
1
tt
÷
Et
[r
t
]
Var
t
[r
t
]
.
where 1
tt
= 1
1
tt
is the riskless return from date t to t. If we de…ne aggregate consumption
returns and expected aggregate consumption returns as
1
a
tt
=
r
t
\
a
tt
. 1
a
tt
=
Et
[r
t
]
\
a
tt
.
17
Normally distributed aggregate consumption does not go well with standard utility functions de…ned
only on positive consumption—really, only the negative exponential utility function works (see the following
section). Instead we could assume that aggregate consumption is lognormally distributed and write the
risk-adjustment in (24) as
Cov
t
[roi

, ¡
t
(r

)] = Cov
t
[roi

, /
t
(.

)],
where /
t
(.

) = ¡
t
(exp(.

)) and .

= ln(r

), and then proceed in a similar manner from there. In this
case, most …nance researchers’ favorite utility function, the power utility function, would be the choice
for similar general equilibrium analyses as in the subsequent sections. For simplicity, we stick with the
normal distribution in the text. The general equilibrium analysis with lognormally distributed aggregate
consumption and power utilities is worked out in Appendix B.
Note that a lognormal distribution would not be a good choice for future residual operating income, since
these income numbers may very well be negative (which is precluded by the lognormal distribution).
18
Here we explicitly use that the valuation index is a function of a normally distributed random variable
namely aggregate consumption. If in‡ation is stochastic, then the nominal valuation index is only measurable
with respect to the pair of aggregate real consumption and the price level. Hence, in a nominal analysis,
we must make the additional assumption that the nominal valuation index aggregates these two objects
into a normally distributed random variable. This requires assumptions about not only the distributions
of aggregate consumption and the price level, but also about the utility function. We will provide such
assumptions in the nex section.
34
then we can rewrite (32) as
\
t
= oc
t

t
) +
T

t=t+1
1
tt
_
Et
[:oi
c
t
] ÷[
Et
[1
a
tt
] ÷1
tt
]
Cov
t
[:oi
c
t
. 1
a
tt
]
Var
t
[1
a
tt
]
_
(33)
Similarly, we can de…ne residual income returns on net operating assets by normalizing future
spot residual operating income by the current book value of operating assets (like in “trend
analyses”), i.e.,
1c`C¹
c
tt
=
:oi
c
t
oc
t
. 1c`C¹
c
tt
=
Et
[:oi
c
t
]
oc
t
.
Dividing by the current book value of operating assets in (33) yields the following result.
Proposition 8 Assume the conditions in Theorem 5 hold and make the accounting assump-
tions (A)-(D). Furthermore, assume that future spot residual operating income and aggregate
consumption are jointly normally distributed, and that
Et
[[¡
0
tt
(r
t
)[] < ·. Then the market-
to-book ratio for the operating assets is given by
\
t
oc
t
= 1 +
T

t=t+1
1
tt
_
1c`C¹
c
tt
÷[1
a
tt
÷1
tt
]
Cov
t
[1c`C¹
c
tt
. 1
a
tt
]
Var
t
[1
a
tt
]
_
. (34)
There are two unusual objects in this value relation. One is the term structure of excess
returns at date t, i.e.,
1
a
tt
÷1
tt
. t = t + 1. .... 1.
Of course, these are easily determined if one assumes that the expected return on a well
diversi…ed stock index is a good proxy for expected aggregate consumption returns, and
that these excess returns are the same for all maturities (as in applications of the standard
CAPM). However, there is signi…cant empirical evidence that these excess returns are time-
varying and predictable to some extent (see, for example, Campbell and Shiller, 1988a and
1988b, and Lettau and Ludvigson, 2001).
The second unusual object is the term structure of systematic accounting risk, i.e.,
Cov
t
[1c`C¹
c
tt
. 1
a
tt
]
Var
t
[1
a
tt
]
. t = t + 1. .... 1.
Except for the term structure aspect it is interesting that we have “accounting betas.” In the
early seventies of the previous century there was a literature examining accounting betas (see,
for example, Beaver, Kettler and Scholes, 1970).
19
The advantage of this approach is that we
19
In that literature, aggregate earnings on stock indices are used while we are using aggregate consumption
returns de…ned by market values of aggregate consumption claims.
35
avoid the circularity in the estimation of betas using stock returns. However, the drawback is
that we have much less data available to estimate the accounting betas. A possible solution
might be to estimate “industry betas” and then use those. Note that the accounting betas
we need are a¤ected neither by leverage (which is one of the curses of tests of asset pricing
models) nor by scale. The other interesting aspect is the term structure aspect. Here it
may be useful to assume a particular stochastic process for spot residual operating income
returns and aggregate consumption returns and then derive the term structure from there
(see below).
Both of these unusual objects in equity valuation require new empirical work (which is
not a comparative advantage of the authors’). It is an empirical issue whether Proposition
8 is a superior equity valuation model compared to a standard valuation approach using
risk-adjusted discount rates. In any case, Proposition 8 tells us what are the elements we
should be looking for in order to have an equity valuation model consistent with classical
multi-period asset pricing results.
5 An Accounting-based Multi-period Equity Valuation
Model with Exponential Utility
In this section we make additional assumptions on investor preferences in order to get more
insight into the determination of the term structure of excess returns, accounting betas,
and interest rates. Hakansson (1970) and others (see also Christensen and Feltham, 2003,
Proposition 6.6) show the following result.
Theorem 9 Let a Pareto e¢cient equilibrium be given and assume the investors have ho-
mogeneous beliefs and time-additive preferences represented by HARA utility functions with
identical risk cautiousness. Then there are parameters ·
it
and ,
it
such that
c
it
(r
t
) = ,
it
+ ·
it
r
t
. i = 1. .... 1.

1
i=1
·
it
= 1. and

1
i=1
,
it
= 0
Each investor’s fraction of aggregate consumption ·
it
is the same for all dates, i.e., ·
it
= ·
i
,
36
if, and only if, one of the following conditions holds for all i = 1. .... 1 and for all t = 1. .... 1:
n
it
(c
it
) = ÷,
1
it
exp[÷c
it
,j
i
]. ,
1
it
0. j
i
0;
n
it
(c
it
) = ,
1
t
ln(c
it
÷/
it
). ,
1
t
0. c
it
÷/
it
0;
n
it
(c
it
) = ,
1
t
1
c ÷1
[cc
it
÷/
it
]
1

. ,
1
t
0. cc
it
÷/
it
0.
The advantage of having constant fractions of aggregate consumption is that it is very easy
to ensure an e¢cient equilibrium—essentially, a market portfolio and a complete set of zero-
coupon bonds are su¢cient if personal endowments are also spanned (see Christensen and
Feltham, 2003, Proposition 6.7)
Assume investors have negative exponential utility and, for simplicity, that their risk
aversion is time-independent with aggregate risk tolerance j
c
=

i
j
i
. Wilson (1968) shows
that in this case ·
it
= ·
i
= j
i
,j
c
(see also Christensen and Feltham, 2003, Proposition. 4.3).
It then follows from (31) that the valuation index is given by
20
¡
tt
(r
t

t
) =
exp[÷r
t
,j
c
]
Et
[exp[÷r
t
,j
c
]]
. ¸
t
_ ¸
t
. t = t + 1. .... 1.
Di¤erentiating the valuation index with respect to aggregate consumption at date t and
taking conditional expectations yield
Et
_
¡
0
tt
(r
t
)
_
=
Et
_
÷
1
j
c
exp[÷r
t
,j
c
]
Et
[exp[÷r
t
,j
c
]]
_
= ÷
1
j
c
.
20
In a nominal analysis with stochastic in‡ation, the nominal valuation index is given in Corollary 6. If
we assume that the price level j

is determined as a geometric Brownian motion such that
j

= j
t
exp
__

t
j
p
d: +
_

t
o
p
d1
s
_
,
where j
p
and o
p
are the instantaneous expected in‡ation rate and in‡ation volatility, respectively, then the
price level at date t is log-normally distributed given information at date t. That is, ln[j

,j
t
] is normally
distributed with mean j
p
(t ÷t) and variance o
2
p
(t ÷t). Inserting this into the nominal valuation index,
using the exponential utility function, yields that
¡
N
t
(r

, j

[j
t
) =
exp[÷.

,j
o
]
Et
[exp[÷.

,j
o
]]
,
where
.

= r

+ j
o
ln[j

,j
t
] .
Hence, the analysis can proceed with the in‡ation-adjusted normally distributed random variable .

substi-
tuted for real aggregate consumption r

. Although this may seem a minor change in the analysis, it may have
a very substantial impact on empirical implementations of the model. It is well known that the stochastic
variation in real aggregate consumption is too small to …t observed risk premia in asset prices. Stochastic
in‡ation is an additional factor, which may help explaining observed nominal risk premia—standard asset
pricing tests implicitly assume deterministic in‡ation.
37
Inserting this into (32) yields
\
t
= oc
t

t
) +
T

t=t+1
1
tt
¦
Et
[:oi
c
t
] ÷Cov
t
[:oi
c
t
. r
t
,j
c

If we de…ne j as the investors’ average risk tolerance (i.e., j = j
c
,1) and “aggregate con-
sumption per capita” as ccc
t
= r
t
,1, then the risk-adjusted aggregate consumption per
capita is :ccc
t
= ccc
t
,j = r
t
,j
c
.
21
Hence, normalizing by the book value of operating
assets we get the following result.
Proposition 10 Assume the conditions in Theorem 5 hold and make the accounting as-
sumptions (A)-(D). Furthermore, assume that future spot residual operating income and
aggregate consumption are jointly normally distributed, and investors have exponential util-
ity with constant risk tolerances. Then the market-to-book ratio for the operating assets is
given by
\
t
oc
t
= 1 +
T

t=t+1
1
tt
_
1c`C¹
c
tt
÷Cov
t
[1c`C¹
c
tt
. :ccc
t
]
_
. (35)
and the zero-coupon prices are given by
1
tt
=
Et
[n
0
it
(c
it
)]
n
0
it
(c
it

t
))
= ,
c
tt
exp[÷¦
Et
[:ccc
t
] ÷:ccc
t
÷
1
2
Var
t
[:ccc
t
]¦]. (36)
where ,
c
tt
is the investors’ “average personal discount factor” from t to t.
22
In the introduction, we pointed to the fact that the equilibrium risk-adjusted expected cash
‡ows or accounting numbers as in (35) are not, in general, equal to the certainty equivalents
of the cash ‡ows or accounting numbers as certainty equivalents are normally de…ned in the
economics of uncertainty, i.e., the wealth level which makes the decision maker indi¤erent
between receiving this level of wealth with certainty and retaining the uncertain wealth.
First, it is clear from (35), and all other value relations preceding it, that it is not the risk
in the future cash ‡ows or accounting numbers of the individual securities which determines
their risk-adjusted expected cash ‡ows or accounting numbers, but rather how these risks
are correlated with risk-adjusted aggregate consumption per capita or, more generally, with
21
Note that risk-adjusted aggregate consumption per capital is equal to the average relative risk aversion
in the economy (which has an order of magniture of 2-5, empirically).
22
If we let the investors’ personal discount factors be de…ned as ,
P
it
= exp[÷0
i
t] , then the average personal
discount factor is given by
,
o
t
= exp[÷(t ÷t)

i
0
i
j
i
,j
o
] .
38
the valuation index. Of course, this is well known, but it might still be the case that the
equilibrium risk-adjusted expected cash ‡ows on optimally invested wealth would be equal
to the certainty equivalent of optimally invested wealth. The normally distributed aggregate
consumption and exponential utility assumptions make it possible to answer this question.
Since this question and its answer are bit of an aside, we refer the discussion to Appendix
D. However, under the assumptions of this section, we show in Appendix D that the risk-
adjustment in the equilibrium price of optimally invested wealth is twice as big as the risk
premium in the associated certainty equivalent.
6 A VAR Model with Exponential Utility
In order to get additional insights we now consider a simple vector-auto-regressive (VAR)
model of the stochastic properties of spot residual operating income for a particular …rm
and risk-adjusted aggregate consumption per capita. More comprehensive models can be
assumed, but we have chosen a simple one in order to more clearly focus on the central
issues. Assume the spot residual operating income returns and the risk-adjusted aggregate
consumption per capital are the only information available and that they follow a …rst-
order vector auto-regressive process with mean-reversion to a deterministic exponential trend,
i.e.,
23
1c`C¹
c
tt
÷1c`C¹
c
t
(1 +c)
tt
= .
v
[1c`C¹
c
t1,t
÷1c`C¹
c
t
(1 +c)
t1t
] +
t
. .
v
¸ [0. 1). (37a)
:ccc
t
÷:ccc
c
t
(1 +¸)
tt
= .
o
[:ccc
t1,t
÷:ccc
c
t
(1 +¸)
t1t
] + o
t
. .
o
¸ [0. 1). (37b)
where
t
and o
t
are zero-mean normally distributed and serially uncorrelated, and with
conditional variances and contemporaneous conditional covariance: Var
t
[
t
] = o
2
v
, Var
t
[o
t
] =
o
2
o
, Cov
t
[
t
. o
t
] = o
vo
. Solving these equations recursively yields
1c`C¹
c
tt
= 1c`C¹
c
t
(1 +c)
tt
+ .
tt
v
[1c`C¹
c
tt
÷1c`C¹
c
t
] +
t1t

c=0
.
c
v

tc
.
:ccc
t
= :ccc
c
t
(1 +¸)
tt
+ .
tt
o
[:ccc
t
÷:ccc
c
t
] +
t1t

c=0
.
c
o
o
tc
.
23
That is, the trends for the two processes are 1c·O¹
o
t
(1 +c)
t
and racc
o
t
(1 +¸)
t
with growth rate
c and ¸, respectively.
39
This speci…cation of the information dynamics allows us to calculate all the terms in the
exponential utility accounting-value relation (35) explicitly.
Proposition 11 The information dynamics in (37) imply that
1c`C¹
c
tt
= 1c`C¹
c
t
(1 +c)
tt
+ .
tt
v
[1c`C¹
c
tt
÷1c`C¹
c
t
].
Var
t
[1c`C¹
c
tt
] = o
2
v
1 ÷.
2[tt]
v
1 ÷.
2
v
.
Et
[:ccc
t
] = :ccc
c
t
(1 +¸)
tt
+ .
tt
o
[:ccc
t
÷:ccc
c
t
].
Var
t
[:ccc
t
] = o
2
o
1 ÷.
2[tt]
o
1 ÷.
2
o
.
and that the date t risk-adjustment in (35) is given by

c
tt
= Cov
t
[1c`C¹
c
tt
. :ccc
t
] = o
vo
1 ÷(.
o
.
v
)
tt
1 ÷.
o
.
v
. (38)
Note that the risk-adjustment, 1¹
c
tt
, is an increasing function of t (for .
o
.
v
0), but with
an upper limit of
o
vo
1 ÷(.
o
.
v
)
tt
1 ÷.
o
.
v
÷o
vo
1
1 ÷.
o
.
v
. (39)
Moreover, if spot residual operating income is serially uncorrelated, i.e., .
v
= 0, then the
risk-adjustment is due solely to the contemporaneous correlation with risk-adjusted aggregate
consumption, i.e., the risk-adjustment is o
vo
independently of maturity date t. The reason
for this latter result, of course, is that no new information about 1c`C¹
c
tt
is revealed
until date t—the claim to 1c`C¹
c
tt
is like a zero-coupon bond maturing at date t ÷ 1
plus a one-period claim from t ÷ 1 to t with risk o
vo
. The important lesson from (38) is
that, in general, the risk-adjustment depends not only on the contemporaneous conditional
covariance between 1c`C¹
c
tt
and risk-adjusted aggregate consumption, o
vo
, but also on
their time-series properties, i.e., .
v
and .
o
.
Note that the investors’ average risk tolerance is not directly empirically observable, and
it enters into the de…nition of risk-adjusted aggregate consumption per capita. There are
various ways of estimating the average risk tolerance. For example, it can be estimated based
on a time-series of average excess returns (probably giving rise to an equity premium puzzle
and many other empirical problems). The obvious alternative is to determine it implicitly
from a calibration to the readily observable current term structure of interest rates using
the zero-coupon prices in (36)—in a general equilibrium, the risk tolerances determining
zero-coupon prices are the same as those determining excess returns on equities! Once the
40
implicit risk tolerance is determined, the parameters in the process for risk-adjusted aggregate
consumption per capita can be estimated using a time-series of aggregate consumption per
capita. Again, these parameters may alternatively be backed out from the current term
structure of interest rates. We are not aware of empirical analyses determining the preference
parameters and the parameters in the stochastic process for aggregate consumption from an
observable term structure of interest rates. Of course, it is an empirical issue whether this
approach works better than the standard time-series approaches. In Appendix C we consider
a setting with preferences exhibiting habit formation. This allows for more ‡exibility in the
types of equilibrium term structures of interest rates.
7 Comparison to the Standard Textbook Approach
The standard textbook approach uses a constant risk-adjusted cost of capital (WACC) to
discount expected residual operating income, independently of the maturity. Of course, using
a constant risk-adjusted discount rate which does not re‡ect the current term structure of
interest rates is obviously a very bad choice. Therefore, in our comparison we will take the
liberty to interpret the textbook approach as just assuming a constant risk premium. Hence,
allowing for a general term structure of interest rates, the risk-adjusted discount factor has
the form
1
tt
= [1 + t
tt
+ :j]
(tt)
< 1
tt
= [1 + t
tt
]
(tt)
. for :j 0 and t t.
where t
tt
is the zero-coupon riskless interest rate for maturity t as of date t, and :j is a
constant risk premium (i.e., excess return times beta) typically estimated on the basis of
short-term stock returns. The present value at date t of the spot residual operating income
return on operating assets at date t then has the form 1
tt
1c`C¹
c
tt
. Hence, comparing
to (35), the comparable implicit risk-adjustment for maturity date t with this approach,

T1
tt
, is a solution to the equation
1
tt
1c`C¹
c
tt
= 1
tt
[1c`C¹
c
tt
÷1¹
T1
tt
].
Solving this equation for the implicit risk-adjustment yields

T1
tt
= 1c`C¹
c
tt
_
1 ÷
_
1 +t
tt
1 +t
tt
+ :j
_
tt
_
. (40)
41
Comparing the implicit risk-adjustment in (40) to the risk-adjustment of the exponential
utility model in (38) shows a number of striking di¤erences. First, note that the implicit
risk-adjustment 1¹
T1
tt
depends on the deterministic trend in spot residual operating income
returns (through 1c`C¹
c
tt
) and, obviously, it should not! For example, increasing (decreas-
ing) the growth rate in spot residual operating income returns c. increases (decreases) the
risk-adjustment and, thus, lowers (increases) the net present value of future residual operat-
ing income returns. Hence, high (low) growth …rms will tend to be undervalued (overvalued)
using the textbook approach. Note that our assumption, that the variance of future spot
residual operating income returns is independent of growth, is important for this relation.
Secondly, if there is low persistence in residual operating income returns, i.e., a high degree
of mean reversion, then the risk-adjustment in (38) is almost independent of maturity t.
i.e., close to being equal to the contemporaneous covariance with risk-adjusted aggregate
consumption, whereas the implicit risk-adjustment in (40) compounds the constant risk pre-
mium, :j. (t ÷t) times. Hence, low persistence …rms will tend to be undervalued using the
textbook approach (compare Brennan and Xia, 2006). Note also that even if there is high
persistence in residual operating income returns, the risk adjustment in (38) is bounded from
above as t goes to in…nity (see (39)), while the implicit risk-adjustment in (40) approaches
the expected spot residual operating income returns, 1c`C¹
c
tt
. These latter relations high-
light the fact that proper risk-adjustments should not only re‡ect the contemporaneous
covariance with risk-adjusted aggregate consumption, but also the time-series properties of
residual operating income returns and, in particular, the degree of persistence of these re-
turns. The key problem with the textbook approach is that it does not re‡ect when new
information about future residual operating income returns is revealed!
Will these di¤erences between the textbook approach and the exponential utility model
yield noticeable di¤erences in equity values, or are they merely second-order e¤ects? In
the following, we try to answer this question by means of using a numerical example with
reasonable parameters. The example is based on the set of parameters for risk-adjusted
aggregate consumption shown in Table 2 and a personal discount rate of 2%.
Current :ccc
t
Trend :ccc
c
t
Growth rate ¸ Persistence .
o
Variance o
2
o
2.25 2.00 2% 90% 1%
Table 2: Parameters of risk-adjusted aggregate consumption.
Using Propositions 10 and 11, these parameters result in the term structure of interest rates
shown in Figure 1.
We assume the …rm has a horizon of 25 years, and that the spot residual operating income
return process in our base case is characterized by the parameters shown in Table 3.
42
0%
2%
4%
6%
8%
0 5 10 15 20 25
Maturity
Figure 1: Term structure of interest rates.
Current 1c`C¹
c
t
Trend 1c`C¹
c
t
Growth rate c Persistence .
v
Covariance o
vo
15% 10% 0% 90% 1%
Table 3: Parameters of residual operating income return process for base case.
For the textbook approach we assume the …rm has a beta of one, and that the excess return
on the market portfolio is 5%. Using Propositions 10 and 11, this …rm has a market-to-book
of operating assets of 2.14, whether the textbook approach or the exponential utility model
is used. The risk-adjustments for both approaches are shown in Figure 2.
0%
2%
4%
6%
8%
0 5 10 15 20 25
Maturity
Textbook Model Exponential Model
Figure 2: Risk-adjustments in base case.
Note that even though the market-to-book values are the same (indeed, we have chosen
the parameters for this to be the case), the patterns across maturities are quite di¤erent.
In particular, in the textbook approach the implicit risk-adjustment is increasing without
bounds, while the risk-adjustment in the exponential model quite quickly approaches its
43
upper bound of (see (39))
o
vo
1
1 ÷.
o
.
v
= 0.01
1
1 ÷0.9 0.9
= 5.26%
despite the rather high persistence rates. Consider now two alternative cases: 5% growth,
and 5% decline in spot residual operating income returns, i.e., c = 5% and c = ÷5%,
respectively. The market-to-book ratios are 2,64 and 3.15 in the 5% growth case, and 1.86
and 1,63 in the 5% decline case for the textbook and exponential utility model, respectively.
The risk-adjustments are shown in Figures 3 and 4 for the c = 5% and c = ÷5% cases,
respectively.
0%
5%
10%
15%
20%
25%
0 5 10 15 20 25
Maturity
Textbook Model Exponential Model
Figure 3: Risk-adjustments in 5% growth case.
0%
2%
4%
6%
0 5 10 15 20 25
Maturity
Textbook Model Exponential Model
Figure 4: Risk-adjustments in 5% decline case.
As noted earlier the textbook approach leads to undervaluation of growth, while it over-
values declining expected spot residual operating income returns, and the di¤erences are far
44
from being trivial.
Finally, consider a case with low persistence in residual operating income returns (i.e.,
high mean reversion), .
v
= 10%. Compared to a market-to-book ratio of 2.14 for both
approaches in the base case, the market-to-book ratio now becomes 1.91 and 2.25 for the
textbook and exponential utility model, respectively. The risk-adjustments are shown in
Figure 5.
0%
2%
4%
6%
8%
0 5 10 15 20 25
Maturity
Textbook Model Exponential Model
Figure 5: Risk-adjustments in 10% persistence case.
Even though there is no growth, i.e., c = 0, the textbook approach risk-adjusts the
expected spot residual operating income returns far too much – due to the compounding
of the constant risk premium. The exponential utility model risk-adjusts with an almost
constant amount equal to the upper bound of
o
vo
1
1 ÷.
o
.
v
= 0.01
1
1 ÷0.9 0.1
= 1.10%.
Above we have asked the question of what the implicit risk-adjustment is in the textbook
approach assuming a constant risk premium. We can also ask the opposite question, i.e., what
is the risk premium for maturity t, :j
tt
, using a term structure of risk-adjusted discount
rates (like Ang and Liu, 2004, and Brennan and Xia, 2006) which makes this approach
equivalent to discounting risk-adjusted expected residual operating income returns with the
zero-coupon riskless interest rates. That is, :j
tt
is determined by the equation
1c`C¹
c
tt
(1 +t
tt
+ :j
tt
))
(tt)
=
_
1c`C¹
c
tt
÷1¹
c
tt
¸
(1 +t
tt
)
(tt)
(41)
Is there always a solution to this equation? Unfortunately, the answer is NO! To see why,
assume 1c`C¹
c
tt
0. and (t ÷t) is even. Then the left-hand side of (41) is positive for
45
any value of :j
tt
, i.e., the net present value of spot residual operating income returns at date
t is positive. However, the right-hand side is negative if 1¹
c
tt
1c`C¹
c
tt
. i.e., the net
present value of spot residual operating income returns at date t is negative. Unfortunately,
this latter condition is likely to be satis…ed for …rms in which future spot residual operating
income is expected to be small (i.e., low market-to-book …rms). Hence, we can only use a
term structure of risk-adjusted discount rates with great caution! In the following, assume

c
tt
< 1c`C¹
c
tt
such that there is a solution for :j
tt
. Solving (41) with respect to :j
tt
(using Propositions 10 and 11) yields
:j
tt
1 +t
tt
=
_
1c`C¹
c
tt
1c`C¹
c
tt
÷1¹
c
tt
_
1¸(tt)
÷1
=
_
1c`C¹
c
t
(1 +c)
tt
+ .
tt
v
[1c`C¹
c
tt
÷1c`C¹
c
t
]
1c`C¹
c
t
(1 +c)
tt
+ .
tt
v
[1c`C¹
c
tt
÷1c`C¹
c
t
] ÷o
vo
1(.a.r)
t
1.a.r
_
1¸(tt)
÷1
=
1
_
1 ÷
ora
1c.O¹
o
t
(1+c)
t
+.
t
r
[1c.O¹
s
tt
1c.O¹
o
t
]
1(.a.r)
t
1.a.r
_
1¸(tt)
÷1.
Note that the risk premiumis increasing in the contemporaneous covariance between 1c`C¹
c
tt
and :ccc
t
. while it is decreasing in the growth rate in residual operating income returns.
For the base case of the numerical example examined above the risk premium is shown
in Figure 6.
0%
2%
4%
6%
8%
10%
0 5 10 15 20 25
Maturity
Figure 6: Risk-premium in risk-adjusted discount rate for base case.
Hence, there must be a quite substantial reduction in the risk premium for larger matu-
rities to make this approach consistent with exponential utility model. Figure 7 shows the
risk premium for the other cases examined above.
46
0%
5%
10%
15%
20%
25%
0 5 10 15 20 25
Maturity
5% growth 5% decline 10% persistence
Figure 7: Risk premium in risk-adjusted discount rate for noted cases.
Note that the risk premium in the risk-adjusted discount rate is not de…ned for maturities
larger than 16 years for the case with declining residual operating income returns—for these
maturities 1¹
c
tt
1c`C¹
c
tt
and, thus, the risk premium is not de…ned.
References
[1] Ang, A., and J. Liu. (2004) “How to Discount Cash Flows with Time-varying Expected
Returns,” The Journal of Finance 59, 2745-2783.
[2] Beaver, W, P. Kettler, and M. Scholes. (1970) “The Association between Market
Determined and Accounting Determined Risk Measures,” The Accounting Review
45, 654-682.
[3] Brennan, M. J. (1997) “The TermStructure of Discount Rates,” Financial Management
26, 81-90.
[4] Brennan, M. J., and Y. Xia. (2006) “Risk and Valuation Under an Intertemporal
Capital Asset Pricing Model,” Journal of Business 79, 1-36.
[5] Caks, J. (1977) “The Coupon E¤ect on Yield to Maturity,” The Journal of Finance
32, 103-115.
[6] Campbell, J. Y., and J. H. Cochrane. (1999) “By Force of Habit: A Consumption-based
explanation of aggregate Stock Market Behavior,” Journal of Political Economy 107,
205-251.
[7] Campbell, J. Y., and R. J. Shiller. (1988a) “Stock Prices, Earnings, and Expected
Dividends,” The Journal of Finance 43, 661-676.
47
[8] Campbell, J. Y., and R. J. Shiller. (1988b) “The Dividend-price Ratio and Expectations
of Future Dividends and Discount Factors,” Review of Financial Studies 1, 195-228.
[9] Christensen, P. O., L. E. de la Rosa, and G. A. Feltham. (2008) “Information and the
Cost of Capital: An Ex-Ante Perspective,” Working Paper, University of Aarhus.
[10] Christensen, P. O., and G. A. Feltham. (2003) Economics of Accounting: Volume I –
Information in Markets, Springer Science+Business Media, Inc.
[11] Christensen, P. O., and J. Aa. Nielsen. (1986) “The Bond-type E¤ect on Yield to
Maturity,” The Scandinavian Journal of Economics 89, 193-208.
[12] Copeland, T., and J. Weston. (1988) Financial Theory and Corporate Policy, Addison-
Wesley.
[13] Cox, J., J. Ingersoll, and S. Ross. (1985) “A Theory of the Term Structure of Interest
Rates,” Econometrica 53, 385-467.
[14] Feltham. G. A., and J. A. Ohlson. (1999) “Residual Earnings Valuation With Risk
and Stochastic Interest Rates,” The Accounting Review 74, 165-183.
[15] Grinblatt, M., and S. Titman. (2002) Financial Markets and Corporate Strategy, second
edition, New York: McGraw-Hill/Irwin.
[16] Hakansson, N. H. (1970) “Optimal Investment and Consumption Strategies under Risk
for a Class of Utility Functions,” Econometrica 38, 587-607.
[17] Hughes, J., J. Liu, and J. Liu. (2008) “On the Relation Between Expected Returns and
Implied Cost of Capital,” Working Paper, UCLA.
[18] Kraus, A., and J. S. Sagi. (2006) “Asset Pricing with Unforeseen Contingencies,”
Journal of Financial Economics 82, 417-453.
[19] Lettau, M., and S. Ludvigson. (2001) “Consumption, Aggregate Wealth, and Expected
Returns,” The Journal of Finance 56, 815-849.
[20] Lundholm, R. J., and R. G. Sloan. (2004) Equity Valuation and Analysis, New York:
McGraw-Hill.
[21] Nekrasov, A., and P. K. Shro¤. (2006) “Fundamentals-Based Risk Measurement in
Valuation,” Working Paper, University of Minnesota.
[22] Ohlson, J. A., and B. Juettner-Nauroth. (2005) “Expected EPS and EPS Growth as
Determinants of Value,” Review of Accounting Studies 10, 349-365.
[23] Penman, S. H. (2007) Financial Statement Analysis and Security Valuation, third edi-
tion, New York: McGraw-Hill.
48
[24] Rubinstein, M. (1976) The Valuation of Uncertain Income Streams and the Pricing of
Options,” Bell Journal of Economics 7, 407-425.
[25] Wachter, J. A. (2006) “A Consumption-based Model of The Term Structure of Interest
Rates,” Journal of Financial Economics 79, 365-399.
[26] Wilson, R. (1968) “The Theory of Syndicates,” Econometrica 36, 119-132.
Appendix A: Proof of Theorems
In the proof of Theorem 3 the following lemma is useful. It follows from the de…nition of the
forward probability measures (see Christensen and Feltham, 2003, proof of Proposition 9.1).
Lemma 12 Let .
t1
be some quantity measurable with respect to date t ÷ 1 information.
Then
E
Q
tt
[1
t,t1
.
t1
] = 11
t1,t E
Q
t1,t
[r
t1
] .
where 11
t1,t
is one plus the current forward rate from t ÷1 to t. i.e.,
11
t1,t
=
1
t1,t
1
t,t
.
Proof of Theorem 3. The free cash ‡ow value relation (23) follows directly from Theorem
1, since the free cash ‡ows are equal to the dividends on a portfolio of consisting of the equity
and the …nancial assets.
In the proof of the spot residual operating income relation (24), we substitute out the
free cash ‡ows in (23) using the operating asset relation, i.e.,
\
t
=
T

t=t+1
1
tt E
Q
tt
[oi
t
÷(oc
t
÷oc
t1
)]
=
T

t=t+1
1
tt E
Q
tt
[oi
t
÷oc
t
] +
T

t=t+1
1
tt E
Q
tt
[oc
t1
] .
49
Using Lemma 12, we can rewrite the last term such that
\
t
=
T

t=t+1
1
tt E
Q
tt
[oi
t
÷oc
t
]
+
T

t=t+1
1
tt
_
11
t1,t E
Q
t1,t
[oc
t1
] ÷
E
Q
tt
[t
t1
oc
t1
]
_
=
T

t=t+1
1
tt E
Q
tt
[:oi
c
t
÷oc
t
] +
T

t=t+1
1
tt
11
t1,t E
Q
t1,t
[oc
t1
] .
Using the de…nition of forward rates and collecting terms, we get
\
t
=
T

t=t+1
1
tt E
Q
tt
[:oi
c
t
÷oc
t
] +
T

t=t+1
1
t1,t E
Q
t1,t
[oc
t1
]
=
T

t=t+1
1
tt E
Q
tt
[:oi
c
t
÷oc
t
] +
T1

t=t
1
tt E
Q
tt
[oc
t
]
= oc
t
+
T

t=t+1
1
tt E
Q
tt
[:oi
c
t
] + 1
Tt E
Q
Tt
[oc
T
]
= oc
t
+
T

t=t+1
1
tt E
Q
tt
[:oi
c
t
] .
where the last equality comes from the clean surplus relation, i.e., oc
T
= 0. This proves the
spot residual operating income relation (24).
50
The proof of the spot abnormal operating income growth relation (25) starts by rewriting
spot residual operating income relation (24),
\
t
= oc
t
+
T

t=t+1
1
tt E
Q
tt
[:oi
c
t
]
= oc
t
+ 1
t+1,t
_
E
Q
t+1,t
[oi
t+1
] ÷t
t
oc
t
_
+
T

t=t+2
1
tt E
Q
tt
[:oi
c
t
]
= 1
t+1,t
_
E
Q
t+1,t
[oi
t+1
] + oc
t

t
)
_
+
T

t=t+2
1
tt E
Q
tt
[:oi
t
]
=
1
t+1,t
t
t
_
E
Q
t+1,t
[oi
t+1
] + oc
t

t
)
_
[1
t+1,t
÷1] +
T

t=t+2
1
tt E
Q
tt
[:oi
c
t
]
=
1
t
t
_
E
Q
t+1,t
[oi
t+1
] ÷1
t+1,t
_
E
Q
t+1,t
[oi
t+1
] + oc
t

t
) ÷1
t+1,t
oc
t

t
)
__
+
T

t=t+2
1
tt E
Q
tt
[:oi
c
t
]
=
1
t
t
_
E
Q
t+1,t
[oi
t+1
] +
_
÷1
t+1,t E
Q
t+1,t
_
:oi
c
t+1
¸
+ t
t
T

t=t+2
1
tt E
Q
tt
[:oi
c
t
]
__
. (42)
The term in the inner bracket can be rewritten as
÷1
t+1,t E
Q
t+1,t
_
:oi
c
t+1
¸
+ t
t
T

t=t+2
1
tt E
Q
tt
[:oi
c
t
]
= ÷1
t+1,t E
Q
t+1,t
_
:oi
c
t+1
¸
+ [1
t+1,t
÷1]
T

t=t+2
1
tt E
Q
tt
[:oi
c
t
]
= 1
t+1,t
T

t=t+2
1
tt E
Q
tt
[:oi
c
t
] ÷
T

t=t+1
1
tt E
Q
tt
[:oi
c
t
]
= 1
t+1,t
T

t=t+2
1
tt E
Q
tt
[:oi
c
t
] ÷
T+1

t=t+2
1
t1,t E
Q
t1,t
_
:oi
c
t1
¸
=
T

t=t+2
_
1
t+1,t
1
tt E
Q
tt
[:oi
c
t
] ÷1
t1,t E
Q
t1,t
_
:oi
c
t1
¸
_
÷1
Tt E
Q
Tt
[:oi
c
T
]
=
T

t=t+2
1
t+1,t
1
tt
_
E
Q
tt
[:oi
c
t
] ÷
1
t1,t
1
t+1,t
1
tt
E
Q
t1,t
_
:oi
c
t1
¸
_
÷1
Tt E
Q
Tt
[:oi
c
T
] .
51
Using Lemma 12 it follows that
÷1
t+1,t E
Q
t+1,t
_
:oi
c
t+1
¸
+ t
t
T

t=t+2
1
tt E
Q
tt
[:oi
c
t
]
=
T

t=t+2
1
t+1,t
1
tt
_
E
Q
tt
[:oi
c
t
] ÷
E
Q
tt
_
1
t,t1
1
t+1,t
:oi
c
t1
__
÷1
Tt E
Q
Tt
[:oi
c
T
] .
Substituting into (42) yields
\
t
=
1
t
t
_
E
Q
t+1,t
[oi
t+1
] +
T

t=t+2
1
t+1,t
1
tt
_
E
Q
tt
[:oi
c
t
] ÷
E
Q
tt
_
1
t,t1
1
t+1,t
:oi
c
t1
__
÷1
Tt E
Q
Tt
[:oi
c
T
]
_
=
1
t
t
_
E
Q
t+1,t
[oi
t+1
] +
T

t=t+2
1
t+1,t
1
tt E
Q
tt
[coiq
c
t
] ÷1
Tt E
Q
Tt
[:oi
c
T
]
_
which is (25).
Proof of Theorem 4. It follows from the free cash ‡ow value relation (23) that
\
t
=
T

t=t+1
1
tt Et
[,c,
t
] +
T

t=t+1
1
tt
Cov
t
[,c,
t
. ¡
t
].
Using the operating asset relation and the de…nition of forward residual operating income,
the …rst term on the right-hand side can be rewritten such that
T

t=t+1
1
tt Et
[,c,
t
] =
T

t=t+1
1
tt Et
[oi
t
] ÷
T

t=t+1
1
tt Et
[oc
t
] +
T

t=t+1
1
tt Et
[oc
t1
]
=
T

t=t+1
1
tt Et
[oi
t
] ÷
T

t=t+1
1
tt Et
[oc
t
] +
T

t=t+1
1
tt Et
[(1 +,:
t1,t
) oc
t1
]
÷
T

t=t+1
1
tt Et
[,:
t1,t
oc
t1
].
52
Using the de…nition of forward residual operating income and the fact that 1
tt
(1 +,:
t1,t
) =
1
t1,t
, we get that
T

t=t+1
1
tt Et
[,c,
t
] =
T

t=t+1
1
tt Et
[:oi
)
tt
] ÷
T

t=t+1
1
tt Et
[oc
t
] +
T

t=t+1
1
t1,t Et
[oc
t1
]
=
T

t=t+1
1
tt Et
[:oi
)
tt
] ÷
T

t=t+1
1
tt Et
[oc
t
] +
T1

t=t
1
tt Et
[oc
t
]
= oc
t
+
T

t=t+1
1
tt Et
[:oi
)
tt
] ÷1
Tt Et
[oc
T
]
= oc
t
+
T

t=t+1
1
tt Et
[:oi
)
tt
].
where the last equality follows from the clean surplus relation, i.e., oc
T
= 0. This shows
(27).
The …rst two terms of the forward residual operating income value relation,
\
t
= oc
t
+
T

t=t+1
1
tt Et
[:oi
)
tt
] +
T

t=t+1
1
tt
Cov
t
[,c,
t
. ¡
t
].
can be rewritten as in (42) such that
oc
t
+
T

t=t+1
1
tt Et
[oi
t
÷,:
t1,t
oc
t1
]
=
1
t
t
_
Et
[oi
t+1
] +
_
÷1
t+1,t Et
[(oi
t+1
÷t
t
oc
t
)] + t
t
T

t=t+2
1
tt Et
[:oi
)
tt
]
__
. (43)
53
The term in the inner bracket can be rewritten such that
÷1
t+1,t Et
[(oi
t+1
÷t
t
oc
t
)] + t
t
T

t=t+2
1
tt Et
[:oi
)
tt
]
= ÷1
t+1,t Et
[(oi
t+1
÷t
t
oc
t
)] + (1
t+1,t
÷1)
T

t=t+2
1
tt Et
[:oi
)
tt
]
= 1
t+1,t
T

t=t+2
1
tt Et
[:oi
)
tt
] ÷
T

t=t+1
1
tt Et
[:oi
)
tt
]
= 1
t+1,t
T

t=t+2
1
tt Et
[:oi
)
tt
] ÷
T+1

t=t+2
1
t1,t Et
[:oi
)
t1,t
].
Using the fact that 1
tt
(1 +,:
t1,t
) = 1
t1,t
, we get that
÷1
t+1,t Et
[(oi
t+1
÷t
t
oc
t
)] + t
t
T

t=t+2
1
tt Et
[:oi
)
tt
]
= 1
t+1,t
T

t=t+2
1
tt Et
[:oi
)
tt
] ÷
T

t=t+2
1
tt
(1 +,:
t1,t
)
Et
[:oi
)
t1,t
] ÷1
Tt Et
[:oi
)
Tt
]
=
T

t=t+2
1
tt
_
1
t+1,t Et
[:oi
)
tt
] ÷(1 +,:
t1,t
)
Et
[:oi
)
t1,t
]
_
÷1
Tt Et
[:oi
)
Tt
]
=
T

t=t+2
1
t+1,t
1
tt
_
Et
[:oi
)
tt
] ÷
t1,t Et
[:oi
)
t1,t
]
_
÷1
Tt Et
[:oi
)
Tt
].
which when substituted into (43) yields that
oc
t
+
T

t=t+1
1
tt Et
[oi
t
÷,:
t1,t
oc
t1
]
=
1
t
t
_
Et
[oi
t+1
] +
T

t=t+2
1
t+1,t
1
tt Et
[coiq
)
tt
] ÷1
Tt Et
[:oi
)
Tt
]
_
This shows (28).
54
Appendix B: Power Utility and Log-normal Distribu-
tions
In this appendix we examine the case in which investors have power utilities. In lieu of
Theorem 9, we assume investors have the same risk cautiousness c (the slope of the linear
risk tolerance function), identical personal discount factors ,
1
t
, and, for simplicity, that their
personal minimum consumption levels /
it
are time-independent, i.e.,
n
it
(c
it
) = ,
1
t
1
c ÷1
[cc
it
÷/
i
]
1

. ,
1
t
0. cc
it
÷/
i
0.
The former two assumptions imply that the market is e¤ectively dynamically complete if
consumption endowments are spanned, and the investors can trade in a complete set of
zero-coupon bonds and the market portfolio. Note that the investors’ relative risk aversion
is increasing (decreasing) if /
it
< () 0, while the constant relative risk aversion (CRRA)
model has /
i
= 0. Let /
c
=

i
/
i
denote the aggregate minimum consumption level. In
this case, Wilson (1968) shows that e¢cient consumption plans are on the form (see also
Christensen and Feltham, 2003, Proposition 4.3)
c
it
(r
t
) = ,
i
+ ·
i
r
t
. i = 1. .... 1. where ,
i
= (/
i
÷·
i
/
c
),c.
Scaling by the number of investors and de…ning the average minimum consumption level by
/ = /
c
,1, the marginal utility of consumption for any investor i is
n
0
it
(c
it
) = ,
1
t

i
1)
1¸c
[cccc
t
+ /]
1¸c
.
Inserting this into (31) we get
¡
tt
(r
t

t
) =
[cccc
t
÷/]
1¸c
Et
[[cccc
t
÷/]
1¸c
]
. ¸
t
_ ¸
t
. t = t + 1. .... 1.
De…ne “log-aggregate consumption” per capita as |ccc
t
= ln(cccc
t
÷ /), and assume this
object is normally distributed. Then
/(|ccc
t

t
) = ¡
tt
(r
t

t
) =
exp[÷|ccc
t
,c]
Et
[exp[÷|ccc
t
,c]]
. ¸
t
_ ¸
t
. t = t + 1. .... 1.
55
Di¤erentiating the valuation index for log-aggregate consumption per capita and taking
expectations yields
Et
_
/
0
tt
(|ccc
t
)
_
= ÷
1
c
.
Inserting this into the accounting-value relation (32) and de…ning risk-adjusted aggregate
consumption per capita as :ccc
t
= |ccc
t
,c yields the following result.
Proposition 13 Assume the conditions in Theorem 5 hold and make the accounting as-
sumptions (A)-(D). Furthermore, assume that future spot residual operating income and
log-aggregate consumption per capita are jointly normally distributed, and investors have
power utilities with identical and constant risk cautiousness and personal discount factors,
and constant minimum consumption levels. Then the market-to-book ratio for the operating
assets is given by
\
t
oc
t
= 1 +
T

t=t+1
1
tt
_
1c`C¹
c
tt
÷Cov
t
[1c`C¹
c
tt
. :ccc
t
]
_
. (44)
and the zero-coupon prices are given by
1
tt
=
Et
[n
0
it
(c
it
)]
n
0
it
(c
it

t
))
= ,
1
tt
exp[÷¦
Et
[:ccc
t
] ÷:ccc
t
÷
1
2
Var
t
[:ccc
t
]¦]. (45)
where ,
1
tt
is the investors’ common personal discount factor from t to t.
Comparing to the exponential utility model in the text, note that the only di¤erence is
a re-de…nition of risk-adjusted aggregate consumption per capita: average risk tolerance
is substituted by the common risk cautiousness (= relative risk tolerance with / = 0), and
aggregate consumption per capital is substituted with log-aggregate consumption per capita.
Note that if / = 0 (i.e., the CRRA model), then |ccc
t
= ln(c) + ln(ccc
t
) and, consequently,
the VAR model for risk-adjusted aggregate consumption can be estimated directly based on
log-aggregate consumption per capita without any preference-dependent parameters like in
the exponential model (the risk cautiousness is just a time-independent mean adjustment).
However, if / ,= 0, then the VAR model for risk-adjusted aggregate consumption per capital
must be estimated simultaneously with a calibration to the term structure of interest rates
using (45).
56
Appendix C: Habit Formation
The analysis in Appendix B assumes that the investors’ minimum consumption levels are
constant across time. Allowing for time-dependent minimum consumption levels is straight-
forward. The next step is to allow for preferences exhibiting habit formation such that more
variability in the types of equilibrium term structures of interest rates can be obtained. The
literature distinguishes between internal and external habit formation. In the internal habit
formation models, the investor’s utility depends not only on his current consumption but
also on his past consumption (the habit) and, in turn, his current choice of consumption
a¤ects his future habit level of consumption. In these models, the investors’ utility functions
are no longer time-additive, and general equilibrium analyses of this type are very rare (if
they exist at all).
In the external habit formation models, the investor’s date t utility depends on his current
consumption as well as on some exogenous habit level of consumption, for example, on some
weighted average of past aggregate consumption per capita (keeping-up-with-the-Joneses).
That is, the marginal utility of consumption is increasing in the external habit level of
consumption. Although several models of this type assuming a representative agent economy
have been considered in the literature (see, for example, Wachter, 2006, and Campbell
and Cochrane, 1999), we have not been able to locate a general equilibrium analysis with
heterogeneous agents (see, however, Kraus and Sagi, 2006, for an extension to more general
event-contingent preferences). The issue is to establish aggregation, but as the following
analysis demonstrates, this can be established in a similar fashion as with deterministic
minimum consumption levels (see Christensen and Feltham, 2003, Chapters 4 and 6).
Consider date t utility functions of the type
n
it
(c
it
. ¸
t
) = ,
1
t
1
c ÷1
[cc
it
÷/
it

t
)]
1

. ,
1
t
0. cc
it
÷/
it

t
) 0. (46)
Note that this is the standard power utility function except that we allow the minimum con-
sumption level to be date- and event-contingent, i.e., the utility function is event-dependent.
In the following, we allow /
it

t
) to be some general date- and event-contingent habit for-
mation function. Of course, in an application of the model, we could let this function be
some weighted average of current and past aggregate consumption. First, we derive Pareto
e¢cient consumption plans (assuming homogeneous beliefs as in the preceding analysis) and,
secondly, we derive the equilibrium prices in an e¤ectively dynamically complete market.
Note that the investors’ utility functions are time-additive with external habit formation
(as opposed to internal habit formation). Hence, the Borch …rst-order conditions character-
57
izing necessary and su¢cient conditions for Pareto e¢cient risk sharing are
`
i
n
0
it
(c
it

t
). ¸
t
),(¸
t
) = j
t

t
). \¸
t
. t = 1. .... 1; i = 1. .... 1; (47)

1
i=1
c
it

t
) = r
t
. \¸
t
. t = 1. .... 1; (48)
where `
i
is the weight assigned to investor i, and j
t

t
) is the multiplier for the aggregate
consumption constraint in a central planner’s optimal risk sharing problem (see, for example,
Christensen and Feltham, 2003, Chapter 4). De…ning `
it
= `
i
,
1
t
and inserting (46) in (47)
yield
`
it
[cc
it

t
) ÷/
it

t
)]

1

=
j
t

t
)
,(¸
t
)
=cc
it

t
) ÷/
it

t
) = `
c
it
_
j
t

t
)
,(¸
t
)
_
c
. (49)
Summation across investors using (48) yields
cr
t
÷/
ct

t
) = `
c
ct
_
j
t

t
)
,(¸
t
)
_
c
.
where /
ct

t
) =

i
/
it

t
) and `
c
ct
=

i
`
c
it
. Substitution back into (49) yields the investors’
e¢cient consumption plans,
cc
it

t
) ÷/
it

t
) =
`
c
it
`
c
ct
[cr
t
÷/
ct

t
)] =
`
c
i
`
c
c
[cr
t
÷/
ct

t
)] =
c
it

t
) = ,
it

t
) + ·
i
r
t
. \¸
t
. t = 1. .... 1; i = 1. .... 1;
where ,
it

t
) = [/
it

t
) ÷ ·
i
/
ct

t
)],c and ·
i
= `
c
i
,`
c
c
. Note that the e¢cient consumption
plans have a similar form as in the standard power utility case in Appendix B. Each investor
consumes a constant fraction of aggregate consumption (as a result of the assumed identical
personal discount factors). However, the “…xed component” with habit formation is event-
contingent such that e¢cient individual consumption is not necessarily measurable with
respect to contemporaneous aggregate consumption. This also implies that trading in the
market portfolio and a complete set of zero-coupon bonds is no longer su¢cient to ensure an
e¤ectively dynamically complete market—there must be claims which allow implementation
of the personal event-contingent “…xed components” associated with the habit formation.
Of course, if the habit levels are generated by current and past aggregate consumption per
capita, a su¢ciently varied set of aggregate consumption claims will do.
Scaling by the number of investors and de…ning the average habit level by /
t

t
) =
/
ct

t
),1, the marginal utility of consumption for any investor i is
n
0
it
(c
it

t
)) = ,
1
t

i
1)
1¸c
[cccc
t
÷/
t

t
)]
1¸c
.
58
Inserting this into (31) assuming an e¤ectively dynamically complete market we get the
equilibrium valuation index,
¡
tt

t

t
) =
[cccc
t
÷/
t

t
)]
1¸c
Et
[[cccc
t
÷/
t
]
1¸c
]
. ¸
t
_ ¸
t
. t = t + 1. .... 1.
Note that the valuation index for date t may not be measurable with respect to aggregate
consumption per capita as in the standard power utility case. Furthermore, the equilibrium
prices of zero-coupon bonds are given by
1
tt

t
) =
Et
[n
0
it
(c
it
)]
n
0
it
(c
it

t
))
= ,
1
tt
Et
_
cccc
t
÷/
t

t
)]
1¸c
¸
[cccc
t
÷/
t

t
)]
1¸c
.
Hence, both the equilibrium valuation index and the equilibrium prices of zero-coupon
bonds are independent of both the distribution of initial wealth (as re‡ected in the equi-
librium fractions of aggregate consumption, ·
i
= `
c
i
,`
c
c
) and the distribution of event-
contingent habit levels, /
it

t
). That is, power utilities with identical risk cautiousness (and
personal discount factors) and a general external habit formation allow aggregation. Us-
ing this result and assuming that “habit-adjusted log-aggregate consumption” per capita,
|ccc
t

t
) = ln(cccc
t
÷/
t

t
)), is normally distributed, the analysis proceeds as in the stan-
dard power utility case in Appendix B(see Wachter, 2006, and Campbell and Cochrane, 1999,
for particular parameterization of “habit-adjusted log-aggregate consumption” per capita).
Appendix D: Risk-adjusted Expected Cash Flows and
Certainty Equivalents
In this appendix we discuss the relationship between risk-adjustments to expected cash ‡ows
or accounting numbers in the determination of equilibriumprices and the associated certainty
equivalents. For this discussion, a single-consumption-date economy with no discounting
will be su¢cient and, without loss of generality, we will use dividend value relations. In the
economics of uncertainty, the certainty equivalent of uncertain wealth n is de…ned as the
wealth level which has the same utility as the expected utility of the uncertain wealth, i.e.,
n(CE[n]) =
_
n(n)d(n).
59
where (n) is the distribution of n. Similarly, the risk premium is de…ned as the di¤erence
between the expected wealth and the certainty equivalent, i.e.,
:(n) =
E
[n] ÷CE[n] .
Note that certainty equivalents and risk premia do not, in general, satisfy the value additivity
principle due to the non-linearity of the utility function. The only exceptions are risk neutral
investors, of course, and investors with exponential utility and independence of the individual
cash ‡ows, i.e., if n = . + n with . and n independent, then
_
n(n)d(n) =
_
÷exp [÷(. + n) ,j] d(.)d(n)
= ÷
_
exp [÷.,j] d(.)
_
exp [÷n,j] d(n).
and, hence,
CE[n] = CE[.] + CE[n] . :(n) = :(.) + :(n).
Hence, in general, the concepts of certainty equivalents and risk premia can only be applied
to total wealth, and not individual assets in a portfolio of securities. Moreover, we can only
explicitly calculate certainty equivalents for particular combinations of utility functions and
distributions of wealth. Two such combinations are (A) exponential utility and normally
distributed wealth, and (B) power utility and log-normally distributed wealth. For both
types of utility functions, we know from a simpli…ed version of Proposition 9 that in an
e¤ectively complete market, each investor’s optimally invested wealth will be a linear function
of aggregate consumption. Moreover, we can represent the equilibrium prices of securities
as that of a representative agent economy with a risk tolerance equal to the average risk
tolerance in the economy, j, for the exponential utility case, and a risk cautiousness equal
to the common risk cautiousness in the economy, c, for the power utility case (see Section 5
and Appendix B, respectively).
(A) Exponential utility and normally distributed wealth
In this setting, the price of aggregate consumption per capita follows directly from the
analysis in Section 5 and is given by
\ (ccc) =
E
[ccc] ÷Cov[ccc. ccc,j] =
E
[ccc] ÷Var[ccc],j.
The certainty equivalent for the representative investor with risk tolerance j follows directly
from the standard formula for calculating certainty equivalents with exponential utility and
60
normally distributed wealth (see, for example, Christensen and Feltham, 2003, Appendix
3A), i.e.,
CE[ccc] =
E
[ccc] ÷
1
2
Var[ccc],j.
Hence, the risk premium in the equilibrium price is twice as large as the risk premium in
the certainty equivalent. The reason is that the risk premium in the equilibrium price is
determined such that the representative agent has no incentive to deviate from holding the
aggregate consumption level per capita, while the risk premium in the certainty equivalent
re‡ects the required risk premium to hold the aggregate consumption level per capita instead
of going to a riskless position. We may refer to this di¤erence as the “investors’ surplus,”
since the driving forces are the same as those underlying the consumers’ surplus in consumer
theory.
(B) Power utility and log-normally distributed wealth
In this setting, we cannot directly apply the equilibrium prices derived in Appendix B,
since these are the equilibrium prices on securities with normally distributed cash ‡ows or
accounting numbers, whereas here we must determine the equilibrium price on aggregate con-
sumption per capital, which is log-normally distributed. However, we can apply a simpli…ed
version of Theorem 5, i.e.,
\ (ccc) =
E
[ccc] + Cov
_
n
0
(ccc)
E
[n
0
(ccc)]
. ccc
_
.
For simplicity, assume that the utility function is a CRR utility function, i.e.,
n(ccc) =
1
1 ÷¸
c

. ¸ 0.
where ¸ = c
1
is the coe¢cient of relative risk aversion. Hence,
\ (ccc) =
E
[ccc] + Cov[ccc
¸
. ccc],
E
[ccc
¸
]
=
E
[ccc] +
_
E
[ccc

] ÷
E
[ccc
¸
]
E
[ccc]
¸
,
E
[ccc
¸
]
=
E
[ccc

],
E
[ccc
¸
].
Using the fact that for any scalar q,
ccc
j
= exp [q ln [ccc]] .
61
and the assumption that aggregate consumption per capita is log-normally distributed, i.e.,
. = ln [ccc] ~ `(j. o
2
).
we can calculate the expected values in the equilibrium price using the above-mentioned
formula for calculating expected values of exponentials of normally distributed random vari-
ables. That is,
\ (ccc) =
exp
_
(1 ÷¸)
_
j +
1
2
(1 ÷¸) o
2
¸¸
exp
_
÷¸
_
j ÷
1
2
¸o
2
¸¸
= exp
_
j +
1
2
(1 ÷¸) o
2
÷
1
2
¸o
2
_
.
In order to calculate the certainty equivalent, we must …rst calculate the expected utility
using that aggregate consumption per capita is log-normally distributed, i.e.,
E
[n(ccc)] =
1
1 ÷¸
E
[ccc

]
=
1
1 ÷¸
exp
_
(1 ÷¸)
_
j +
1
2
(1 ÷¸) o
2
__
=
1
1 ÷¸
_
exp
_
j +
1
2
(1 ÷¸) o
2
__

By the de…nition of the certainty equivalent, this implies that
CE[ccc] = exp
_
j +
1
2
(1 ÷¸) o
2
_
.
Taking the log of the ratio between the certainty equivalent and the equilibrium price of
aggregate consumption per capita yields
ln
_
CE[ccc]
\ (ccc)
_
= ln
_
exp
_
j +
1
2
(1 ÷¸) o
2
¸
exp
_
j +
1
2
(1 ÷¸) o
2
÷
1
2
¸o
2
¸
_
=
1
2
¸o
2
.
Since the coe¢cient of relative risk aversion ¸ is strictly positive, we get that
CE[ccc] \ (ccc).
and, thus, the risk premium in the equilibrium price is higher than the risk premium in
62
the certainty equivalent again illustrating the investors’ surplus in the equilibrium prices of
optimally invested wealth.
63

Contents
1 Introduction 2 Risk-adjusted Discount Rates 2.1 Enterprise Valuation . . . . . . . . . . . . . 2.2 Estimating the Cost of Capital . . . . . . . 2.2.1 The WACC-method . . . . . . . . . 2.2.2 The APV-method . . . . . . . . . . . 2.2.3 Estimating the equity cost of capital 2.3 Value Additivity . . . . . . . . . . . . . . . 1 7 8 11 13 15 17 19 21 22 24 28 30 33

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

3 Multi-period Asset Pricing Theory and Accounting Relations 3.1 Fundamental Theorem of Asset pricing . . . . . . . . . . . . . . . 3.2 Accounting Relations . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 Alternative Formulations of Accounting Value Relations . . . . . . 3.4 Equilibrium Valuation Indices . . . . . . . . . . . . . . . . . . . . 4 An Accounting-based Multi-period Equity Valuation Model

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

5 An Accounting-based Multi-period Equity Valuation Model with Exponential Utility 36 6 A VAR Model with Exponential Utility 7 Comparison to the Standard Textbook Approach References Appendix A: Proof of Theorems Appendix B: Power Utility and Log-normal Distributions Appendix C: Habit Formation 39 41 47 49 55 57

Appendix D: Risk-adjusted Expected Cash Flows and Certainty Equivalents 59

1

Introduction

The valuation of uncertain income streams is at the heart of …nancial and accounting research as well as in almost all areas of business. Financial statement analysis and equity valuation challenge market prices of traded stocks, and the valuation of a …rm’ equity is the key in any s equity deal be it an acquisition, a merger, or a private equity deal. The signi…cant increase in the capitalization of private equity funds in recent years also calls for good models of valuing equity when there is no market price. Basic capital budgeting is yet another prominent example in which the valuation of uncertain investment returns is at the heart of the issue. Fundamentally, the question in all of these settings is how to calculate the value today of a stream of uncertain future cash ‡ ows or earnings. The key issues involved are how to account for taxes, in‡ ation, payment patterns, and not least for risk. In this paper, we …rst provide a short review of standard valuation models as they are presented in textbooks and used in practice. These models are based on estimates of expected future cash ‡ ows or earnings and a risk-adjusted cost of capital estimated from time-series of stock returns using some form of capital asset pricing model. We argue that this approach is based on single-period results which do not extend to multi-period settings. For example, the concept of the weighted average cost of capital is based on the well-known result from single-period portfolio theory that the expected return of a portfolio of two assets (such as equity and debt) is equal to the weighted average of the expected returns on the individual assets. On the other hand, in a multi-period setting, the cost of capital is like an internal rate of return on a sequence of expected cash ‡ ows. However, the internal rate of return on a portfolio of two assets is not equal to the weighted average of the internal rates of returns on the individual assets except under very restrictive assumptions such as the same constant growth in perpetuity of the expected cash ‡ ows on the individual assets. If the latter is indeed descriptive, then a simple valuation model like the Constant Growth model may be used, i.e., we may use a simple multiples-based model. Moreover, if the cost of capital is estimated based on short-term stock returns, then leverage, interest rates and risk premia must all be constants. Recall that interest rates change continuously, and that leverage and, thus, equity risk premia change whenever the stock price changes. We develop a comprehensive model based on contemporaneous multi-period asset pricing theory. The purpose is twofold. Firstly, we take the initial steps in developing better valuation models which can explicitly recognize key economic phenomena like stochastic interest rates, di¤erences between short- and long-term interest rates, time-varying risk premia, and the fact that the time-series properties of cash ‡ ows and earnings a¤ect the proper riskadjustments. Equities are long-term claims. Why is it that the information in a readily

1

However. this is indeed the unifying principle used in the pricing of derivatives. Secondly. would be plus or minus in…nity. ow obviously. “The Valuation of Uncertain Income Streams and the Pricing of Options. whereas these factors do not directly a¤ect the risk-adjusted discounts rates in standard valuation models. persistence and mean reversion of cash ‡ ows or earnings are key determinants of not only expected future cash ‡ ows or earnings but also of proper risk-adjustments. we discount risk-adjusted expected cash ‡ ows or earnings using zero-coupon interest rates from a readily observable term structure of interest rates (which re‡ ects expected future spot interest rates and their risk). A key contribution of this paper is to explicitly identify the risk-adjustments in a general multi-period equilibrium model. we use the comprehensive multi-period valuation model for a set stylized examples to shed light on the types and the magnitudes of errors which may be made using standard valuation models based on risk-adjusted discount rates. Instead. these errors are far from being trivial. The forward price is determined such that the current value of the contract is equal to zero. the only risk-adjusted discount rate. risk-adjusting the expected payo¤ on the contract yields that the forward price is equal to the risk-adjusted expected spot price. As it will become clear. Of course. We also teach them that they should be very careful distinguishing cash receipts from revenues 2 . In particular. we show that growth. In order to illustrate this fact. However. Unless the expected spot price is equal to the forward price (which would imply risk-neutral pricing). We teach our students in their …rst …nance class that the value today of an investment is the net present value of the incremental future cash ‡ ows it generates. This ow approach requires that the expected cash ‡ and its current value have the same sign and. which would give a current price of zero. In this paper.” Mark Rubinstein was the …rst to explicitly show more than thirty years ago how closely equity prices and the prices of …xed income claims are related in a multi-period setting. this need not be the case. a simple example is the valuation of a forward contract. and the payo¤ is the di¤erence between future spot price and the forward price. new empirical work is needed in order to …nd the best empirical speci…cation of the model. we reintroduce this link for the purpose of practical equity valuation. We also show that in certain cases. We frame our valuation question in terms of valuing a …rm’ common equity. these factors may indirectly a¤ect the estimates of the cost of capital from time-series of stock returns. it s should be noted that our approach is applicable for the general question of how to …nd the value today of a stream of future cash ‡ ows or earnings in other settings. such as in basic capital budgeting. a risk-adjusted discount rate can simply not be used to …nd the current value of an uncertain future cash ‡ (or accounting number). More generally. The basic idea is instead of discounting expected cash ‡ ows or earnings with risk-adjusted discount rates.observable term structure of interest rates is not used in practical equity valuation? In his seminal paper.

the …rm may still have positive earnings and even positive residual income (which measures earnings in excess of the capital cost of using the book value of invested capital). In the …rst place.and expenditures from costs— the di¤erence being the non-cash accruals. of course. 3 . For example. and …rms do not recognize the net present value of a capital investment as an asset on the balance sheet at the investment date. accuracy and reliability. to provide information about value creation and performance of its employees— in a timely manner and for each period separately. thus. Instead. For example. The existence of these accounting rules can only be understood by considering the stewardship role of accounting and the incentives of managers and …rms to misrepresent information. plant and equipment. It is important to recognize that value creation (or performance) as re‡ ected in …nancial statements does not perfectly re‡ value creation in market value terms due to the ect particular accounting rules applied. An advantage of accounting-based models over the discounted cash ‡ models is that ow value creation is re‡ ected earlier in the accounting-based models. whereas ow understanding the implications of …nancial statement articulation requires more training. The key. these bene…ts will be recognized as increases in future earnings as they are realized. Accrual accounting recognizes these investments as assets on the balance sheet and. The cash ‡ is a very easy concept to understand for …rst year students. is the impact of the time value of money. and that any “errors”made will be perfectly o¤set in the net present value of subsequent residual income numbers. These accounting-based models provide exactly the same net present values as the discounted cash ‡ ows models. in high growth …rms. in recent years there has been an increased interest in accounting-based valuation models which instead of discounting future cash ‡ ows determine the net present value by discounting future residual income or abnormal earnings growth anchored by current book value or capitalized one-period ahead earnings. The determination of the particular accounting rules must strike a balance between timeliness. The beauty of accounting-based valuation models is that they recognize value creation in the short run although imperfectly. Nevertheless. …rms have for several centuries used accrual accounting instead of cash ‡ accounting (which would fully expense capital ow investment expenditures like a reduction in free cash ‡ ows). For example. research and development expenditures are typically fully expensed. respectively. The reason for this equivalence is the articulation of …nancial statements within and across periods. a conservative assessment of the book value of an asset today will eventually show up as higher residual income numbers in future periods— the “error”in current book value is perfectly balanced against the increase in the net present value of future residual income numbers. the free cash ‡ ows are typically negative in the short run due to high capital investments in new property.

should be of relevance to both an accounting and a …nance audience.As a practical matter. These are the key inputs to the accounting-based valuation models. From these forecasts. these forecasts of free operating cash ‡ ows are no more reliable than the forecasts of the accounting numbers on which they are based. Our starting point is the implications of no-arbitrage in perfect multi-period …nancial markets. An immediate consequence of the value additivity principle is that the value of a sequence of future uncertain cash ‡ ows must be equal to the sum of the values of each of the future uncertain cash ‡ ows. Hence. Obviously. our approach is signi…cantly in‡ uenced by Rubinstein (1976). i. the choice between discounted free cash ‡ models versus accounting-based models is merely a choice between how the results and ow the analysis are presented. Instead. Hence. As mentioned above. ending book value of operating assets equals opening book value of operating assets plus operating income minus the free cash ‡ from operations ow (which is the amount that can be used to pay the debt and the shareholders). Although important issues. book values of operating assets. pro…t margins. these models can be applied directly by calculating future residual operating income or abnormal operating income growth from forecasts of operating income. our focus is on discounting ow cash ‡ ows or accounting numbers in a theoretically sound manner in a multi-period setting under uncertainty. the value of a portfolio of two uncertain income streams must be equal to the sum of the values of the individual income streams. If the discounted free cash ‡ models are used.. If calculated at the future date. future operating earnings and book values of operating assets can be calculated. Instead. hence.e. ow A likely reason is that past free cash ‡ numbers contain a lot of transitory noise due.. i. A fundamental property of no-arbitrage pricing is that the valuation operator is a linear functional. given the same set of forecasts the models are mathematically equivalent (if the growth assumptions and the discounting are not messed up). the latter 4 . and the capital charge on opening book values or last period’ free cash ‡ s ows. the paper is on the foundations of valuation and discounting under uncertainty and. Similarly. ow for example. and asset turnover.e. …nancial analysts and others evaluating equities do typically not forecast free cash ‡ ows directly even if they are using the discounted free cash ‡ model. In that sense. a common approach is to forecast value drivers like sales. That is. the forecasts of free cash ‡ ow ows are obtained from the operating asset relation. to the particular timing of past capital investment expenditures. the focus of this paper is not to discuss the most e¢ cient ways of forecasting future accounting numbers or free cash ‡ ows or to advocate accounting-based valuation models over discounted free cash ‡ models. or just the value additivity principle. the value of a given future uncertain cash ‡ ow must be equal to the value of the expected cash ‡ plus the value of di¤erence between the ow uncertain cash ‡ and the expected cash ‡ ow ow.

In the economics of uncertainty.is referred to as (minus) the cash ‡ow risk premium. its current value must be equal to its ow discounted value using the current zero-coupon interest rate for the particular future date as the discount rate. we show that this implies that the risk-adjusted expected cash ‡ ows are lower than the certainty equivalents. In a general equilibrium model with HARA utilities. thus. The di¤erence between the expected cash ‡ and the cash ‡ risk premium is the risk-adjusted expected cash ‡ow.e. No-arbitrage pricing is very useful to …nd the relative prices between 5 . However. the meaning should be clear: the certainty equivalent of an uncertain cash ‡ is the certain ow cash ‡ which has the same equilibrium value as the uncertain cash ‡ ow ow. This concept is not equal to the risk-adjusted expected cash ‡ in equilibrium valuation models (as it might appear from reading at least ow some …nance textbooks). One way to see this is to note that certainty equivalents do not satisfy the value additivity principle and.. If the language of certainty equivalents is retained for risk-adjusted expected cash ‡ ows. violate the implications of no-arbitrage.e. the certainty equivalent of optimal invested wealth is the value the decision maker attaches to total wealth (like in a take-it-or-leave-it o¤er). Of course. Hence. the certainty equivalent of uncertain wealth is de…ned as the level of wealth which has the same utility as the expected utility of the uncertain wealth. this de…nition of certainty equivalents is moot until it is speci…ed how it can be calculated. this method is referred to as the certainty equivalent method— with the certainty equivalent being equal to the risk-adjusted expected cash ‡ ow. it s is impossible to determine the certainty equivalents of individual securities in a meaningful way even if the investors’utility functions are known— the certainty equivalent of an asset depends on the other risks the decision maker has in his portfolio (and in his other income). exponential utility). no-arbitrage implies that the current value of a sequence of future uncertain cash ‡ ows is equal to the sum of the sequence of future risk-adjusted expected cash ‡ ows discounted by the associated zero-coupon interest rates. Hence. Unfortunately. i. the wealth level which makes the decision maker indi¤erent between receiving this level of wealth with certainty and retaining the uncertain wealth. no-arbitrage alone provides no guide to actually determining the risk-adjustments to expected cash ‡ ows. whereas the equilibrium prices of risky securities are determined such that no investor has any incentive to change his optimal portfolio of securities (on the margin).. such as Grinblatt and Titman (2002). one should be careful about the language when it comes to actually determining the certainty equivalents. Since the ow ow risk-adjusted expected cash ‡ is a certain number. Moreover. In standard …nance textbooks. The certainty equivalent of a portfolio of two uncertain cash ‡ ows is only equal to the sum of their individual certainty equivalents if these cash ‡ ows are stochastically independent and the decision maker’ preferences have no wealth e¤ects (i.

it is well-known (see. the risk-adjustment is determined as the covariance between the cash ‡ and ow (log-)aggregate consumption times this valuation-index factor. Given the analysis in the preceding section. In the latter approach. Hence. Furthermore.e. we derive an accounting-based multi-period equity valuation model in Section 4 with equilibrium risk-adjustments determined by the prices of aggregate consumption claims. For purpose of illustration. However. and aggregate consumption and residual operating income are jointly normally distributed. for example. Section 5 includes a general equilibrium analysis of a setting in which the investors have exponential utility. and Appendix D discuss the relationship between risk-adjusted expected cash ‡ ows (or accounting numbers) and certainty equivalents. Appendix B shows how the analysis is adjusted if aggregate consumption is lognormally distributed and investors have power utility (average risk tolerance is simply replaced by the common risk cautiousness). i. normally distributed aggregate consumption goes well with exponential utilities. The latter can be determined from the market prices of aggregate consumption claims (like in the standard CAPM). In the …rst section we give a short review of standard valuation models based on risk-adjusted discount rates with an emphasis on the many implicit assumptions made in this approach. then Stein’ Lemma can be used to separate the valuation index out of s the covariance as a factor equal to the expected value of the derivative of the valuation index. We assume that equilibrium prices are determined in an e¤ectively dynamically complete market (such that equilibrium allocations are Pareto optimal). whereas the latter is the preferred combination in …nance. In that case. In Section 3 we review key results from multi-period asset pricing theory in discrete-time. Based on these results. by assuming a particular set of investor preferences. Section 6 examines a simple setting in which residual operating income and 6 . an equilibrium model is required to determine the prices of the primitive assets spanning the market risks. we will show both. Christensen and Feltham. The former combination is probably the best known in accounting research. if it is assumed that aggregate (or log-aggregate) consumption and the cash ‡ are jointly norow mally distributed.. or by using a general equilibrium approach.redundant derivative securities and the primary securities. whereas log-aggregate consumption goes well with power utilities. a similar model can be derived using free cash ‡ ows. Hence. The remainder of the paper is organized as follows. and show how equity valuation models can equivalently be based on free cash ‡ ows or accrual accounting numbers. 2003. and that the investors have homogeneous beliefs and time-additive preferences. Chapter 6) that the risk-adjustment to expected cash ‡ ows can be determined as the covariance between the cash ‡ and a valuation inow dex measurable with respect to aggregate consumption at that date. Appendix C extends the setting to preferences with external habit formation (which recently has gained popularity in asset pricing theory).

residual operating income.e. the value of the …rm (also denoted the enterprise value) at the valuation date t. This construct (or plug number) may be useful for strategy evaluation within the …rm. The reader is probably already su¢ ciently confused at this point— but hopefully at a higher level. we review the standard approaches to (equity) valuation using risk-adjusted discount rates. Instead. while an equity cost s of capital is used in the former model. We skip the concluding remarks. is determined as the discounted expected free cash ‡ ows conditional on information available at date t. Emphasis will be on the many implicit assumptions made in this approach. The common practice in equity valuation is to use some kind of dividend. i. using free cash ‡ ows. but it is certainly not useful for security valuation or net present value calculations within the …rm— in these cases we are trying to determine the current value of the sequence of future uncertain cash ‡ ows. you should immediately stop reading and go to the next section. or abnormal operating income growth discount model. If you at any point reading through this section have had enough of outrageous assumptions. the implied cost of capital.aggregate consumption are given by a simple …rst-order vector-autoregressive model with mean-reversion to a deterministic exponential trend. the …rm’ cost of capital is merely an s internal rate of return. we must determine the …rm’ cost of capital directly. For example. This analysis is the basis for the comparison of the general equilibrium analysis and the standard approach of using a time-independent risk-adjustment to the required rates of returns in Section 7. we will be using the free cash ‡ discount model to illustrate the standard valuation ow approaches. s In this section. taking the value of the s …rm and the expected future free cash ‡ ows as given.. the discount rate used is the …rm’ cost of capital. The following sections aim at developing such alternative approaches. free cash ‡ ow. i. Vt . f cf . 1 7 . These latter models will take center stage in subsequent sections. but we could equally well have used the accounting-based models.1 Vt = 1 X Et [f cf ] . KFt t =t+1 (1) where KFt is the …rm’ gross cost of capital at the date t.. In the latter three models. 2 Risk-adjusted Discount Rates In this section. this section is written in order to motivate you to be open-minded for alternative approaches to equity valuation. Indeed. The analysis stresses the importance of both the contemporaneous correlation between residual operating income and aggregate consumption and the time-series properties of these processes.e. Of course.

Ingersoll. by the formula for covariances and iterated expectations.t ] Et [f cft+2 + Vt+2 ] Et [Et+1 [RFt+2. this correlation does not appear to be trivial.t+1 ] 1 ]] + Et [RFt+1. ow this procedure requires a model for how the future cum-dividend value of free cash ‡ ows. Hence. If the expected returns are de…ned as the spot riskless interest rate plus a risk premium. i. using one-period ahead expected returns iteratively for valuation of multi-period cash ‡ streams is problematic for several reasons. Secondly. a random variable given date t information. ]. ::: Using this relation iteratively starting at date t. Et [Et+1 [RFt+2.t+1 ] 1 ] + Et [f cft+2 + Vt+2 ] Et [RFt+1.t+1 ] 1 ] Et [RFt+1. like the Cox. as it appears from the last term. t + 1. for the moment.. are correlated with future conditional expected returns.t+1 ] 1 ] + Covt Et+1 [f cft+2 + Vt+2 ].t ] Clearly.t+1 ] 1 ] + Et [RFt+1. t + 1. f cft+2 + Vt+2 . ::: (see below).t ] Et [RFt+1.t ] Et [Et+1 [RFt+2. First. in a single-factor interest rate model.t+1 ] 1 Et [Et+1 [RFt+2. given the available information at date . E [RF +1.t ] Covt [Et+1 [f cft+2 + Vt+2 ]. and Ross (1985) model.1 Enterprise Valuation Suppose. even if we assume that cum-dividend values are uncorrelated with expected 8 . in general.t ] Vt = The one-period ahead conditional expected return at t + 1 is.t ] Et [f cft+1 ] = Et [RFt+1. we get that Et [f cft+1 ] Et [Et+1 [f cft+2 + Vt+2 ] Et+1 [RFt+2. we are able the estimate the expected one-period ahead gross returns on …rm value. = t.e. Et+1 [RFt+2.t+1 ] 1 ] + Et [RFt+1. it follows that V = E [f cf +1 + V +1 ] . it follows that Vt = Et [f cft+1 ] Et [RFt+1. a lower spot interest rate would lead to a higher ex-dividend value of future cash ‡ ows even if these are non-stochastic (due to lower discount rates). ] = t. E [RF +1. By the de…nition of conditional expected returns.t ] Et+1 [RFt+2.2. For example.

i.t+1 ]] Et [RFt+1. we can determine the current value of the …rm as2 Vt = 1 X =t+1 Deterministic one-period ahead expected returns require that future spot interest rates and risk premia are deterministic.returns..s ] = (1 + ) t Note that with deterministic interest rates.t ] Et [RFt+1. Suppose the risk premia are the same in all future periods. Hence. the zero-coupon interest rates.t ] Et [RFt+2. t . Hughes. the discount factor in the second term is higher than if the discount factor is based on the expected returns based on date t information (due to Jensen’ inequality). at date t for maturity dates . are given by (1 + t) t Q s=t (1 + s ) = Q s=t (1 + s ) : 2 A transversality condition is also at work here in an in…nite horizon setting. “average”expected future returns tend to overstate the implied cost of capital in (1). > Et [RFt+1. In that case. Liu and Liu (2008) develop these ideas more fully and show how it a¤ects the interpretation of empirical …ndings for the relationship between …rms’cost of capital and the informational environment. the discount factors may be written as Q s=t Q Et [f cf ] : s=t Es [RFs+1. s 1 1 Et [Et+1 [RFt+2.e. in this setting. thus.t ] Et [Et+1 [RFt+2.t+1 ] 1 ] = . 9 . one-period forward rates are equal to the associated future spot interest rates and. We may then write these expected returns as Es [RFs+1. Valuation based on estimates of risk-adjusted short-term expected returns is immensely simpli…ed if it is assumed that future expected returns are deterministic.t+1 ] with the di¤erence being determined by the uncertainty in next period’ conditional expected s return. Es [RFs+1.s ] = (1 + s ) (1 + ) where s is the spot one-period interest rate at date s and Hence.s ] is the constant risk premium.

respectively). it is well appreciated that the term structure of interest rates is not ‡ and the recommendation in almost all textbooks is to use a at. The last line in Table 1 shows the price and the cost of capital of a portfolio of the two cash ‡ sequences A and B. this is due to the fact that the cash ‡ ows in the increasing sequence A “load” relatively more on the high twoperiod zero-coupon interest rate than on the low one-period zero-coupon interest rate. for example. In that case.73 KFt 1 WACC 5. the cost of capital on the portfolio is somewhere ow This is well known in the …xed income literature: the yield-to-maturity on a bond does not only depend on its maturity. growth s in free cash ‡ ows? In order to illustrate this as simply as possible.g. e. 10-year treasury yield as a proxy for the riskless interest rates. This illustrates the fact that with non-‡ term structures of interest rates the cost of capital at used to discount multi-period sequences of cash ‡ ows must re‡ not only risk but also the ect expected growth in cash ‡ ows.30% 5. and Christensen and Nielsen. Caks. This implies that the increasing sequence A has a higher cost of capital. 1977.and bond-type e¤ects on yield-to-maturity. i.Hence.e. Hence. i. there are no risk premia...08 140.27% Table 1: Cost of capital for a non-‡ term structure of interest rates. Of course. Note …rst that the term structure of interest rates is assumed to be increasing (which is the most common empirically). ow Date t + 1 Date t + 2 Price. Of course. 1987. for the so-called coupon.58% 4. Vt Interest rates. in the …xed income literature discounting is always done using the term structure of zero-coupon interest rates. it also depends on its payment pro…le (see. internal rate of return.e. It is less appreciated that the cost of capital must also re‡ the growth in the sequences of cash ‡ ect ows which are to be discounted— one can only guess the size of the error introduced. how is the cost of capital KFt a¤ected by. Table 1 shows the prices of two such two-period cash ‡ sequences using a given set of zero-coupon interest rates. 3 10 . = 0.3 In implementations of the standard approaches to equity valuation using risk-adjusted discount rates. the value of the …rm may be written as Vt = 1 X Et [f cf ] [(1 + t ) (1 + )] =t+1 t: Equating this expression for the value of the …rm with the …rm value in the de…nition of the …rm’ cost of capital (1). t Cash ‡ sequence A ow Cash ‡ sequence B ow Portfolio of A and B 4% 50 100 150 6% 100 50 150 137. than the decreasing sequence B.96% 5.. consider an example in which the …rm’ future free cash ‡ s ows are deterministic (like a default-free bond). at There are several important things to note from this very simple example.65 277.

This re‡ ects that. the WACC underestimates the cost of capital on the portfolio. for multi-period sequences of cash ‡ ows. we may estimate the cost of capital s for debt and equity as the implied cost of debt and equity. KFt WACCt = Dt St KEt + KDt . KDt t =t+1 where u and z are the net payments to equity. it does not hold— the internal rate of return on a portfolio is not 11 . i. debt and equity. and this only holds in single-period settings and in settings in which the sequences of expected cash ‡ ows are assumed to grow at the same rate in perpetuity. the implied cost of capital for the two sources of capital are determined by 1 X Et [u ] . 2. the cost of capital on a portfolio of these sequences is a highly non-linear weighted average of the cost of capital on the individual sequences. St + Dt S t + Dt (2) where St and Dt are the market values of the …rm’ common equity and net debt. in general. respectively. That is. s with St + Dt = Vt . respectively. otherwise.. taking the market values of debt and equity and their expected net cash ‡ ows as given. with the sum being equal to the free cash ‡ by the …rm’ budget constraint. as a weighted average cost of capital (WACC) using the relative values as weights. respectively.e. The WACC calculation assumes it is a linear weighted average. A common approach is to determine it as the weighted average cost of capital. f cf = u + z . in the example.between the cost of capital on the individual cash ‡ sequences.and debtholders at date . the …rm’ cost of capital KFt is not estimated directly. Note that.2 Estimating the Cost of Capital In practice. The common approach ow in equity valuation is to calculate the cost of capital for portfolios of. it is estimated s indirectly by …rst estimating the the equity cost of capital KEt and the debt cost of capital KDt and then from these calculate the …rm’ cost of capital using accounting identities and s value additivity principles. then it is easy to show that the …rm’ cost of capital is indeed equal to the weighted average cost s of capital but. ow s If the expected payments are all in…nite sequences growing at the same constant rate. Instead. for example. Like for the …rm’ cost of capital. St = KEt t =t+1 1 X Et [z ] Dt = .

V = UA + T S + BC + ::: (4) (3) The latter is the so-called adjusted present value (APV) method. and Vt = Et [f cft+1 ]=(KFt Using the …rm’ budget constraint. plus any other debt related items. and as the sum of the values of the comprehensive operating assets. Then G): G). UA TS BC : V S D V The debits are the value of an otherwise identical unleveraged …rm. so this approach clearly does not work for equity— we must determine the cost of equity directly.. it might be a reasonable approach to estimate the cost of debt as its implied cost of capital.e..5 Before we go into the details of how the cost of equity is determined empirically. it is useful to look at a simple T-account (in market value terms) in order to keep the language straight. we may determine the value of the …rm both as the sum of debt and equity values. for example. Dt = Et [zt+1 ]=(KDt 1. whereas most accounting textbooks seem to 4 Let both Et [u ] and Et [z ] grow at a constant rate g = G St = Et [ut+1 ]=(KEt G). i. the value of the future corporate tax savings on net debt. the value of future bankruptcy costs. i. Grinblatt and Titman. Hence. V =S +D . we get s KFt = [Et [ut+1 ] + Et [zt+1 ]] =Vt + G = [St (KEt G) + Dt (KDt G)] =Vt + G = (St =Vt )KEt + (Dt =Vt )KDt which is (2). 5 Did you notice the circularity in the WACC calculation? 12 . 2002).equal to the weighted average of the individual assets’internal rates of return.4 If debt is close to being riskless debt. T S .e. UA . BC . Combining the two relations yields the value of equity as S = (UA + T S + BC + :::) D : (5) The APV-method seems to be the recommended method in standard …nance textbooks (see. we are trying to determine the value of equity. However.

of course. The APV-method has a comprehensive de…nition of operating assets. operating income. 6 13 . This gives much more ‡ exibility in how. as the net cash ‡ paid to debt and ow equity. The advantage of the APV-method is that it directly follows the value additivity principle such that the value of the …rm is given as the sum of the values of the di¤erent components.e. ] 1] D : By the de…nition of the …rm’ expected one-period ahead return.. bankruptcy cost must not be taken to mean higher yields on debt due to default risk. the di¤erence between the two is the tax savings from net interest. operating income. and in the operating income in accounting-based models. i. i.recommend the WACC-method to discount free cash ‡ ows from operations with an after-tax weighted average cost of capital (see.e.6 The conceptual di¤erence between the two approaches is what is included in the operating assets. 2007. In the following discussion we will.e. f cf o .1 The WACC-method As above. 2004). for simplicity. E [RD +1. how the corporate tax savings on net interest payments are dealt with. ] 1. f cf +1 = f cf o +1 + tc [E [RD +1. Bankruptcy costs are deadweight losses associated with default. times the opening value of net debt. i. for simplicity. and it requires a particular and strong assumption about the future debt policy (see below).. the “unleveraged free cash s ‡ ows. ] +1. On the other hand. ] 1] D : Here.”If there are no bankruptcy costs.. de…ne comprehensive free cash ‡ows. 2.2. f cf . i.e. as everything in the comprehensive free cash ‡ ows which has nothing to do with the …rm’ …nancing. but it should be clear from the discussion above how to include these in the APV-method. is most relevant for highly leveraged …rms. and free cash ‡ows from operations. for example. Penman. the free cash ‡ ows. one wants to model the tax savings from debt and how to include an estimated value of future bankruptcy costs. whereas the WACC-method uses a narrow de…nition of operating assets. Assume. ignore any bankruptcy costs. that these tax savings are equal to the corporate tax rate tc times one minus the expected one-period ahead return on debt.. and free cash ‡ ows. for example. The latter. the WACC-method has problems in including bankruptcy costs consistently. in particular. and free cash ‡ ows. excluding bankruptcy costs and the corporate tax savings on net interest (and any associated deferred tax bene…ts). and Lundholm and Sloan. it then follows that s V = E [f cf o +1 +V +1 ] + tc [E [RD E [RF +1.

t ] =t+1 t t .t ] St V Et [f cf ot+1 + Vt+1 ] + Et [RDt+1. Et [WACCtAT ] 1] Dtt V (6) Et [REt+1. Vt = where Et [WACCt+1. We can now try to use (6) iteratively as we did in Section 2.t ] St + [Et [RDt+1.t ]. but the weights must also be constants. at each instant the stock price 14 . i. f cf ot+1 .t ] Substitution then yields that Vt = or.. must be constant (which also means that S =V is a constant.t ] AT Dt St + Et [RDt+1. Vt = 1 X Et [f cf o ] AT E [WACCt+1.t ] : 1] Dtt V Note from (3) that we can view the value of the …rm as the value of a portfolio of the net debt and the equity.Starting at the valuation date t and rewriting.e. E [WACC AT ] = Et [WACCt+1.. we get that Vt = Et [f cf ot+1 + Vt+1 ] Et [RFt+1.t ] : Vt Vt t Et [REt+1. not only are the expected one-period ahead expected returns on debt and equity constants. Et [WACCt+1.t ] V Et [f cf ot+1 + Vt+1 ] . ::: This means that +1.t ] = Et [REt+1. equivalently.t ] Dtt tc [Et [RDt+1. = t + 1. since V = S + V ).e. D =V . and we will encounter exactly the same problems with stochastic expected one-period ahead returns as well as some additional problems. and recognizes the tax shield from net debt in the after-tax discount factor. leverage.t ] 1]] Dt : Vt (7) That is. we must assume that the weighted average of expected one-period ahead after-tax returns AT are constant over time..t ].t ] tc [Et [RDt+1. Hence. Hence.1.t ] Vt tc [Et [RDt+1.e. the weighted average AT of expected one-period ahead after-tax returns. In order to use (6) to derive the familiar WACC-valuation formula. i. i. the WACC-approach takes the free cash ‡ ows from operations. i.e.. Et [RFt+1. it follows (from basic portfolio theory) that the expected one-period ahead return on the …rm is equal to the weighted average of the expected one-period ahead returns on the individual assets in the portfolio.

the answer is +1. Two common assumptions in the literature are that (A) the expected amounts of future tax shields are equal to the expected current tax shield. 2. where E [RUA +1. and (B) the amount of future debt is continuously adjusted so that the tax shields are proportional to the free cash ‡ ows from operations— but. it is less restrictive to assume that the expected one-period ahead returns on the unleveraged assets are constant. compared to the WACC-method. ] is +1 ] . 15 .2. any other assumption is also feasible. Of course. ] +1 ] . an advantage of the APV-method over the WACC-method is that the assumptions on future tax shields can be made independently. so we may determine the value of the unleveraged assets as UAt = 1 X Et [f cf o ] E [RUAt+1. ] is the expected one-period ahead return on the tax savings. where E [RT S +1.2 The APV-method The APV-method calculates the values of the free cash ‡ ows from operations and the tax shields of interest on net debt separately. changes in leverage is closely related to changes in expected one-period ahead expected returns on debt and equity. Hence. i. Would it be possible that these changes are o¤setting such that the expected discount factors E [WACC AT ] are independent of leverage? Unfortunately.e. As noted above. ] 1] D + E [T S E [RT S +1. of course.t ] =t+1 t t : The value of the tax shields of interest on net debt is similarly determined as TS = tc [E [RD +1. ] is the expected one-period ahead return on the unleveraged assets.. Note that the expected one-period ahead returns on the unleveraged assets must be independent of leverage— this is Modigliani/Miller at work. no. This is because the discount factors include the tax shields from net debt. new debt has to be issued or existing debt must be retired to keep the leverage constant (since debt has a stock price sensitivity di¤erent from one). V = UA + T S The value of the free cash ‡ ows from operations at date UA = E [f cf o +1 + UA E [RUA +1.changes.

t ] Et [RDt+1.e. ]: The determination of the …rm value can therefore be based on the comprehensive free cash ‡ ows discounted at the expected one-period ahead returns on the unleveraged assets.t ] 1 + Dt T St St + Et [RT St+1. we get T St = tc [Et [RDt+1. Hence.. and the conditional variance of the factor. E [RT S +1.t ]) St (1 tc ) Dt : St = Et [RUAt+1.t ] St St (1 tc ) Dt .t ] Dt T St Et [RDt+1. 7 16 . the risks associated with future tax shields are the same as the risks associated with future free cash ‡ ows from operations and. (B) Proportional tax shields In this setting. the expected one-period ahead return on the tax savings is the same as the expected one-period ahead return on debt.. and assuming this is a constant in perpetuity. i. Note that the factor does not have to be the return on the market portfolio. E [RT S +1. Et [REt+1.t ] (10) Recall that the expected one-period ahead returns on the unleveraged assets are independent The beta of an asset is de…ned as the ratio of the conditional covariance between the asset’ return and s some factor. ] = E [RUA +1.(A) Constant expected tax shields In this setting.t ] = Et [RUAt+1. ] = E [RD +1.t ] 1] Dt = tc Dt : 1 Using (5) and the explicit value of the tax savings.t ] or in “beta”terms7 E t (8) = UA t + UA t D t (9) These are the standard relations to unlever expected equity returns and equity betas.t ] + (Et [RUAt+1.e. there is no implicit assumption of the CAPM being true at this point. but recall that we have made a very strong assumption about constant expected one-period ahead returns on debt in order to establish these relations. hence. i. Et [RDt+1. we can express the expected one-period ahead return on equity as follows. Vt = Et [f cft+1 + Vt+1 ] : Et [RUAt+1. ].

t ] = Et [RUAt+1.t ] St Dt + Et [RDt+1. St D t Et [RDt+1. the tax shields are re‡ ected in the comprehensive free cash ‡ ows.t ] is also relatively precise..t ]) . we get 1 X Vt = Et [f cf ] BT E [WACCt+1. then the estimate of the before-tax Et [WACCt+1. i.t ] T St St Dt Et [RDt+1. Hence.t ] + (Et [RUAt+1.t ]: s But again.t ] : Vt Vt (13) BT Note that the before-tax Et [WACCt+1. we can also rewrite (11) to obtain a before-tax weighted average of expected one-period ahead returns.t ] = Et [REt+1.t ] =t+1 t t . the CAPM says that the expected one-period ahead return on a stock is equal to the riskless return plus the 17 . contrary to WACC-method.of leverage.2.t ] is independent of leverage (as opposed to the afterAT tax Et [WACCt+1.3 Estimating the equity cost of capital The common approach to determine the equity cost of capital is to use one of the standard asset pricing models such as the CAPM or the APT model. why should the expected one-period ahead returns on the unleveraged assets be constant in a multi-period setting under uncertainty? 2.t ] Dt St (11) = Et [RUAt+1. and using the one-period value relation (10) iteratively. Et [WACCt+1. BT and comparing to (2) it follows that the …rm’ cost of capital is KFt = Et [WACCt+1. Assuming that these expected unleveraged returns are constant. if the expected equity returns are estimated using a time-series of stock returns over a period with relatively constant leverage. these relations are precisely the same as in a before tax setting.t ] and that E t = UA t + UA t Dt : St (12) That is. since it is determined as the expected returns in an otherwise identical unleveraged …rm.t ] BT Et [RUAt+1. For example. and the expected BT equity returns are unleveraged by (11). Note that. This is also re‡ ected in how the expected one-period ahead returns on equity and equity betas are unleveraged in this setting. and not in the discount factor.e.t ] in (7)).t ] 1 + Dt T St St + Et [RT St+1. Again. Of course. using (5) we get that Et [REt+1.

= E t + E t RM +1. is the return on an optimally invested portfolio. ] E . i. whether the estimated alpha in (16) is di¤erent from zero. + [E [RM +1.e. RE +1.e. it is well known that the CAPM is only valid in a multi-period setting if the investment opportunity set is deterministic or investors are myopic. R +1. s E [RE +1.e. respectively. there is an omitted variable in (15).e. and RM +1. = s. The …rst uses a time-series of raw returns over the estimation period. ] The …rst step is to estimate the equity beta as the slope of a time-series regression of equity returns on the returns of some stock index (as a proxy for the market portfolio). i. ]=R +1. i. t 1.8 The second recognizes that the riskless spot interest rate varies over the estimation period and. That is. this assumes that the equity beta is constant over the estimation period. . . ] R +1. = s. uses excess returns. ] : Var [RM +1. = 1 + is one plus the riskless spot interest rate. RM +1. ]+" +1. thus. this requires that leverage is relatively constant over the estimation period (and that the unleveraged equity and debt betas are also constant). 1 where RP +1. have log utilities. = Cov [RE +1. s + 1. i. are the observed returns in period on the stock and the stock index. and E +1.. and the slope. :::. E [RP +1. (15) where RE +1.e. On the other hand. . while (16) provides a direct test of the validity of the CAPM. which is the We are not going to discuss the econometric problems in the estimation such as the fact that using a stock index as proxy for the market portfolio induces an errors-in-variables problem resulting in a downward biased beta estimat. i. s + 1. A deterministic investment opportunity set means that the so-called capital market line. 1 2 ] Var [RM ] Var [RP +1.. + [E [RM +1. As we noted above. can only shift in a deterministic manner.. RM on the market portfolio. ]=R +1.. +1. i.. the intercept.expected one-period ahead excess return on the market portfolio times the stock’ beta. RE +1. (14) is the gross return where R +1.e. ]2 . Of course. ] R +1.. 8 18 . = E t + E t [RM +1. R +1. t 1: (16) Under the assumptions of the CAPM with varying interest rates. :::. +" +1. Two types of regression equations are typically used. which is the riskless return.

but stochastic expected oneperiod ahead returns remain a problem for iteratively using single-period value relations to value a multi-period sequence of future cash ‡ ows or accounting numbers. There is signi…cant empirical evidence that excess returns are time-varying and to some extent predictable (see. Hence. then only the starting and ending cum-dividend values of the index over the estimation period a¤ect the estimate. a Fama-MacBeth type cross-section regression of average returns on beta-sorted portfolios on their estimated betas is typically used to estimate the expected excess returns on the market portfolio.. throwing away potentially signi…cant covariances.. The following quote is from Penman (2007). such estimates are very unreliable. Instead. the value of a portfolio of two sequences of cash ‡ ows must be equal to the sum of the values of the individual sequences of cash ‡ ows. RP i = 0 + M bi + ui : Among the empirical stylized facts for these regressions are that the estimates of the market risk premium M are typically too low compared to average excess returns on broad stock indices. i. and Lettau and Ludvigson. i. The next step is to estimate the expected one-period ahead excess return on the market portfolio. violating Jensen’ inequality. the game in town is time-varying risk premia. No one knows what the market premium is. for example. Campbell and Shiller. and that there is a lot of variation in the estimates across di¤erent empirical studies. However. Hence.e. 1988a and 1988b. page 691: “Compound the error in beta and the error in the risk premium and you have a considerable problem. These models contain a strong element of smoke and mirrors. i.” In the recent asset pricing literature.”must both be non-stochastic. relying on the CAPM. And adopting multifactor pricing models adds more risk premiums and betas to estimate. even if true.e.. 2. if continuously compounded returns are used in the estimation (which is generally recommended). there is an implicit assumption in the test for a zero alpha in (16) that spot interest rates are deterministic. The CAPM. 2001). is quite imprecise when applied. This may be a useful way to proceed.e. etc.3 Value Additivity As noted above a fundamental property of valuation in perfect markets is the value additivity principle. It might be tempting to estimate this expected return simply by calculating the average excess return on the stock index over the estimation period. non-‡ term structures of s at interest rates.“market price of risk. The standard valuation approaches 19 .

is determined as 1 X Et [f cf ] . only by the contemporaneous covariance with the market return but also by the time-series 20 .. Besides not satisfying the value additivity principle. 2006). For example. as the sum of the value of the individual future cash ‡ ows. Ang and Liu (2004) recognize that one should not discount future expected cash ‡ ows with the same discount rate. and we are reasonably certain that it is constant through time. That is. rather. it appears as if they determine the value of the sequence of future cash ‡ ows as the sum of the values of the individual future cash ‡ ows using speci…c risk-adjusted discount rates for each future expected cash ‡ ow. unlevering it to get a risk-adjusted discount rate for the operations. They parameterize their model with speci…c stochastic processes for asset betas and market risk premia (in return form) and …nd that signi…cant pricing errors can occur if a constant discount rate is used. More direct evidence that “there is something rotten in the State of Denmark” is that the value additivity principle and (17) imply that the current value of f cf for some particular future date must be positive if Et [f cf ] > 0 (at least if t is even). Hence. Even if we are successful in estimating the expected one-period ahead return on equity (and debt). the risk-adjusted discount rate used to discount a particular future expected cash ‡ does not re‡ the risk in that cash ‡ but. the value of the sequence of future cash ‡ ows is determined as the value of the portfolio of future cash ‡ ows and not. the trouble with this kind of model is that everything is hidden in the date-speci…c risk-adjusted discount rates— and in a very convoluted way. say free cash ‡ ows. the value of a sequence of uncertain future cash ‡ ows. Hence. this needs not to be the case. and Brennan and Xia. KF t 1. because the starting assumption is a model for how market risk premia and how the return betas for the whole sequence of cash ‡ ows evolve over time.reviewed above do not satisfy this principle. the date-speci…c market price of risk.e. are date-speci…c (like zero-coupon rates). then the risk-adjusted discount rate re‡ ects the risk in the operations from now until perpetuity. However. 1997. Hence. Vt = KF t t =t+1 (17) where the risk-adjusted discount rates. and the date-speci…c systematic risk of the cash ‡ where the latter is determined not ow. this might in fact not be the case. They propose a model in which a term structure of required rates of return is used to discount future expected payments (see also Brennan. the information in the current directly observable term structure of interest rates is not used— the date-speci…c risk-adjusted discount rate is some complicated mixture of the zero-coupon interest rate for that date. i. Obviously. necessarily. the risk in the ow ect ow whole sequence of future cash ‡ ows.

q ] . This latter object is the Radon-Nikodym derivative for the two probability measures. In order to sketch the approach. The approach described in the following sections aims at separating out these various e¤ects on the current value of some future uncertain cash ‡ (or residual operating inow come/abnormal operating income growth). i. Hence. the discounting for time and the adjustment for risk are separated— the risk-adjustment is made in the numerator. Obviously. and such that the value of a sequence of future cash ‡ ows can be determined as a sum of the values of the individual cash ‡ ows. Vt (ff cf g >t ) = 1 X Vt (f cf ): =t+1 This requires that we go back to the fundamentals of asset pricing in multi-period settings in which we determine the risk and the values of the individual future cash ‡ ows.e. These relations are the basis for the equivalence of the discounted free cash ‡ ow and accounting-based valuation models.properties of cash ‡ ows and market returns. or the so-called valuation index. The section concludes with the determination of 21 . (1 + t ) t where t is the zero-coupon interest rate at date t for maturity . and the discounting for time is done through the zero-coupon interest rates in the denominator. we can use the well known result that no-arbitrage in the …nancial market implies that the current value of a future cash ‡ f cf is given as ow Vt (f cf ) = Et [f cf ] + Covt [f cf . and how fundamental risks associated with future cash ‡ ows are priced in the market through the valuation index. We then introduce the basic accounting relations for …nancial statement articulation within and across periods. 3 Multi-period Asset Pricing Theory and Accounting Relations In this section we …rst provide a short review of multi-period asset pricing theory starting with the implications of assuming no-arbitrage in the …nancial markets.. This requires that we specify the stochastic process for the future cash ‡ ows (but not the returns which depend on future prices). the task at hand then is to determine the risk-adjustment in the numerator— the zero-coupon interest rates are almost directly observable. and q is the ratio between the risk-neutral probabilities based on zero-coupon bonds (the so-called forward probabilities) and the true conditional probabilities.

except that we assume a continuous state space (in order to allow for continuous distributions such as normal and log-normal distributions). Theorem 1 No-arbitrage in the securities market (and mild regularity conditions) implies that there exists a strictly positive event-price measure P (y jyt ) such that the ex-dividend price at date t of any marketed security can be written as Vjt (yt ) = T X Z =t+1 dj (y ) dP (y jyt ). That is. 3. T denote future dates. J: (18) The linearity of the pricing functional is evident at this point. 2003.9 Date t will throughout denote the current date at which the valuation is performed. :::. . and denote probabilizable events observable at date by y 2 . the dividend dj and the ex-dividend price Vj of security j = 1. issues of asset pricing in in…nite horizon settings. This is the value additivity principle in perfect (although not necessarily complete) …nancial markets. One set of representations normalizes the event-prices. but in all cases such that value additivity principle is preserved. As is well known. the fundamental no-arbitrage asset pricing relation in (18) can be given alternative representations..T . t + 1. Trading and consumption take place at dates = t. The following analysis re-expresses this fundamental asset pricing result in several di¤erent ways. J at date can be written as functions of the observable events at that date.1 Fundamental Theorem of Asset pricing Our model closely follows the discrete-time model in Christensen and Feltham. Hence. and dates = t + 1. :::. respectively. i.equilibrium valuation indices in e¤ectively dynamically complete markets with time-additive utility and homogenous expectations.e. j = 1. ) with homogeneous investor beliefs be given. mainly technical.:::. dj = dj (y ) and Vj = Vj (y ). Dividends and ex-dividend prices of marketed securities are stochastic processes adapted to the …ltration = f g =t. :::. The fundamental theorem of asset pricing can now be stated as follows. The event-price measure applies to any stream of marketed future dividends. the prices of securities and portfolios of securities are determined as a “summation”of its date-event speci…c dividends times the date-event speci…c event prices. Let the standard probability space ( . T conditional on public information . and another set of representations re-expresses the dividends in terms of free cash ‡ ows or We assume a …nite horizon setting here in order to avoid additional. and summations over future dates and integrals over the future events are linear operators. 9 22 . :::. Chapters 6 and 9.

The following theorem states alternative representations of the …rst type. it follows that Et [q t ] = 1. 10 23 . for example. B t . =t+1 where B t is the price at date t of a zero-coupon bond paying one unit of account at date . T such that (19) B t EQt [dj ]. T such that (20) Et [ dj ]: =t+1 (c) There exist valuation indices q Vjt = t for dates = t + 1. we will in the following suppress the dependence of. T such that (21) T X =t+1 B t fEt [dj ] + Covt [dj . q t ]g . while in (b) the event-prices are normalized by the conditional densities. where the valuation indices are given by the Radon-Nikodym derivative between the probability measures Q t and . In the following. In (c) the eventprices are normalized by both the discount factor and the probabilities and. (b) There exist event-price de‡ators for dates Vjt = T X = t + 1. and EQt [ ] is the conditional expectations operator under Q t . :::. qt= dQ t (y jyt ): d (22) In (a) the event-prices are normalized by the riskless discount factor.accounting numbers using accounting relations. :::. :::.10 Theorem 2 The fundamental asset pricing relation (18) is equivalent to each of the following: (a) There exist forward probability measures Q Vjt = T X t for dates = t + 1.e. furthermore. prices on the speci…c events yt whenever no confusion should be possible. d (y jyt ). we use these equivalent formulations of asset prices interchangeably.. For notational simplicity. i. for the period from t to (in order to transform the event-prices into probability densities).

Chapter 9. 2003.. 1 [fi ]= 1 fa 1: 24 .e. for further discussion of these relations). ni is net income in period .e. we focus on a particular …rm and.. and these assets are increased by …nancial income and free cash ‡ ows denoted f cf . bvT = 0. T (see also Christensen and Feltham. fa = fa 1 + fi + f cf d + : (C) Financial Assets Marked-to-Market (FAM): The risk-adjusted expected …nancial income equals the riskless spot interest rate (denoted 1 ) times the opening book value of the …nancial assets.. bv = bv 1 + ni d + . Following Feltham and Ohlson (1999) we assume the following accounting articulation relations of future income statements and balance sheets = t+1. i. i.2 Accounting Relations As noted above. ni = fi + oi : (B) Financial Asset Relation (FAR): All transfers to common equityholders are made through the …nancial assets. d is the net-dividend paid to equityholders prior to closing the books at date . i. (A) Clean Surplus Relation (CSR): All changes in the book value of equity except transactions with common equityholders go through the income statement.e. we drop the subscript j on asset prices in the remainder of the paper. and is the market value of issued equity at date through anticipated equity transactions. therefore. E Q .e. the set of alternative representations of no-arbitrage prices of equity claims of the second type builds on a set of accounting relations. bv = fa + oa . where bv is the book value of equity at date . :::.. i. For notational simplicity. the book value of equity and net income are separated into …nancial and operating activities.3. Furthermore.

We are not going to invoke the Modigliani-Miller assumptions of capital structure irrelevance. while it is not important that past …nancial statements satisfy this relation. if past reformulated …nancial statements are used to forecast future statements. then consistency would imply that these past statements are also on a comprehensive income basis. so that the relative prices can be determined on a before tax basis (as it is common in standard approaches to equity valuation). Note that we are using the same set of risk-adjusted probabilities Q t to price the before-investor-tax dividends on all claims including both …nancial and equity claims.e.. which are marked-to-market on the balance sheet and earn a competitive risk-adjusted expected return as implied by (C). implying that corporate taxes (or corporate tax savings) on …nancial income (or net …nancial expenses) are part of operating income. there is an implicit assumption here that all claims are taxed equally in the hands of investors. Our approach is similar to the so-called adjusted-present-value If di¤erential personal taxation on equities and bonds is viewed as an important institutional feature.4). Hence. The issue of how to consistently account for corporate tax savings of …nancial debt is central to practical equity valuation.(D) Operating Asset Relation (OAR): The operating assets are increased by operating income and reduced by the free cash ‡ ows transferred to the …nancial assets. 11 25 . …nancial income is income from the …nancial assets before corporate taxes. This is di¤erent from the WACC-approach in which tax shields of interest payments on net debt are part of …nancial income. Of course. Furthermore. In that approach.11 Therefore. Financial assets (and the associated …nancial income) are those assets. oa = oa 1 + oi f cf : (A) is the well known clean surplus relation.” since our objective is to value common equity. Note that it must hold for all future dates. any recognition of deferred taxes (or deferred tax bene…ts) is classi…ed as an operating liability (asset). i. a quick …x is to mark down the zero-coupon interest rates used in (21) by the relative personal tax disadvantage on equities compared to bonds. then the accounting must be “super-clean. If there are anticipated equity transactions (such as outstanding executive stock options). The assets are divided into operating assets and …nancial assets. Assets not satisfying these conditions are classi…ed as operating assets. the tax bene…ts of debt are re‡ ected in a lower required rate of return used to discount the free cash ‡ ows from operations or residual operating income from operations— a somewhat strange idea. i. 2003.e.. the equity concept is “pure-equity”(see Christensen and Feltham. Section 9. which states that the net income concept is comprehensive income.

the risk-adjusted expected residual …nancial income is equal to zero. we de…ne spot residual income for the …nancial and operating activities as income minus the riskless spot interest rate times the opening book value. additivity of these concepts in the sense that ri = rfi + roi requires that all assets including the operating assets are marked-to-market (which is obviously not the case).t 1+ 1+ 1 t 1+ as s 1: 1 t: De…ne spot abnormal operating income growth in period aoig st = rois s 1. Note that with our classi…cation of all taxes as part of the operations. in which the tax advantage to debt is valued separately as an operating asset potentially generating value (see Section 2. While the residual income models essentially work with the price-to-book ratio. (b) an after-tax cost of capital for residual …nancial income.t . ris = ni rfis = fi rois = oi 1 bv 1 fa 1 oa 1. (D) and (B). 1: Note the di¤erence here compared to the standard approach of using risk-adjusted capital charges for calculating residual income: (a) an equity cost of capital is used for residual income. In our de…nition of the three residual income measures. In addition to these accounting relations.. 1.(APV) technique. namely the price-earnings ratio. Furthermore. 1. be de…ned as s 1.2). and (c) a weighted average cost of capital for residual operating income. note that by the FAM relation. the capital charge on opening book values of all assets is the beginning-of-period before-tax riskless spot interest rate. the free cash ‡ is equal to operating cash minus cash investments in operating assets (as usual) ow minus total paid taxes. the abnormal income growth models work with another common multiple. Let the spot yield-spread between the spot riskless interest rate for some future period and the current spot riskless interest rate. Using these approaches. The free cash ‡ f cf is like a dividend from the operating assets to the …nancial assets.t roi 26 .t 1+ s 1. The abnormal operating income growth model is the latest addition to the arsenal of accounting-based equity valuation models. ow cf. and it is equal to the cash ‡ distributed to the …nancial assets net ow of payments to common equity holders.e. i.

and that expectations are calculated using the forward risk-adjusted probabilities. s roi 1 The following theorem establishes the relations between the current market value of common equity St = Vt + f at . Using the de…nition of the valuation indices. and Ohlson and Juettner-Nauroth. and future accounting numbers (see also Feltham and Ohlson (1999) and Christensen and Feltham. CSR. which is the left-hand side of the value relations in Theorem 3. the risk-adjusted expectations can also be represented as the non-adjusted expectations plus the covariances with the valuation indices. and OAR are su¢ cient for the following free operating cash ‡ow and spot accounting-value relations Vt = T X B t EQt [f cf ] = T X =t+1 T X B t [Et [f cf ] + Covt [f cf . 2003.t [oit+1 ] + Q Rt B t EQt [aoig t ] (25) =t+2 where the risk-adjusted probabilities and the valuation indices are the same as in the dividendvalue relations in Theorem 2. the spot abnormal operating income growth in some future period is positive if. FAM. risk-adjustments are made in the numerator as opposed to in the discount rates in the denominator. 12 All proofs are given in Appendix A. Given mark-to-market accounting of the …nancial assets. T # (24) " =t+1 =t+1 Et+1. Hence. q t ]] BT t (yt ) EQt [roiT ] . FAR.4. 27 . Propositions 9. and only if. The key di¤erences are that the calculation of capital charges and discounting (and the capitalization of the net present value in (25)) are performed using before-tax spot and zero-coupon interest rates. future free cash ‡ ows.This is equivalent to the usual de…nition of abnormal operating income growth except for the spot yield-spread (and the fact that residual operating income is based on spot interest rates). These value relations appear like the similar relations in the standard approaches calculation net present values by discounting expected cash ‡ ows or accounting numbers by risk-adjusted discount rates.12 Theorem 3 No-arbitrage. the growth in spot residual operating income exceeds the yield spread (for rois 1 > 0): rois s > s 1.3 and 9.t : aoig (y ) > 0 . q t ]] T X (23) =t+1 = oat + = 1 t B Q s t E t [roi ] T X = oat + B t [Et [rois ] + Covt [rois . 2005). That is. the task in equity valuation is the determination of the value of operations.

in principle. however. = t + 1. These are easily estimated using current prices of traded treasury bonds and your favorite term structure model. at the valuation date t are the prices of zero-coupon bonds at the valuation date.. these covariances may interact with the risk-adjustment terms. oa 1 ]] : 28 . such as Covt [rois .13 The book values of the …nancial and operating assets are also readily observable from (reformulated) …nancial statements. 0=B 1.e.t FR 1. s Et [roi ] = Et [oi ] = Et [oi ] 13 Et [ [Et [ 1 oa 1] 1] 1 ] Et [oa + Covt [ 1 . B 1.t 1 + fr 1. T: (26) F R 1.t for period is de…ned as the interest rate on a contract for a single-period deposit from date 1 to . However. q t ] .3. :::. T . that future free cash ‡ ows are independent of accounting policies. in subtle ways.t : Hence. The current forward interest rate f r 1. and derive the associated accounting value relations.3 Alternative Formulations of Accounting Value Relations In Theorem 3.t = Bt Note that. in the calculation of expected residual operating income or expected abnormal operating income growth. we do not need a fully dynamic model of the stochastic process for interest rates— we only need to calibrate to the current term structure of interest rates.. which may “hide”the full consequences of changes in accounting policy on risk-adjusted expected future accounting numbers.e.t B t . These relations also highlight the fact that it is the underlying free cash ‡ ows and their time-series properties which are determining the risk-adjustments. the covariances with the future spot interest rates may be a¤ected by the choice of accounting policy. which implies that the contract has a value of zero at the current date t. and the covariance between spot interest rates and book values of operating assets or operating income. moreover. i. 14 For example.t . Note. :::. moving from historical cost towards fair value accounting for some operating assets. we need estimates of both the expected future spot interest rates. FR 1. the discount factors B t . for example.14 These terms are not usually estimated in standard security analyses and. we rede…ne residual operating income and abnormal operating income growth using current forward interest rates as opposed to future spot interest rates. = t + 1. Expected future free cash ‡ ows are outcomes of a strategic and …nancial statement analysis of the …rm. Hence. the current forward interest rates are readily observable from the current zero-coupon prices. i. In this section.

15 see also Christensen. of course. the forward accounting-value relations also highlight the fact that accounting policy choice has no impact on risk premia (given that the valuation indices are una¤ected).t . there is no need to forecast future spot interest rates and how they are correlated with future operating income and operating assets. q ] T X (27) BT t Et [roif t ] T # (28) " =t+1 Et [oit+1 ] + T X Rt+1. de la 15 Of course.t t 1 + fr 1.t roi f 1. for example. From a theoretical point of view. q ]: =t+1 One advantage of these forward accounting-value relations from an implementability point of view.t t: We can now restate the accounting value relations in Theorem 3 as follows. forecasts of operating income and operating assets.e. where the current forward yield-spread is de…ned as f 1. Hence. and then. Theorem 4 No-arbitrage.t 1 + fr 1+ 1. CSR. not equal to expected future spot interest rates (unless. in general.t oa 1. FAM.t B t Et [aoig t ] f =t+2 + B t Covt [f cf . Now. respectively.. That is. the net present value of the risk-adjustments to future free cash ‡ ows. since the future free cash ‡ ows are independent of the accounting policy choice. of course. since the ex-ante …rm value can also be determined as the 29 . this is a somewhat trivial result. FAR. is that they only require an estimation of the current term structure of interest rates. and OAR are su¢ cient for the following forward accounting-value relations Vt = oat + = 1 t T X B t f Et [roi ] + Covt [f cf . oi fr 1.t 1+ f 1. the latter are deterministic). i. a future change in accounting policy to earlier recognition of information about future cash ‡ ows (with. fair value accounting) has no impact on risk premia (nor on the ex-ante value of the …rm). current zero-coupon bond prices from which the forward rates can be derived using (26).Note that forward interest rates are. rede…ne forward residual operating income and forward abnormal operating income growth to be roif t and aoig ft roif t f 1.

Since we (in the current version of the paper) do not investigate the impact of changes in accounting policies. the operating asset relation (OAR) can be used to substitute out the free cash ‡ ows. there is an additional risk-adjustment component in (29) if forward residual operating income is also used for risk adjustments.t oa 1.4 Equilibrium Valuation Indices The remaining item in the value relations is the risk-adjustment due to the covariance with the valuation index. free cash ‡ ow. on the spot residual operating income relation (24). A disadvantage of the forward accounting-value relations is that they “mix”future accrual accounting numbers with free cash ‡ ows. net present value of future free cash ‡ ows. q ] . 3. and abnormal income growth relations. no-arbitrage alone does not tell us much about the valuation indices. the forward residual operating income relation (27) can be restated as follows Vt = oat + = oat + = oat + T X B t f Et [roi ] + Covt [oi (oa oa 1) . q ] i Covt [aoag ft . we will in the following (also for notational ease) perform our analysis on the basis of the spot accounting-value relations and. compared to the spot residual operating income relation (24). Unfortunately. We will assume that equilibrium prices are formed in an e¤ectively dynamically complete market. q ] =t+1 B t =t+1 h f f Et [roi ] + Covt [roi . For example.Rosa. All we know is that they are positive and have expected values of one— at this point they are merely mathematical constructs without economic content (except they are derived from an assumption of no-arbitrage). in particular.t oa 1: Hence. q ] =t+1 T X T X B t f Et [roi ] + Covt [oi fr 1t oa 1 oa + F R 1. 30 . Of course. Similar analyses can be performed using any of the other dividend-. (29) where forward abnormal operating asset growth is de…ned as aoag ft oa oa FR 1 1. and Feltham (2008). There are various ways of getting more information about the valuation indices.

u0it (cit (yt )) are given by (30) = t + 1.By this we mean that there are su¢ cient trading opportunities for the investors to trade to a Pareto e¢ cient risk sharing. for example. i = 1. and the references therein). i. Hence. or abnormal operating income growth with the “scarcity” of aggregate consumption. Assume investors have homogeneous beliefs. (a) Individual equilibrium consumption plans are measurable with respect to aggregate consumption. i = 1. residual operating income. I: (31) This theorem is the key result in consumption-based asset pricing models. I. but rather use a general equilibrium approach.. T . The …rst-order condition for the investors’decision problems (with time-additive utility) implies that each investor’ valuation index is determined by the investor’ marginal utility of consumption. the risk-adjustments in the preceding value relations relate to the covariance of dividends. xt . the Fama-French portfolios as factors (see Nekrasov and Shro¤. free cash ‡ ows. and cit (xt ) is an increasing function of xt . de…ned on Ci = [ci . cit = cit (xt ). s s The additional assumptions of homogeneous beliefs and e¢ cient risk sharing imply that all investors’valuation indices are perfectly aligned and measurable with respect to aggregate consumption. This leads to the following result (see. :::. uit (cit ). Chapter 6. Theorem 5 Let an equilibrium in an e¤ectively dynamically complete market be given. for cit # ci . i. and di¤erentiable. :::. Moreover. :::. time-additive utility functions of event-contingent consumption. 31 .e. for example.. at each date. strictly increasing. 1). q t (x jyt ) = u0i (ci (x )) .e. E[u0i (ci )jyt ] = t + 1. i = 1. We are not going to take that route. (b) The prices of zero-coupon bonds paying one unit of account at date B t (yt ) = E[u0i (ci )jyt ] . :::. :::. Christensen and Feltham. But how can we empirically measure the “scarcity” of aggregate consumption? One approach is to make an ad hoc assumption that the valuation index is given by a linear factor structure using. T . 2003. and concave. I: (c) The valuation indices based on zero-coupon bonds are measurable with respect to aggregate consumption at each date and given by the“scarcity” of aggregate consumption as measured by the investors’marginal utility of consumption. 2006). we will assume investors have time-additive utility functions and homogeneous beliefs. where u0it (cit ) ! 1.

T . the following corollary indicates the key ingredients of an analysis in which the prices of nominal zero-coupon bonds are used as the numeraire (instead of the real prices of real bonds). (a) Individual equilibrium consumption plans are measurable with respect to real aggregate consumption. we do not develop a full analysis of the impact of uncertain in‡ ation.. In the current version of the paper. some stocks may be a better hedge against in‡ ation than others. T . :::. i = 1. i = 1. q N (x t u0i (ci (x ))=p . Moreover. T . = t + 1. :::. = t. cit = cit (xt ). at each date.e. :::. :::. If in‡ ation is stochastic. t + 1.e. we should note that the zero-coupon bond prices in (30) and the valuation indices in (31) are given in “real terms. :::. i = 1. p jyt ) = E[u0i (ci )=p jyt ] = t + 1. :::. . Let p denote the price level of the consumption good at date . in practice. and cit (xt ) is an increasing function of xt . (b) The nominal prices of zero-coupon bonds paying one unit currency at date by E[u0i (ci )=p jyt ] . 16 32 . because the set of traded real bonds is very limited and. because.. respectively.e. i. secondly. forecasts of future accounting numbers and free cash ‡ ows are in nominal terms.” i. the units of accounts are units of the consumption good. this is a problem empirically— …rstly. such that the real prices of securities and the real dividends are given as V N =p and dN =p . I. thus. it may not be possible to estimate the real term structure of interest rates reliably from traded bonds and. q N ] : t t =t+1 We present this result without proof— it essentially follows from the investors’ …rst-order conditions recognizing that their preferences are de…ned over units of real consumption. I: (d) The nominal prices of securities are given as VtN = T X B N Et [dN ] + Covt [dN . xt . However. I: B N (yt ) = t u0it (cit (yt ))=pt are given (c) The nominal valuation indices based on nominal zero-coupon bonds are measurable with respect to the pair of real aggregate consumption and the price level at each date. i..16 Corollary 6 Let an equilibrium in an e¤ectively dynamically complete market be given (as in Theorem 5).Before closing this section.

Of course. more securities may be needed to ensure that investors can achieve a Pareto e¢ cient allocation— they need to be able to insure in‡ ation risk. there are two sources of the risk-adjustment to expected nominal dividends: the covariance with the scarcity of real consumption. using the accounting relations. 1976. the nominal dividend value relation can be re-expressed to obtain the accounting value relations as above. p which is the standard “Fisher relation”for the relationship between nominal and real interest rates. Standard distributions in asset pricing models are normal distributions and lognormal distributions. Of course. then the prices of nominal bonds will not only re‡ the expected in‡ ect ation but also include an in‡ation risk premium due to the covariance between the marginal utility of real consumption and the purchasing power of one unit of nominal currency. we will not pursue these issues much further in this version of the paper. First. Hence. One the other hand. we will include a few footnotes describing the necessary adjustments to the analysis in order to account for stochastic in‡ ation. and a term which re‡ ects the nominal dividends’ ability to hedge in‡ ation risk. 4 An Accounting-based Multi-period Equity Valuation Model Identi…cation of the valuation index requires more assumptions on distributions and (or) preferences. the nominal valuation index used to calculate the nominal risk premia for risky nominal securities not only re‡ ects the scarcity of real consumption but also the purchasing power of nominal dividends. i. the presence of uncertain in‡ ation does not a¤ect the allocation of real aggregate consumption in an e¤ectively dynamically complete market. which is (a). However. then B N (yt ) = B t t pt . As noted above. we need a mathematical result for normal distributions known as Stein’ Lemma (see Rubinstein. 1=p .e. s Lemma 7 If X and Y are jointly normally distributed random variables and F ( ) is some 33 . for a proof). If the price level is deterministic. Note that the no-arbitrage value relations shown in the preceding sections do not rely on which unit of account we are using— it is only when we wish to identify the equilibrium valuation index in terms of real consumption that we need to be speci…c about the units of account. However.. Similarly. if in‡ ation is stochastic.

Note that a lognormal distribution would not be a good choice for future residual operating income. we stick with the normal distribution in the text. we must make the additional assumption that the nominal valuation index aggregates these two objects into a normally distributed random variable. If in‡ ation is stochastic. most …nance researchers’ favorite utility function. In this case. This requires assumptions about not only the distributions of aggregate consumption and the price level. we get that the price of this claim is Vx =B t o n 0 Et [x ] + Et [q t (x )] Covt [x . 18 Here we explicitly use that the valuation index is a function of a normally distributed random variable namely aggregate consumption. x ] (32) Recognizing that the valuation indices in the dividend-value relation (21) and the accountingvalue relation (24) are the same. since these income numbers may very well be negative (which is precluded by the lognormal distribution). Instead we could assume that aggregate consumption is lognormally distributed and write the risk-adjustment in (24) as Covt [roi . The general equilibrium analysis with lognormally distributed aggregate consumption and power utilities is worked out in Appendix B. F (Y )) = E[F (Y )] Cov[X. the power utility function. only the negative exponential utility function works (see the following section). h t (z )]. If we de…ne aggregate consumption returns and expected aggregate consumption returns as Rxt x . Vx t Normally distributed aggregate consumption does not go well with standard utility functions de…ned only on positive consumption— really. but also about the utility function. Vx t R x t Et [x ] . where h t (z ) = q t (exp(z )) and z = ln(x ). then the nominal valuation index is only measurable with respect to the pair of aggregate real consumption and the price level. Hence.di¤erentiable real valued function with E[jF (Y )j] < 1. in a nominal analysis. x ] . and then proceed in a similar manner from there. q t (x )] = Covt [roi . t 0 Et [q t (x )] = V x R t Et [x ] t . 17 34 . then Stein’ Lemma and the accounting-value relation (24) s imply that17.18 Vt = oat + T X 0 0 =t+1 n o 0 s s B t Et [roi ] + Et [q t (x )] Covt [roi . would be the choice for similar general equilibrium analyses as in the subsequent sections. Y ]: If we assume that future spot residual operating income and aggregate consumption are jointly normally distributed. We will provide such assumptions in the nex section. For simplicity. and that a claim on aggregate date consumption can be priced using (21). then Cov(X. Vart [x ] where R t B t1 is the riskless return from date t to .

and that Et [jq t (x )j] < 1. Beaver. T: Except for the term structure aspect it is interesting that we have “accounting betas. Kettler and Scholes. Vart [Rxt ] = t + 1. for example. ReN OA t : ReN OAs t oat oat Dividing by the current book value of operating assets in (33) yields the following result. 1988a and 1988b. and that these excess returns are the same for all maturities (as in applications of the standard CAPM). ). 2001). i. Rxt ] Vart [Rxt ] : (34) There are two unusual objects in this value relation. Proposition 8 Assume the conditions in Theorem 5 hold and make the accounting assumptions (A)-(D). :::. assume that future spot residual operating income and aggregate 0 consumption are jointly normally distributed. for example. we can de…ne residual income returns on net operating assets by normalizing future spot residual operating income by the current book value of operating assets (like in “trend analyses” i. aggregate earnings on stock indices are used while we are using aggregate consumption returns de…ned by market values of aggregate consumption claims. there is signi…cant empirical evidence that these excess returns are timevarying and predictable to some extent (see. Covt [ReN OAs t . Rxt ] R t] Vart [Rxt ] (33) Similarly. Campbell and Shiller.. Then the marketto-book ratio for the operating assets is given by T X Vt B =1+ oat =t+1 t ReN OA s t [R x t R t] Covt [ReN OAs t . Furthermore.”In the early seventies of the previous century there was a literature examining accounting betas (see. x R t R t. The second unusual object is the term structure of systematic accounting risk. these are easily determined if one assumes that the expected return on a well diversi…ed stock index is a good proxy for expected aggregate consumption returns. One is the term structure of excess returns at date t.19 The advantage of this approach is that we In that literature. 1970).e. However. T: Of course.then we can rewrite (32) as Vt = oat (yt ) + T X B t Et [roi ] s [Et [Rxt ] =t+1 Covt [rois . rois s Et [rois ] . 19 35 . and Lettau and Ludvigson.e. Rxt ] ..e. = t + 1.. :::. i.

Then there are parameters vit and fit such that cit (xt ) = fit + vit xt . Both of these unusual objects in equity valuation require new empirical work (which is not a comparative advantage of the authors’ It is an empirical issue whether Proposition ). Proposition 8 tells us what are the elements we should be looking for in order to have an equity valuation model consistent with classical multi-period asset pricing results. 5 An Accounting-based Multi-period Equity Valuation Model with Exponential Utility In this section we make additional assumptions on investor preferences in order to get more insight into the determination of the term structure of excess returns.6) show the following result. A possible solution might be to estimate “industry betas” and then use those. PI i=1 fit = 0 Each investor’ fraction of aggregate consumption vit is the same for all dates. 8 is a superior equity valuation model compared to a standard valuation approach using risk-adjusted discount rates. i = 1. and interest rates. Proposition 6. Note that the accounting betas we need are a¤ected neither by leverage (which is one of the curses of tests of asset pricing models) nor by scale.avoid the circularity in the estimation of betas using stock returns. :::. s 36 . PI and i=1 vit = 1. However. the drawback is that we have much less data available to estimate the accounting betas. I. Hakansson (1970) and others (see also Christensen and Feltham. Here it may be useful to assume a particular stochastic process for spot residual operating income returns and aggregate consumption returns and then derive the term structure from there (see below). Theorem 9 Let a Pareto e¢ cient equilibrium be given and assume the investors have homogeneous beliefs and time-additive preferences represented by HARA utility functions with identical risk cautiousness. In any case. 2003. vit = vi . The other interesting aspect is the term structure aspect. accounting betas..e. i.

then the price level at date is log-normally distributed given information at date t. T : uit (cit ) = uit (cit ) = uit (cit ) = P t P t P it exp[ cit = i ]. 4. 1 P it P t > 0. cit P t . I and for all t = 1.7) Assume investors have negative exponential utility and. 1 [ cit bit ] 1 > 0. 37 . That is. Although this may seem a minor change in the analysis. the analysis can proceed with the in‡ ation-adjusted normally distributed random variable z substituted for real aggregate consumption x . Stochastic in‡ ation is an additional factor. :::. Proposition 6. Wilson (1968) shows that in this case vit = vi = i = o (see also Christensen and Feltham. :::. and only if. If we assume that the price level p is determined as a geometric Brownian motion such that Z Z p = pt exp p dBs . it may have a very substantial impact on empirical implementations of the model. :::. It then follows from (31) that the valuation index is given by20 q t (x jyt ) = exp[ x = o ] . p ds + t t 20 where p and p are the instantaneous expected in‡ ation rate and in‡ ation volatility. Et [exp[ x = o ]] y yt . a market portfolio and a complete set of zerocoupon bonds are su¢ cient if personal endowments are also spanned (see Christensen and Feltham. which may help explaining observed nominal risk premia— standard asset pricing tests implicitly assume deterministic in‡ ation. = t + 1. 2003. Et [exp[ z = o ]] ln [p =pt ] : Hence. ln[p =pt ] is normally distributed with mean p ( t) and variance 2 ( t). that their risk P aversion is time-independent with aggregate risk tolerance o = i i . i > 0. bit > 0: ln(cit bit ). cit The advantage of having constant fractions of aggregate consumption is that it is very easy to ensure an e¢ cient equilibrium— essentially. for simplicity. T: and Di¤erentiating the valuation index with respect to aggregate consumption at date taking conditional expectations yield i h 0 Et q t (x ) = Et 1 exp[ x = o ] = o Et [exp[ x = o ]] 1 o : In a nominal analysis with stochastic in‡ ation. one of the following conditions holds for all i = 1. 2003. respectively.3). p jyt ) = t where z x + o exp[ z = o ] . p using the exponential utility function. bit > 0. Proposition. It is well known that the stochastic variation in real aggregate consumption is too small to …t observed risk premia in asset prices. the nominal valuation index is given in Corollary 6. yields that q N (x . > 0.if. Inserting this into the nominal valuation index.

empirically). that it is not the risk in the future cash ‡ ows or accounting numbers of the individual securities which determines their risk-adjusted expected cash ‡ ows or accounting numbers. 2 to t. and investors have exponential utility with constant risk tolerances. in general.Inserting this into (32) yields Vt = oat (yt ) + T X =t+1 B t fEt [rois ] Covt [rois . assume that future spot residual operating income and aggregate consumption are jointly normally distributed. Furthermore. = o =I) and “aggregate consumption per capita” as acc = x =I. with Note that risk-adjusted aggregate consumption per capital is equal to the average relative risk aversion in the economy (which has an order of magniture of 2-5.. but rather how these risks are correlated with risk-adjusted aggregate consumption per capita or.21 Hence. more generally. =1+ (35) oat =t+1 and the zero-coupon prices are given by Bt= where o t Et [u0i (ci )] = u0it (cit (yt )) o t exp[ fEt [racc ] racct 1 Vart [racc ]g].22 (36) is the investors’“average personal discount factor” from In the introduction. Proposition 10 Assume the conditions in Theorem 5 hold and make the accounting assumptions (A)-(D). and all other value relations preceding it.e. Then the market-to-book ratio for the operating assets is given by T X Vt s B t ReN OA t Covt [ReN OAs t . racc ] . it is clear from (35). equal to the certainty equivalents of the cash ‡ ows or accounting numbers as certainty equivalents are normally de…ned in the economics of uncertainty. we pointed to the fact that the equilibrium risk-adjusted expected cash ‡ ows or accounting numbers as in (35) are not. the wealth level which makes the decision maker indi¤erent between receiving this level of wealth with certainty and retaining the uncertain wealth. 22 If we let the investors’personal discount factors be de…ned as P exp [ i t] . x = o ]g If we de…ne as the investors’ average risk tolerance (i.e.. First. normalizing by the book value of operating assets we get the following result. then the average personal it discount factor is given by P o t) i i i = o ] : t = exp [ ( 21 38 . then the risk-adjusted aggregate consumption per capita is racc = acc = = x = o . i.

However. but we have chosen a simple one in order to more clearly focus on the central issues. Of course. racco (1 + ) t 1 t ! r 2 [0. and with conditional variances and contemporaneous conditional covariance: Vart [" ] = 2 .the valuation index. (37a) (37b) where " and are zero-mean normally distributed and serially uncorrelated. ! a 2 [0. Solving these equations recursively yields a . ReN OA t = s ReN OAo (1 t racco (1 t + ) + ) t + !r t [ReN OAs tt ReN OAo ] t + 1 Xt s=0 + 1 Xt s=0 s: !s " r s. i. but it might still be the case that the equilibrium risk-adjusted expected cash ‡ ows on optimally invested wealth would be equal to the certainty equivalent of optimally invested wealth.t t t 1 t = ! r [ReN OAs racco (1 + ) t ReN OAo (1 + ) t = ! a [racc 1. Since this question and its answer are bit of an aside. ]+ . we refer the discussion to Appendix D. 1). this is well known. 23 t and racco (1 + ) t t with growth rate 39 . Covt [" . the trends for the two processes are ReN OAo (1 + ) t and . Vart [ ] = r 2 ] = ra .e. 6 A VAR Model with Exponential Utility In order to get additional insights we now consider a simple vector-auto-regressive (VAR) model of the stochastic properties of spot residual operating income for a particular …rm and risk-adjusted aggregate consumption per capita.. under the assumptions of this section.23 ReN OAs t racc ReN OAo (1 + ) t 1. racc = t + ! a [racct t racco ] t !s a That is. we show in Appendix D that the riskadjustment in the equilibrium price of optimally invested wealth is twice as big as the risk premium in the associated certainty equivalent. Assume the spot residual operating income returns and the risk-adjusted aggregate consumption per capital are the only information available and that they follow a …rstorder vector auto-regressive process with mean-reversion to a deterministic exponential trend. The normally distributed aggregate consumption and exponential utility assumptions make it possible to answer this question. respectively. More comprehensive models can be assumed.t ]+" . 1).

in general. racc ] = s 1 ra (! a ! r ) 1 !a!r t : (38) Note that the risk-adjustment. the risk-adjustment is ra independently of maturity date . ! r and ! a . it can be estimated based on a time-series of average excess returns (probably giving rise to an equity premium puzzle and many other empirical problems). The important lesson from (38) is that. RAs t . is an increasing function of an upper limit of 1 (! a ! r ) t 1 ! ra : ra 1 !a!r 1 !a!r (for ! a ! r > 0). then the risk-adjustment is due solely to the contemporaneous correlation with risk-adjusted aggregate consumption. i. The reason for this latter result.. but with (39) Moreover. The obvious alternative is to determine it implicitly from a calibration to the readily observable current term structure of interest rates using the zero-coupon prices in (36)— in a general equilibrium. i. t 2[ t] 2[ t] + ! a t [racct !a . is that no new information about ReN OAs t is revealed until date — the claim to ReN OAs t is like a zero-coupon bond maturing at date 1 plus a one-period claim from 1 to with risk ra . the risk tolerances determining zero-coupon prices are the same as those determining excess returns on equities! Once the 40 .e.This speci…cation of the information dynamics allows us to calculate all the terms in the exponential utility accounting-value relation (35) explicitly. but also on their time-series properties. and it enters into the de…nition of risk-adjusted aggregate consumption per capita. Note that the investors’average risk tolerance is not directly empirically observable. ra . of course.. the risk-adjustment depends not only on the contemporaneous conditional covariance between ReN OAs t and risk-adjusted aggregate consumption.e.. There are various ways of estimating the average risk tolerance. Proposition 11 The information dynamics in (37) imply that ReN OA t = ReN OAo (1 + ) t s t + ! r t [ReN OAs tt 21 r ReN OAo ]. t Vart [ReN OAs t ] = o Et [racc ] = racct (1 + ) t !r . ! r = 0. if spot residual operating income is serially uncorrelated. For example.e. 1 !2 r racco ]. i. 1 !2 a Vart [racc ] = and that the date 21 a risk-adjustment in (35) is given by RA s t Covt [ReN OA t .

allowing for a general term structure of interest rates. these parameters may alternatively be backed out from the current term structure of interest rates. independently of the maturity. We are not aware of empirical analyses determining the preference parameters and the parameters in the stochastic process for aggregate consumption from an observable term structure of interest rates. In Appendix C we consider a setting with preferences exhibiting habit formation. Of course. is a solution to the equation B t ReN OA t = B t [ReN OA s s t RATtB ]: Solving this equation for the implicit risk-adjustment yields RATtB = ReN OA s t " 1 1+ t 1 + t + rp t # : (40) 41 . Hence. excess return times beta) typically estimated on the basis of short-term stock returns. the risk-adjusted discount factor has the form B t = [1 + t + rp] ( t] ( t) t) < B t = [1 + . it is an empirical issue whether this approach works better than the standard time-series approaches. the comparable implicit risk-adjustment for maturity date with this approach. the parameters in the process for risk-adjusted aggregate consumption per capita can be estimated using a time-series of aggregate consumption per capita.. in our comparison we will take the liberty to interpret the textbook approach as just assuming a constant risk premium. comparing to (35). This allows for more ‡ exibility in the types of equilibrium term structures of interest rates. 7 Comparison to the Standard Textbook Approach The standard textbook approach uses a constant risk-adjusted cost of capital (WACC) to discount expected residual operating income.e. for rp > 0 and > t. Of course. Again. where t is the zero-coupon riskless interest rate for maturity as of date t. Hence. using a constant risk-adjusted discount rate which does not re‡ the current term structure of ect interest rates is obviously a very bad choice. RATtB . The present value at date t of the spot residual operating income s return on operating assets at date then has the form B t ReN OA t . and rp is a constant risk premium (i. Therefore.implicit risk tolerance is determined.

Note that our assumption. thus. 42 .25 Trend racco t 2. lowers (increases) the net present value of future residual operating income returns. Note also that even if there is high persistence in residual operating income returns. if there is low persistence in residual operating income returns. low persistence …rms will tend to be undervalued using the mium.. in particular. Secondly. ReN OA t : These latter relations highlight the fact that proper risk-adjustments should not only re‡ the contemporaneous ect covariance with risk-adjusted aggregate consumption. it should not! For example. Hence. Hence. We assume the …rm has a horizon of 25 years. rp. that the variance of future spot residual operating income returns is independent of growth. increases (decreases) the risk-adjustment and.00 Growth rate 2% Persistence ! a 90% Variance 1% 2 a Table 2: Parameters of risk-adjusted aggregate consumption. Current racct 2. 2006). a high degree of mean reversion. the risk adjustment in (38) is bounded from above as goes to in…nity (see (39)).. but also the time-series properties of residual operating income returns and. i. while the implicit risk-adjustment in (40) approaches s the expected spot residual operating income returns. increasing (decreasing) the growth rate in spot residual operating income returns . Using Propositions 10 and 11. is important for this relation. The example is based on the set of parameters for risk-adjusted aggregate consumption shown in Table 2 and a personal discount rate of 2%. then the risk-adjustment in (38) is almost independent of maturity . these parameters result in the term structure of interest rates shown in Figure 1. we try to answer this question by means of using a numerical example with reasonable parameters. i. whereas the implicit risk-adjustment in (40) compounds the constant risk pret) times. ( textbook approach (compare Brennan and Xia. note that the implicit risk-adjustment RATtB depends on the deterministic trend in spot residual operating income s returns (through ReN OA t ) and. close to being equal to the contemporaneous covariance with risk-adjusted aggregate consumption.e. and that the spot residual operating income return process in our base case is characterized by the parameters shown in Table 3. the degree of persistence of these returns. First.Comparing the implicit risk-adjustment in (40) to the risk-adjustment of the exponential utility model in (38) shows a number of striking di¤erences.e. high (low) growth …rms will tend to be undervalued (overvalued) using the textbook approach. The key problem with the textbook approach is that it does not re‡ when new ect information about future residual operating income returns is revealed! Will these di¤erences between the textbook approach and the exponential utility model yield noticeable di¤erences in equity values. obviously. or are they merely second-order e¤ects? In the following.

in the textbook approach the implicit risk-adjustment is increasing without bounds. In particular.14. The risk-adjustments for both approaches are shown in Figure 2. For the textbook approach we assume the …rm has a beta of one.8% 6% 4% 2% 0% 0 5 10 15 Maturity 20 25 Figure 1: Term structure of interest rates. Using Propositions 10 and 11. whether the textbook approach or the exponential utility model is used. and that the excess return on the market portfolio is 5%. Current ReN OAs t 15% Trend ReN OAo t 10% Growth rate 0% Persistence ! r 90% Covariance 1% ra Table 3: Parameters of residual operating income return process for base case. while the risk-adjustment in the exponential model quite quickly approaches its 43 . 8% 6% 4% 2% 0% 0 5 10 15 Maturity 20 25 Textbook Model Exponential Model Figure 2: Risk-adjustments in base case. Note that even though the market-to-book values are the same (indeed. this …rm has a market-to-book of operating assets of 2. the patterns across maturities are quite di¤erent. we have chosen the parameters for this to be the case).

15 in the 5% growth case.e. and 5% decline in spot residual operating income returns. 6% 4% 2% 0% 0 5 10 15 Maturity Textbook Model 20 25 Exponential Model Figure 4: Risk-adjustments in 5% decline case.86 and 1. and 1.64 and 3. respectively. As noted earlier the textbook approach leads to undervaluation of growth. while it overvalues declining expected spot residual operating income returns.. = 5% and = 5%. i.upper bound of (see (39)) ra 1 1 = 0:01 !a!r 1 1 0:9 0:9 = 5:26% despite the rather high persistence rates. 25% 20% 15% 10% 5% 0% 0 5 10 15 Maturity Textbook Model 20 25 Exponential Model Figure 3: Risk-adjustments in 5% growth case.63 in the 5% decline case for the textbook and exponential utility model. and the di¤erences are far 44 . The market-to-book ratios are 2. The risk-adjustments are shown in Figures 3 and 4 for the = 5% and = 5% cases. respectively. respectively. Consider now two alternative cases: 5% growth.

i. The risk-adjustments are shown in Figure 5. 2004. Then the left-hand side of (41) is positive for 45 . ! r = 10%: Compared to a market-to-book ratio of 2.91 and 2.14 for both approaches in the base case. the answer is NO! To see why. what is the risk premium for maturity .e. and ( t) is even. using a term structure of risk-adjusted discount rates (like Ang and Liu. rp t .e.. That is.25 for the textbook and exponential utility model. the textbook approach risk-adjusts the expected spot residual operating income returns far too much – due to the compounding of the constant risk premium. i. The exponential utility model risk-adjusts with an almost constant amount equal to the upper bound of ra 1 1 = 0:01 !a!r 1 1 0:9 0:1 = 1:10%: Above we have asked the question of what the implicit risk-adjustment is in the textbook approach assuming a constant risk premium. respectively. consider a case with low persistence in residual operating income returns (i. We can also ask the opposite question.e. the market-to-book ratio now becomes 1. and Brennan and Xia. 2006) which makes this approach equivalent to discounting risk-adjusted expected residual operating income returns with the zero-coupon riskless interest rates. 8% 6% 4% 2% 0% 0 5 10 15 Maturity 20 25 Textbook Model Exponential Model Figure 5: Risk-adjustments in 10% persistence case. high mean reversion). Finally.from being trivial.. Even though there is no growth. rp t is determined by the equation ReN OA t (1 + s t + rp t )) ( t) = ReN OA s t RAs t (1 + t) ( t) (41) Is there always a solution to this equation? Unfortunately. s assume ReN OA t > 0.. = 0.

e. Unfortunately. Figure 7 shows the risk premium for the other cases examined above. low market-to-book …rms).. For the base case of the numerical example examined above the risk premium is shown in Figure 6. Hence. Solving (41) with respect to rp t (using Propositions 10 and 11) yields rp t 1+ t " Note that the risk premium is increasing in the contemporaneous covariance between ReN OAs t and racc . Hence. there must be a quite substantial reduction in the risk premium for larger maturities to make this approach consistent with exponential utility model.e.e. the net present value of spot residual operating income returns at date s is positive. i. However. while it is decreasing in the growth rate in residual operating income returns. this latter condition is likely to be satis…ed for …rms in which future spot residual operating income is expected to be small (i. we can only use a term structure of risk-adjusted discount rates with great caution! In the following.. the right-hand side is negative if RAs t > ReN OA t . 10% 8% 6% 4% 2% 0% 0 5 10 15 Maturity 20 25 #1=( t) s ReN OA t = 1 s ReN OA t RAs t " ReN OAo (1 + ) t + ! r t [ReN OAs ReN OAo ] t tt t = 1 (! a ! r ) o o t + ! t [ReN OAs ReN OAt (1 + ) ReN OAt ] ra 1 ! a ! r tt r 1 =h i1=( t) 1: 1 (! a ! r ) t ra 1 ReN OAo (1+ ) t +! t [ReN OAs ReN OAo ] 1 !a !r t r tt t t #1=( t) 1 Figure 6: Risk-premium in risk-adjusted discount rate for base case.any value of rp t .. i. 46 . the net present value of spot residual operating income returns at date is negative. assume s RAs t < ReN OA t such that there is a solution for rp t .

Xia. [3] Brennan. 1-36. References [1] Ang. thus. J..” The Accounting Review 45. P. J.”The Journal of Finance 59. and M. [7] Campbell. [4] Brennan. Scholes. [2] Beaver. and J. and R.. [6] Campbell. 103-115. J. Y. M. J.. the risk premium is not de…ned. 661-676. A.”Journal of Business 79. (1977) “The Coupon E¤ect on Yield to Maturity. Earnings. (2006) “Risk and Valuation Under an Intertemporal Capital Asset Pricing Model. (2004) “How to Discount Cash Flows with Time-varying Expected Returns.. J. (1999) “By Force of Habit: A Consumption-based explanation of aggregate Stock Market Behavior. (1970) “The Association between Market Determined and Accounting Determined Risk Measures. 47 .”The Journal of Finance 43. Kettler.” The Journal of Finance 32. and Expected Dividends. Liu. (1988a) “Stock Prices. 81-90.25% 20% 15% 10% 5% 0% 0 5 5% growth 10 15 Maturity 5% decline 20 25 10% persistence Figure 7: Risk premium in risk-adjusted discount rate for noted cases. W. 205-251. Y. (1997) “The Term Structure of Discount Rates. H.”Journal of Political Economy 107. Shiller.”Financial Management 26. 2745-2783. J. Cochrane. 654-682. and Y. and J. [5] Caks. M. Note that the risk premium in the risk-adjusted discount rate is not de…ned for maturities larger than 16 years for the case with declining residual operating income returns— for these s maturities RAs t > ReN OA t and.

[8] Campbell, J. Y., and R. J. Shiller. (1988b) “The Dividend-price Ratio and Expectations of Future Dividends and Discount Factors,”Review of Financial Studies 1, 195-228. [9] Christensen, P. O., L. E. de la Rosa, and G. A. Feltham. (2008) “Information and the Cost of Capital: An Ex-Ante Perspective,”Working Paper, University of Aarhus. [10] Christensen, P. O., and G. A. Feltham. (2003) Economics of Accounting: Volume I – Information in Markets, Springer Science+Business Media, Inc. [11] Christensen, P. O., and J. Aa. Nielsen. (1986) “The Bond-type E¤ect on Yield to Maturity,”The Scandinavian Journal of Economics 89, 193-208. [12] Copeland, T., and J. Weston. (1988) Financial Theory and Corporate Policy, AddisonWesley. [13] Cox, J., J. Ingersoll, and S. Ross. (1985) “A Theory of the Term Structure of Interest Rates,”Econometrica 53, 385-467. [14] Feltham. G. A., and J. A. Ohlson. (1999) “Residual Earnings Valuation With Risk and Stochastic Interest Rates,”The Accounting Review 74, 165-183. [15] Grinblatt, M., and S. Titman. (2002) Financial Markets and Corporate Strategy, second edition, New York: McGraw-Hill/Irwin. [16] Hakansson, N. H. (1970) “Optimal Investment and Consumption Strategies under Risk for a Class of Utility Functions,”Econometrica 38, 587-607. [17] Hughes, J., J. Liu, and J. Liu. (2008) “On the Relation Between Expected Returns and Implied Cost of Capital,”Working Paper, UCLA. [18] Kraus, A., and J. S. Sagi. (2006) “Asset Pricing with Unforeseen Contingencies,” Journal of Financial Economics 82, 417-453. [19] Lettau, M., and S. Ludvigson. (2001) “Consumption, Aggregate Wealth, and Expected Returns,”The Journal of Finance 56, 815-849. [20] Lundholm, R. J., and R. G. Sloan. (2004) Equity Valuation and Analysis, New York: McGraw-Hill. [21] Nekrasov, A., and P. K. Shro¤. (2006) “Fundamentals-Based Risk Measurement in Valuation,”Working Paper, University of Minnesota. [22] Ohlson, J. A., and B. Juettner-Nauroth. (2005) “Expected EPS and EPS Growth as Determinants of Value,”Review of Accounting Studies 10, 349-365. [23] Penman, S. H. (2007) Financial Statement Analysis and Security Valuation, third edition, New York: McGraw-Hill. 48

[24] Rubinstein, M. (1976) The Valuation of Uncertain Income Streams and the Pricing of Options,”Bell Journal of Economics 7, 407-425. [25] Wachter, J. A. (2006) “A Consumption-based Model of The Term Structure of Interest Rates,”Journal of Financial Economics 79, 365-399. [26] Wilson, R. (1968) “The Theory of Syndicates,”Econometrica 36, 119-132.

Appendix A: Proof of Theorems
In the proof of Theorem 3 the following lemma is useful. It follows from the de…nition of the forward probability measures (see Christensen and Feltham, 2003, proof of Proposition 9.1). Lemma 12 Let z Then
1

be some quantity measurable with respect to date
Q E t [R ; 1z 1]

1 information.

= FR

1;t

E

Q 1;t

[x

1] ;

where F R

1;t

is one plus the current forward rate from FR
1;t

1 to ; i.e.,

=

B B

1;t ;t

:

Proof of Theorem 3. The free cash ‡ value relation (23) follows directly from Theorem ow 1, since the free cash ‡ ows are equal to the dividends on a portfolio of consisting of the equity and the …nancial assets. In the proof of the spot residual operating income relation (24), we substitute out the free cash ‡ ows in (23) using the operating asset relation, i.e., Vt = =
T X

B t EQt [oi B
Q tE t

(oa

oa
T X

1 )]

=t+1

T X

[oi

oa ] +

B t EQt [oa

1] :

=t+1

=t+1

49

Using Lemma 12, we can rewrite the last term such that Vt =
T X

B t EQt [oi
T X

oa ]
Q 1;t Q Et

=t+1

+
T X

B

t

=t+1

h

FR
s

1;t

E

[oa

1]

[

1 oa

1]

=

B

Q tE t

[roi

oa ] +

=t+1

T X

i
1] :

B tF R

1;t

E

Q 1;t

[oa

=t+1

Using the de…nition of forward rates and collecting terms, we get Vt = =
T X T X

B

Q tE t

[roi

s

oa ] +

=t+1

B t EQt [rois
T X

oa ] +

=t+1

T 1 X =t

T X

B

1;t

E

Q 1;t

[oa

1]

=t+1

B t EQt [oa ]

= oat + = oat +

B t EQt [rois ] + BT t EQt [oaT ] T B t EQt [rois ] ;

=t+1

T X

=t+1

where the last equality comes from the clean surplus relation, i.e., oaT = 0. This proves the spot residual operating income relation (24).

50

t = Bt+1. Vt = oat + T X B t EQt [rois ] T i X oat + B t EQt [rois ] t =t+2 =t+1 h = oat + Bt+1.t E Q 1.t = T X h =t+2 B t EQt [rois ] B t EQt [rois ] B t EQt [rois ] B E Q 1.t EQ t+1.t [oit+1 ] + oat (yt ) + =t+2 Q Et+1.t oat (yt ) ## =t+2 Et+1.t rois i 1 1 =t+2 Rt+1.t EQ [oit+1 ] + oat (yt ) t+1.t B t 1.t B t EQt [rois ] Q s E t [roi ] B 1.t rois 1 BT t EQt [rois ] T T BT t EQt [rois ] : T T = T X Rt+1.t h 1] + T X B t EQt [rois ] ii Rt+1.t = = h Bt+1.t T X =t+2 Bt+1.t rois t+1 = + T X t B t EQt [rois ] rois t+1 + [Rt+1.t rois 51 .t EQ t+1.t =t+2 =t+1 T X =t+2 T +1 X 1.t [oit+1 ] + t B t EQt [rois ] : (42) =t+2 The term in the inner bracket can be rewritten as Bt+1.t 1h t i [oit+1 ] + oat (yt ) [Rt+1.t [oit+1 ] T X Q Q Bt+1.t Q E Rt+1.t roit+1 + T X = " =t+2 Et+1.t = Rt+1.The proof of the spot abnormal operating income growth relation (25) starts by rewriting spot residual operating income relation (24).t + 1 t B t EQt [rois ] " s Bt+1.t EQ t+1.t T X 1] T X B t EQt [rois ] =t+2 = Rt+1.t h t T i X Q B t EQt [roi ] Et+1.t B t =t+2 B 1.t EQ [oit+1 ] t+1.

t # 1 BT t EQt [rois ] T T # which is (25). the …rst term on the right-hand side can be rewritten such that T X B t Et [f cf ] = B t Et [oi ] T X =t+1 T X T X B t Et [oi ] B t Et [oa ] + 1 ]: =t+1 T X T X B t Et [oa ] + =t+1 T X T X B t Et [oa 1] 1] =t+1 1.t [oit+1 ] + Q Et+1. 1 s roi Rt+1.t B t EQt [aoig s ] BT t EQt [rois ] T T =t+2 R. It follows from the free cash ‡ value relation (23) that ow Vt = T X B t Et [f cf ] + =t+1 T X B t Covt [f cf .t B t E Q t =t+2 R.t EQ t+1.t 1 BT t EQt [rois ] : T T Substituting into (42) yields Vt = = 1 t " " Et+1.t B t Q s E t [roi ] E Q t =t+2 1 t [oit+1 ] + Rt+1.t roit+1 + T X T X t B t EQt [rois ] Q s E t [roi ] =t+2 = Rt+1.t oa =t+1 B t Et [f r =t+1 52 . 1 s roi Rt+1.Using Lemma 12 it follows that s Bt+1.t ) oa = B t Et [(1 + f r =t+1 =t+1 1.t Q T X T X Rt+1. q ]: =t+1 Using the operating asset relation and the de…nition of forward residual operating income. Proof of Theorem 4.

e. we get that T X 1.t .Using the de…nition of forward residual operating income and the fact that B t (1 + f r B 1. i. =t+1 can be rewritten as in (42) such that oat + = 1 t T X B t Et [oi " fr 1.t oa 1] T X " =t+1 Et [oit+1 ] + Bt+1. The …rst two terms of the forward residual operating income value relation.t Et [(oit+1 t oat )] + t B t Et [roif t ] =t+2 ## : (43) 53 . oaT = 0.t Et [oa 1] =t+1 = =t+1 =t+1 = oat + = oat + B t Et [roif t ] B t Et [roif t ].t ) = B t Et [f cf ] = B t Et [roif t ] T X =t+1 T X T X B t Et [roi t ] B t Et [oa ] + BT t Et [oaT ] f =t+1 T X T X B t Et [oa ] + B t Et [oa ] =t+1 T 1 X =t T X B 1. Vt = oat + T X B t Et [roi t ] + f =t+1 T X B t Covt [f cf . q ]. This shows (27). =t+1 T X =t+1 where the last equality follows from the clean surplus relation..

t =t+2 Using the fact that B t (1 + f r Bt+1. 1 t T X B t Et [oi T X fr 1.The term in the inner bracket can be rewritten such that Bt+1.t Et [roif t ] h f Et [roi t ] B t (1 + f r 1. B t Et [roif t ] B Et [roi f 1.t B t Et [roif t ] B t Et [roif t ] =B 1.t ] BT t Et [roif t ] T =t+2 =t+2 B =t+2 = T X h (1 + f r 1. T which when substituted into (43) yields that oat + = This shows (28).t Et [roi f 1.t oa 1] " =t+1 Et [oit+1 ] + Rt+1.t Et [(oit+1 = Rt+1.t B t Et [aoig t ] f BT t Et [roif t ] T =t+2 # 54 .t = T X t oat )] T X t we get that + t T X B t Et [roif t ] T X =t+2 f B t Et [roi t ] Rt+1.t ] =t+2 i i BT t Et [roif t ] T BT t Et [roif t ].t ] Rt+1.t = Rt+1.t .t ]: =t+2 =t+1 T X =t+2 1.t ) Et [roi f 1.t ) T +1 X 1.t ) Et [roi f 1.t Et [(oit+1 T X 1) T X B t Et [roif t ] =t+2 = Rt+1.t Et [(oit+1 = t oat )] + t T X B t Et [roif t ] + (Rt+1.t B t 1.t T X =t+2 t oat )] Bt+1.

I. = t + 1. = t + 1.e. uit (cit ) = P t 1 1 [ cit bi ] 1 . while the constant relative risk aversion (CRRA) P model has bi = 0. and the investors can trade in a complete set of zero-coupon bonds and the market portfolio. P t > 0. b] 1= ] y yt .Appendix B: Power Utility and Log-normal Distributions In this appendix we examine the case in which investors have power utilities. where fi = (bi vi bo )= : Scaling by the number of investors and de…ning the average minimum consumption level by b = bo =I. for simplicity. and assume this P 1= t (vi I) [ acct + b] 1= : De…ne “log-aggregate consumption” per capita as lacc object is normally distributed. Wilson (1968) shows that e¢ cient consumption plans are on the form (see also Christensen and Feltham. i. Then h(lacc jyt ) q t (x jyt ) = exp[ lacc = ] . Note that the investors’relative risk aversion is increasing (decreasing) if bit < (>) 0. In lieu of Theorem 9. 2003. we assume investors have the same risk cautiousness (the slope of the linear risk tolerance function). In this case. Let bo = i bi denote the aggregate minimum consumption level. T: 55 . T: ln( acc b). i = 1. :::. identical personal discount factors P . the marginal utility of consumption for any investor i is u0it (cit ) = Inserting this into (31) we get q t (x jyt ) = [ acc Et [[ acc b] 1= . Et [exp[ lacc = ]] y yt . cit bi > 0: The former two assumptions imply that the market is e¤ectively dynamically complete if consumption endowments are spanned.. :::. :::. Proposition 4. and. that their t personal minimum consumption levels bit are time-independent.3) cit (xt ) = fi + vi xt .

Di¤erentiating the valuation index for log-aggregate consumption per capita and taking expectations yields i h 0 1 : Et h t (lacc ) = Inserting this into the accounting-value relation (32) and de…ning risk-adjusted aggregate consumption per capita as racc = lacc = yields the following result. Then the market-to-book ratio for the operating assets is given by T X Vt B =1+ oat =t+1 t ReN OA s t Covt [ReN OAs t . Note that if b = 0 (i. Proposition 13 Assume the conditions in Theorem 5 hold and make the accounting assumptions (A)-(D). However. the CRRA model). (44) and the zero-coupon prices are given by Et [u0i (ci )] Bt= 0 = uit (cit (yt )) where P t P t exp[ fEt [racc ] racct 1 Vart [racc ]g]. racc ] . the VAR model for risk-adjusted aggregate consumption can be estimated directly based on log-aggregate consumption per capita without any preference-dependent parameters like in the exponential model (the risk cautiousness is just a time-independent mean adjustment). assume that future spot residual operating income and log-aggregate consumption per capita are jointly normally distributed. Furthermore.. note that the only di¤erence is a re-de…nition of risk-adjusted aggregate consumption per capita: average risk tolerance is substituted by the common risk cautiousness (= relative risk tolerance with b = 0). 2 to t. then the VAR model for risk-adjusted aggregate consumption per capital must be estimated simultaneously with a calibration to the term structure of interest rates using (45). and constant minimum consumption levels. if b 6= 0. and aggregate consumption per capital is substituted with log-aggregate consumption per capita. consequently. then lacct ln( ) + ln(acc ) and. 56 . and investors have power utilities with identical and constant risk cautiousness and personal discount factors. (45) is the investors’common personal discount factor from Comparing to the exponential utility model in the text.e.

2006. and Campbell and Cochrane. the utility function is event-dependent. however. Of course. 2003.and event-contingent. Allowing for time-dependent minimum consumption levels is straightforward. we have not been able to locate a general equilibrium analysis with heterogeneous agents (see. his current choice of consumption a¤ects his future habit level of consumption. Hence. the investors’utility functions are no longer time-additive.Appendix C: Habit Formation The analysis in Appendix B assumes that the investors’ minimum consumption levels are constant across time. for example. we allow bit (yt ) to be some general date. That is. In the internal habit formation models. yt ) = P t 1 1 [ cit bit (yt )] 1 . we derive the equilibrium prices in an e¤ectively dynamically complete market. the investor’ date t utility depends on his current s consumption as well as on some exogenous habit level of consumption. secondly.e. In the external habit formation models. Wachter. but as the following analysis demonstrates. P t > 0. in turn. for an extension to more general event-contingent preferences). cit bit (yt ) > 0: (46) Note that this is the standard power utility function except that we allow the minimum consumption level to be date. Although several models of this type assuming a representative agent economy have been considered in the literature (see.and event-contingent habit formation function. the investor’ utility depends not only on his current consumption but s also on his past consumption (the habit) and. In these models. for example. The issue is to establish aggregation. in an application of the model. we could let this function be some weighted average of current and past aggregate consumption. the marginal utility of consumption is increasing in the external habit level of consumption. The literature distinguishes between internal and external habit formation. Note that the investors’utility functions are time-additive with external habit formation (as opposed to internal habit formation). Chapters 4 and 6). 2006. 1999). i. this can be established in a similar fashion as with deterministic minimum consumption levels (see Christensen and Feltham. In the following. the Borch …rst-order conditions character- 57 . and general equilibrium analyses of this type are very rare (if they exist at all). First. The next step is to allow for preferences exhibiting habit formation such that more variability in the types of equilibrium term structures of interest rates can be obtained. we derive Pareto e¢ cient consumption plans (assuming homogeneous beliefs as in the preceding analysis) and. Consider date t utility functions of the type uit (cit .. Kraus and Sagi. on some weighted average of past aggregate consumption per capita (keeping-up-with-the-Joneses).

Each investor consumes a constant fraction of aggregate consumption (as a result of the assumed identical personal discount factors). I. :::. Note that the e¢ cient consumption plans have a similar form as in the standard power utility case in Appendix B. T . 8yt . ot i Substitution back into (49) yields the investors’ bit (yt ) = [ xt bot (yt )] = i o [ xt bot (yt )] . (47) (48) where i is the weight assigned to investor i. yt )'(yt ) = t (yt ). where fit (yt ) = [bit (yt ) vi bot (yt )]= and vi = i = o . s P Christensen and Feltham. if the habit levels are generated by current and past aggregate consumption per capita. t = 1. and t (yt ) is the multiplier for the aggregate consumption constraint in a central planner’ optimal risk sharing problem (see. cit (yt ) bot (yt ) = P it ot it . :::. t = 1. cit (yt ) bit (yt ) = it t : (49) it [ cit (yt ) '(yt ) '(yt ) Summation across investors using (48) yields xt P where bot (yt ) i bit (yt ) and e¢ cient consumption plans. t (yt ) ot PI i=1 cit (yt ) = xt . 8yt . the “…xed component” with habit formation is eventcontingent such that e¢ cient individual consumption is not necessarily measurable with respect to contemporaneous aggregate consumption. Scaling by the number of investors and de…ning the average habit level by bt (yt ) = bot (yt )=I. However. cit (yt ) = fit (yt ) + vi xt . 2003. De…ning it i t and inserting (46) in (47) yield 1 (yt ) (yt ) bit (yt )] = t . 8yt . T . This also implies that trading in the market portfolio and a complete set of zero-coupon bonds is no longer su¢ cient to ensure an e¤ectively dynamically complete market— there must be claims which allow implementation of the personal event-contingent “…xed components” associated with the habit formation. t = 1. I. '(yt ) . a su¢ ciently varied set of aggregate consumption claims will do. Of course. :::. :::. for example.izing necessary and su¢ cient conditions for Pareto e¢ cient risk sharing are 0 i uit (cit (yt ). :::. i = 1. T . i = 1. the marginal utility of consumption for any investor i is u0it (cit (yt )) = P 1= t (vi I) [ acct bt (yt )] 1= : 58 . Chapter 4).

1999. power utilities with identical risk cautiousness (and personal discount factors) and a general external habit formation allow aggregation. i.e. the equilibrium prices of zero-coupon bonds are given by Et [u0 (ci )] = B t (yt ) = 0 i uit (cit (yt )) P Et t acc [ acct b (y )] bt (yt )] 1= 1= : Hence. the analysis proceeds as in the stanlacc (y ) ln( acc dard power utility case in Appendix B (see Wachter. a single-consumption-date economy with no discounting will be su¢ cient and. for particular parameterization of “habit-adjusted log-aggregate consumption”per capita). = t + 1. vi = i = o ) and the distribution of eventcontingent habit levels. we will use dividend value relations. b ] 1= ] Et [[ acc y yt . Furthermore. That is. 59 . :::. without loss of generality. Using this result and assuming that “habit-adjusted log-aggregate consumption” per capita. b (y )).. For this discussion. q t (y jyt ) = b (y )] 1= [ acc . is normally distributed. In the economics of uncertainty. u (CE [w]) = Z u(w)d (w). both the equilibrium valuation index and the equilibrium prices of zero-coupon bonds are independent of both the distribution of initial wealth (as re‡ ected in the equilibrium fractions of aggregate consumption. bit (yt ). T: Note that the valuation index for date may not be measurable with respect to aggregate consumption per capita as in the standard power utility case. Appendix D: Risk-adjusted Expected Cash Flows and Certainty Equivalents In this appendix we discuss the relationship between risk-adjustments to expected cash ‡ ows or accounting numbers in the determination of equilibrium prices and the associated certainty equivalents.Inserting this into (31) assuming an e¤ectively dynamically complete market we get the equilibrium valuation index. the certainty equivalent of uncertain wealth w is de…ned as the wealth level which has the same utility as the expected utility of the uncertain wealth. 2006. and Campbell and Cochrane.

if w = z + u with z and u independent. and (B) power utility and log-normally distributed wealth. (w) = (z) + (u): Z exp [ (z + u) = ] d (z)d (u) Z Z exp [ z= ] d (z) exp [ u= ] d (u). the concepts of certainty equivalents and risk premia can only be applied to total wealth. in general. Moreover. we can represent the equilibrium prices of securities as that of a representative agent economy with a risk tolerance equal to the average risk tolerance in the economy. The only exceptions are risk neutral investors. of course.. and not individual assets in a portfolio of securities. . (A) Exponential utility and normally distributed wealth In this setting. we can only explicitly calculate certainty equivalents for particular combinations of utility functions and distributions of wealth. acc= ] = E[acc] Var[acc]= : The certainty equivalent for the representative investor with risk tolerance follows directly from the standard formula for calculating certainty equivalents with exponential utility and 60 . Hence. hence. CE [w] = CE [z] + CE [u] . Two such combinations are (A) exponential utility and normally distributed wealth. each investor’ optimally invested wealth will be a linear function s of aggregate consumption. i.. we know from a simpli…ed version of Proposition 9 that in an e¤ectively complete market.where (w) is the distribution of w. .e. the price of aggregate consumption per capita follows directly from the analysis in Section 5 and is given by V (acc) = E[acc] Cov[acc. and a risk cautiousness equal to the common risk cautiousness in the economy. and investors with exponential utility and independence of the individual cash ‡ ows. i. respectively). then Z u(w)d (w) = = and. For both types of utility functions. for the exponential utility case. Similarly. (w) E[w] CE [w] : Note that certainty equivalents and risk premia do not. the risk premium is de…ned as the di¤erence between the expected wealth and the certainty equivalent. in general. satisfy the value additivity principle due to the non-linearity of the utility function. Moreover.e. for the power utility case (see Section 5 and Appendix B.

However. the risk premium in the equilibrium price is twice as large as the risk premium in the certainty equivalent. i. The reason is that the risk premium in the equilibrium price is determined such that the representative agent has no incentive to deviate from holding the aggregate consumption level per capita. 2003.e. Hence. i. we can apply a simpli…ed version of Theorem 5.e.e. (B) Power utility and log-normally distributed wealth In this setting. u(acc) = where = 1 1 1 c1 . acc : V (acc) = E[acc] + Cov E[u0 (acc)] For simplicity. we cannot directly apply the equilibrium prices derived in Appendix B. Appendix 3A).normally distributed wealth (see. i. Christensen and Feltham. while the risk premium in the certainty equivalent re‡ ects the required risk premium to hold the aggregate consumption level per capita instead of going to a riskless position. 1 Var[acc]= : CE[acc] = E[acc] 2 Hence. u0 (acc) .. is the coe¢ cient of relative risk aversion. We may refer to this di¤erence as the “investors’surplus. which is log-normally distributed. 61 . for example. whereas here we must determine the equilibrium price on aggregate consumption per capital.. V (acc) = E[acc] + Cov[acc . assume that the utility function is a CRR utility function. acc]= E[acc ] = E[acc] + E[acc1 ] = E[acc1 ]= E[acc ]: E[acc ] E[acc] = E[acc ] Using the fact that for any scalar g.” since the driving forces are the same as those underlying the consumers’surplus in consumer theory. > 0. since these are the equilibrium prices on securities with normally distributed cash ‡ ows or accounting numbers. accg = exp [g ln [acc]] ..

this implies that CE [acc] = exp + 1 (1 2 ) 2 : Taking the log of the ratio between the certainty equivalent and the equilibrium price of aggregate consumption per capita yields " exp + 1 (1 CE [acc] 2 ln = ln V (acc) exp + 1 (1 ) 2 1 2 = : 2 Since the coe¢ cient of relative risk aversion ) 2 2 1 2 2 # is strictly positive. i. we can calculate the expected values in the equilibrium price using the above-mentioned formula for calculating expected values of exponentials of normally distributed random variables.e. 2 ). V (acc) = exp (1 ) + 1 (1 2 1 exp 2 1 = exp + (1 ) 2 2 ) 2 2 1 2 2 : In order to calculate the certainty equivalent. That is. i..e.and the assumption that aggregate consumption per capita is log-normally distributed. E[u(acc)] = = = 1 1 1 1 1 1 E[acc 1 ] ) + 1 (1 2 + 1 (1 2 ) 2 exp (1 exp ) 1 2 By the de…nition of the certainty equivalent. z ln [acc] N( .. and. thus. we get that CE [acc] > V (acc). the risk premium in the equilibrium price is higher than the risk premium in 62 . we must …rst calculate the expected utility using that aggregate consumption per capita is log-normally distributed.

the certainty equivalent again illustrating the investors’surplus in the equilibrium prices of optimally invested wealth. 63 .