EMPIRICAL ESTIMATION OF THE EXPECTED RATE OF RETURN ON A

PORTFOLIO OF STOCKS

Peter Easton
Ohio State University
and
University of Melbourne
Gary Taylor
University of Alabama
Pervin Shroff
University of Minnesota
Theodore Sougiannis
University of Illinois

July 2000

We thank Jeff Abarbanell, Doug Hanna, John OHanlon, Richard Leftwich, Steven Monahan, Jim
Ohlson, Ken Peasnell, Stephen Penman, Peter Pope, Stefan Reichelstein, Greg Sommers, Paul Taylor,
Martin Walker, Dave Williams, two anonymous referees, and workshop participants at the University
of California, Berkeley, the University of Chicago, Lancaster University, Manchester University, the
University of Minnesota, New York University, the Ohio State University, the University of Oregon,
and the University of Utah for helpful comments on earlier drafts.

ABSTRACT
We invert the residual income valuation model (using current stock prices, current book value
of equity and short-term forecasts of accounting earnings) to obtain an estimate of the expected rate of
return for a portfolio of stocks. Our approach is analogous to the estimation of the internal rate of
return on a bond using market values and coupon payments.
Estimation of the cost of equity capital by inverting the residual income valuation model requires
an estimate of growth in residual income beyond the forecast horizon. The contribution of our method
is that we use the stock price and accounting data to simultaneously estimate the unique implied growth
rate and the internal rate of return. This growth rate provides an adjustment for the fact that our
estimate of the internal rate of return is based on current book value of equity and short-term earnings
forecasts.
Our analysis of DJIA firms yields estimates of expected growth that are considerably higher
than those assumed by earlier studies. Our estimated market premium over the risk-free rate is closer
to the historical premium than that obtained by other studies using earnings forecast data.
After completing the pro-forma forecasting of earnings (as described in, Penman [2000], for
example) and/or after obtaining analysts forecasts of earnings for a number of firms with comparable
operating activities, our method may be used to estimate the markets expectation of the cost of capital
and growth for these firms. These estimates for comparable firms may be used to determine the
intrinsic value of an unlisted firm, a division of a firm, or a firm that is believed to be relatively
over/under-valued.

1.

INTRODUCTION AND SUMMARY

We invert the residual income valuation model (using current stock prices, current book value

of equity and short-term forecasts of accounting earnings) to obtain an estimate of the expected rate of
return for a portfolio of stocks. Our approach is analogous to the estimation of the internal rate of
return on a bond using market values and coupon payments.
Estimation of the cost of equity capital by inverting the residual income valuation model requires
an estimate of growth in residual income beyond the forecast horizon. The contribution of our method
is that we use the stock price and accounting data to simultaneously estimate the unique implied growth
rate and the internal rate of return. This growth rate provides an adjustment for the fact that our
estimate of the internal rate of return is based on current book value of equity and short-term earnings
forecasts.
Recent papers (notably Fama and French [1997]) have demonstrated shortcomings in
estimates of the cost of equity capital based on historical data. Use of the historical market premium to
estimate the expected premium and basing estimates of risk loadings ($s) on a long time-series of data
introduce a high degree of estimation error. Recognizing the problems associated with the use of
historical data, recent studies have relied on analysts earnings forecasts and the residual income
valuation model to form estimates of cost of equity.1 However, as a practical matter, forecasts are
available only for a finite horizon, necessitating assumptions about earnings beyond the forecast horizon.
Most prior studies assume an expected rate of growth in residual earnings in order to calculate

For example, Claus and Thomas [1998], and Gebhardt, Lee, and Swaminathan [1999].

a terminal value. While Claus and Thomas [1998] provide an economic argument for estimating the
rate of growth in residual income beyond the forecast horizon as three percent less than the risk free
rate, they provide no empirical analyses as to whether this estimate is reasonable. Frankel and Lee
[1999], Lee, Myers, and Swaminathan [1999], and Gebhardt, Lee, and Swaminathan [1999] assume
the rate of growth beyond the I/B/E/S forecast horizon that is implied by fading the firms return-onequity to the industry median return-on-equity over varying forecast horizons. Their only justification
for this assumption is that this fade captures the long-term erosion in return-on-equity over time.
They do not examine the empirical validity of this assumption.
Estimates of the cost of equity (r) may be very sensitive to assumptions about the rate of
growth in residual income (g) and the literature provides little guidance as to the appropriateness of any
particular assumed rate. Unlike these papers, which use ad hoc arguments to support an assumed
rate of growth, we estimate the rate of growth that is implied by market prices, book values, and the
finite period forecasts of accounting earnings.2
Our estimate of the expected cost of equity capital (r) is also based on analysts forecasts of
accounting earnings. However, unlike extant studies, we simultaneously estimate the cost of equity as
well as the expected rate of growth in residual income (g) from these data. Our estimate of g is the
unique perpetual growth rate such that market price is equal to the book value plus the discounted

The following argument, suggested by a referee, provides useful intuition here. Surely what we
assign to g is taken away from r (the market has simply given us r-g). If g is artificial then the
corresponding r is also suspect. Noise in the parsing process translates into noise in the discount rate
estimate. The extant literature uses economic arguments as a basis for assigning gs that are not implied
by the market prices, book value and earnings forecasts and the consequent error is parsed to the
estimate of r. In contrast, our method breaks r-g into the unique r and the unique g that are implied by
market prices, book values and analysts earnings forecasts.

present value of residual income. We invert the residual income valuation model and we use forecasts
of earnings obtained from I/B/E/S, recorded book values, and observed market prices to solve for
estimates of the cost of equity and expected growth in residual income.
The advantage of this approach is that the estimate of the cost of equity is not dependent on
assumptions regarding the expected rate of growth. Rather, our estimate of the expected rate of
growth is the rate implied by market prices, book values, and forecasts of earnings. In other words,
simultaneously calculating r and g provides an estimate of r that recognizes and adjusts for the fact that
our estimate is based on book value and short horizon forecasts of earnings.
The residual income valuation model equates price with the sum of book value and the present
value of expected residual income. Our forecast data permit the calculation of aggregate earnings for
the subsequent four-year period and hence residual income for the entire four-year period. We define
g as the perpetual rate of growth in residual income such that market price is equal to book value plus
the present value of this four-year residual income growing in perpetuity. Thus, there are two
unknown variables -- g and the expected rate of return r. Using the residual income valuation model
as the foundation, we express current price as a linear function of current book value and expected
aggregate four-year earnings.3 The multiples on book value and aggregate earnings are both
functions of r and g. Dividing both sides of this linear expression by book value leads to a simple linear
relation between the current price-to-book ratio and the ratio of the sum of the earnings forecasts for
the subsequent four years to current book value. This specification leads to a linear (regression)

We use four years of forecasts because that is the longest time period for which data are readily
available. The method could be readily applied to longer time periods if forecast data became available.

relation in which both the intercept and the slope coefficient are functions of r and g permitting the
estimation of these variables.
To the extent that our sample of I/B/E/S stocks represents the U.S. stock market, application
of our model to this entire sample provides an estimate of the expected return on the market. For the
years 1981 to 1998, our estimates of r range from 11 percent to 16 percent. Our estimated premium
over the risk-free rate (proxied by the 5-year Government T-Bond rate) is between 6 percent and 8
percent in recent years. These estimates are similar to the historical 6 - 8 percent premium in contrast
to the 3 - 4 percent obtained by other studies that use forward-looking data to estimate r.4 Our
estimates of growth in residual earnings g imply rates of growth in earnings that range from 7 percent in
1992 to 11 percent in 1981. For the overall sample, the average estimated growth in earnings is 10.6
percent.5
The simultaneous estimation of r and g is potentially critical to the evaluation of a stock or an
index (such as the Dow Jones Industrial Average -- DJIA). Lee, Myers, and Swaminathan [1999]
use an estimated r and the assumption that g is the rate of growth in residual income that is implied by
fading the firms return-on-equity toward its industry median over a long horizon.6 Based on this
4

The historical premium is the average obtained from Ibbotson Associates [1998]. Examples of
studies that have estimated the ex ante premium are Claus and Thomas [1998] and Gebhardt, Lee, and
Swaminathan [1999]. The relatively low estimate of the premium in these studies may be due to their
somewhat arbitrary assumptions about the rate of growth in residual earnings beyond the forecast horizon.
5

The actual realized growth in earnings for the S&P index over our sample period was 7.92
percent. (See Security Price Index Record Statistical Service, Standard and Poors, [1998]: 298-303).
6

They note the following reason for the potential lack of validity of this assumption: a firms
return-on-equity would not be expected to fade to the historic industry median return-on-equity -- rather it
will fade to the expected industry median. Since the industry return-on-equity will change from being
high to low as the stage of the life-cycle of the firms in the industry changes from the growth phase

assumption, they conclude that the DJIA is over-valued in recent years. However, their conclusion
depends crucially on the validity of their assumption regarding g. We use market prices for the DJIA
stocks to determine the markets expectation of r and g for the DJIA for each of the years for which
we have data. The lowest expected g is 5.8 percent in 1982 and the highest is 13.6 percent in 1998.
The expected rate of return r associated with the expected g of 13.6 percent in 1998 is 15.7 percent
per year in perpetuity. Arguments that the DJIA was rationally priced at the end of 1998 would need
to justify these high expectations.
The most obvious application of our method is the estimation of the internal rate of return from
an investment in a portfolio of stocks (as we demonstrate in this paper). However, the method may
also be useful in firm valuation. After completing the pro-forma forecasting of earnings (as described
in, Penman [2000], for example) and/or after obtaining analysts forecasts of earnings for a number of
firms with comparable operating activities, our method may be used to estimate the markets
expectation of r and g for these firms. These estimates for comparable firms may be used to determine
the intrinsic value of an unlisted firm, a division of a firm, or a firm that is believed to be relatively
over/under-valued.7
through the stable phase and then the decline, the assumption that the historic industry median return-onequity will be equal to the expected return-on-equity may not generally hold as a practical matter.
Further, Lee, Myers, and Swaminathan [1999] assume that, after the return-on-equity has faded to the
industry median the residual income continues at a constant level in perpetuity (that is the rate of growth is
zero). Since book value will generally continue to grow, this implies that the return-on-equity will decline
asymptotically toward the expected cost of equity capital. Zhang [1999] shows that this will not occur
when accounting is conservative.
7

We estimate firm-specific expected rates of return in the latter part of the paper using samples
of firms matched on industry and market capitalization. These matches are a crude attempt at finding
comparable firms. Of course, investment bankers could reasonably be expected to compile a set of
traded firms that are comparable with, say, an initial public offering which they are attempting to value.

2.

THE ESTIMATION PROCEDURE

Our procedure for determining the cost of equity capital using forecasts of earnings, recorded

book values, and stock prices, is derived in this section. The procedure has two essential elements.
First, we use the fact that accounting earnings may be summed over time. In other words, the
forecasts of earnings for, say, each of the next four years may be summed to obtain aggregate earnings
for the entire four-year period and hence residual income for this period.8 Second, using the residual
income valuation model as the foundation, we express current price as a linear function of current
book value and expected aggregate four-year earnings. The multiples on book value and aggregate
forecasted earnings are both functions of r and g. Dividing both sides of this expression by book value
leads to a simple linear relation between the current price-to-book ratio and the ratio of aggregate
forecasted earnings-to-current book value. This specification leads to a linear (regression) relation in
which both the intercept and the slope coefficient are functions of the cost of equity capital and growth
in residual income, permitting the estimation of these variables. The estimate of the cost of equity
capital is the internal rate of return that is, in effect, obtained by inverting the residual income model
using the estimate of growth that is implied by market prices, book value, and the forecasts of
earnings.
2.1.

The Residual Income Valuation Model

The no arbitrage assumption and clean surplus accounting are sufficient to derive the residual

This attribute, which may be simply illustrated by the fact that annual earnings are equal to the
sum of four quarterly earnings, is at the core of the empirical analyses in Easton, Harris, and Ohlson
[1992]. They show that this aggregation of earnings over varying time intervals leads to increasing
explanatory power of earnings for returns as the returns interval increases.

income valuation formula:9

(1)

where P0

is the market price per share at time 0,

B0

is the book value per share at time 0,

E0

is the expectation operator with expectations conditional on the information available at

time 0,

is the expected rate of return,

Xt

is the (comprehensive) earnings per share for fiscal period t-1 to t, and

[Xt - r Bt-1] is the residual earnings per share for period t-1 to t.
Prior studies estimating r from the residual income valuation model assume a forecast horizon
and calculate a terminal value that captures residual earnings beyond the horizon. The terminal value is
generally calculated by treating the horizon residual earnings as a perpetuity growing at an assumed
rate. Since the growth rate is assumed, r is the only unknown variable, which is then inferred from the
model. In contrast, our approach involves the simultaneous estimation of r and the rate of growth that
are implied by the market prices, book values and forecasts of earnings.
Equation (1) may be re-written to isolate the finite period for which we have forecasts of
9

The no arbitrage assumption has been shown by Rubinstein [1976] to be sufficient to derive the
dividend discount model. Clean-surplus accounting requires that reported book value of common equity at
time t is equal to reported book value of common equity at time t-1 plus net income available to common
shareholders for period t-1 to t minus net capital contributions to/from shareholders at time t.

earnings:

(2)

Recognizing that, under clean surplus accounting,

(3)

where dt is the dividend payment per share at time t, the first summation of equation (2) may be rewritten as follows:10

(4)

where R = (1+r)4 is one plus the four-year expected return on equity and XcT is aggregate four year
cum-dividend earnings. This term captures the present value of expected abnormal profitability over
the next four-year period recognizing the notion (underscored in Ohlson [1995]) that, consistent with
the assumption of no arbitrage and the Miller and Modigliani [1961] propositions, payment of
dividends, dt, in period t reduces next period earnings by rdt.
In order to operationalize model (2) we treat the four-year residual earnings [from equation
(4)] as a perpetuity and we define g as the (unknown) annual growth rate such that:

10

The steps in deriving equation (4) involve expression of B3, B2, and B1 as functions of X3, X2,
X1, d3, d2, d1 and B0 and collecting terms.

(5)

where G = (1+g) 4 is one plus the expected rate of growth in four-year residual income.11 The fouryear growth rate of (1+g)4-1 is the rate such that the present value of the growing perpetuity
(beginning with the four-year expected abnormal earnings calculated via equation (4)) explains the
difference between price and current book value. In other words, g is the unique growth rate, which,
if known, would permit the estimation of the internal rate of return implied by the current price, book
value, and the four years of earnings forecasts. The advantage of simultaneously calculating this rate
and the cost of equity capital is that the effects of GAAP accounting which lead to (1) a difference
between book value and price and (2) short term forecasts of earnings being not necessarily indicative
of long run earnings, are taken into account when calculating r. In other words, simultaneously
calculating r and g provides an estimate of r that recognizes and adjusts for the fact that our estimate is

11

We define g as the expected average annual rate of growth in residual income from the date on
which the forecasts of earnings are made. This definition differs from the definition more frequently
encountered when using the residual income model in equity valuation (see, for example, Penman [2000]
and Claus and Thomas [1998]). Previous applications of the residual income model generally define g as
the growth in residual income from the last year for which a forecast of earnings is available. For
example, when four years of forecasts are available, the residual income model is often based on the
following formulation:

where g4 is the growth in residual income from the fourth year onwards. Our estimate of r (together with
current price, book value and forecasts of earnings) may be used to determine the rate of growth g4 by
inverting this formulation of the model. The core issue is estimation of g (or g4) so that the residual
income model can be inverted to obtain an estimate of r.

based on book value and short horizon forecasts of earnings -- g provides this adjustment.12 The
remainder of the analyses in this section is the development of a procedure for simultaneously
estimating g and r.
Dividing equation (5) by book value B0, and re-arranging yields:
(6)

where
(6a)

(6b)

The linear relation between P0/B0 and XcT/B0 in (6) suggests that the average four-year cost of
equity capital (R-1) and the average four-year growth in residual income (G-1) may be estimated from
the intercept and the slope coefficients from a linear regression. For any firm j we can observe price

12

It is evident from the residual income valuation formula (equation (1)) that the difference
between price and book value indicates the sign of future residual income and a high price-to-book ratio
implies high (positive) future residual income. From equation (2) it is evident that, if the forecast of
residual income over the finite forecast-horizon (four years in our example) is small relative to the
difference between price and book value, residual income beyond the forecast horizon will be high relative
to residual income within the forecast horizon -- that is, g will be large. Although g adjusts for the fact
that our estimate of r is based on book value and short horizon forecasts of earnings, providing a
corresponding unique estimate of r, different GAAP may lead to a different estimate of g and a different
estimate of r. This caveat applies to all of the studies cited in this paper that use forecasts of earnings to
estimate r or implicit prices. For example, a change in GAAP that leads to different earnings within the
forecast horizon would lead to different implicit prices in Lee, Myers, and Swaminathan [1999].

10

Pj0 and the corresponding book value Bj0 but we must obtain a proxy for aggregate expected future
earnings XjcT. Our method of estimation of R and G takes this into account. The procedure is as
follows.
We rely on the idea that we can define an unobservable variable XjcT such that:13
(6c)

That is, XjcT is the aggregate cum-dividend earnings that would obtain if all J observations had the same
R and G. Our empirical proxy for XjcT is obtained by using I/B/E/S analysts forecasts as the measure
of E0[Xt], t=1,2,..,4 and assuming E0[dt]= d0, t=1,2,..,4. Thus, there are three sources of error in this
proxy: (1) R and G may vary across the sample of observations, (2) the markets expectations of future
earnings may differ from I/B/E/S forecasts, and (3) expected dividends may not equal dividends paid at
date 0. We design (below) an estimation procedure that overcomes the effects of this error.
As a practical empirical matter we estimate the following regression relation:
(7)

where " 0t is the regression intercept parameter to be used as an estimate of (0, " 1t is the regression
slope parameter to be used as an estimate of (1, and AjcT is the forecasted aggregate cum-dividend

13

That is:

(6d)

11

earnings for the four year period t =1 to t =4, calculated as follows:

(7a)

where E0(I/B/E/S) is the I/B/E/S forecast of earnings (Xjt) for firm j at date 0 and dj0 is the actual dividend
payment for firm j for the period ending at date 0.14
If we could observe XjcT, and if we were to use these observations in regression (7) instead of
AjcT , the R2 would be one. The error introduced by using the empirical proxy AjcT reduces the R2 and
biases the estimate of the regression slope coefficient downward. Since the downward bias in the
estimate of the slope coefficient is directly proportional to the decrease in the R2 from one, an unbiased
estimate of the coefficient may be obtained by dividing its estimated value by the estimate of the R2:15

14

We begin by assuming that the displacement of future earnings due to the payment of dividends
is 12 percent of the dividend payment based on the assumption that, if these dividends had been retained
in the firm, they would have earned roughly the historical market return. Then we use the following
iterative procedure for dealing with the fact that calculation of aggregate cum-dividend earnings should be
based on r rather than an assumed rate -- in our case 12 percent. After calculating r and g based on this
assumption, we then re-calculate expected cum-dividend earnings by compounding the expected dividends
(according to equation (7a)) at our estimated r. We then repeat the regression analysis with this revised
estimate of cum-dividend earnings to obtain revised estimates of r and g. We repeat this procedure until
the revision in the estimate of cum-dividend earnings leads to no change in our estimates of r and g. For
each of our samples, the change in our estimates of r due to this refinement is very small even in the first
iteration. For example, the estimate of r for the entire cross-section of observations in 1985 when we
calculate cum-dividend earnings using a 12 percent rate is 0.12702 and the estimate after one iteration is
0.12839. After two iterations it is 0.12871. Estimates of g are virtually unchanged for this sample. Of
course, it is probable that for a portfolio of firms with very high dividend payout ratios, the estimates of the
cost of equity capital may vary considerably across the iterations.
15

Ryan and Zarowin [1991] make a similar point in the context of overcoming the effects of
measurement error on estimates of the earnings response coefficient. Their paper provides a detailed
analysis of the point that the downward bias in the estimate of the slope coefficient in regression (7) will
be proportional to the decrease in the R2 from one. The essence of the argument is as follows. Let :j 0 =

12

(8)

and
(9)

where " ^1 is the estimate of the slope coefficient from regression (7) and R2 is the coefficient of
determination from this regression, " 0c and " 1c are consistent and unbiased estimates of (0 and (1, and

These estimates of (0 and (1 and equations (6a) and (6b) are used to obtain estimates of r and g.16

3.

DATA AND SAMPLE SELECTION

We illustrate our method of estimating r and g using the following data. The same data have

been used in the numerous other studies based on the residual income valuation model cited elsewhere

XjcT /Bj 0 - AjcT /Bj 0 be the error in the regression independent variable. Assuming that :j 0 is independent of
Pj 0/Bj 0 and XjcT /Bj 0, the estimate of the slope coefficient is FxP /(F2x+F2:) and the estimate of the
regression R2 is (FxP )2/[(F2x+F2:)F2P ] where FxP is the covariance of XjcT /Bj 0 and Pj 0/Bj 0, and F2x, F2:,
and F2P are, respectively, the variances of XjcT /Bj 0, :j 0, and Pj 0/Bj 0. Note that both the estimate of the
slope coefficient and the regression R2 are biased downward from their true values because of the
addition of F2: in the denominator.
16

We take the positive fourth roots of our estimates of (1+g)4 -- that is, G -- and (1+r)4 -- that is,
R. Of course, the negative roots also exist but these are economically meaningless.

13

in this paper. Like these studies we do not attempt to adjust for possible biases in the forecast data.
Accordingly, results must be interpreted with caution. Note, however, that our purpose is to illustrate
how the method could be used if reliable forecasts were available (for example, our method may be
used by analysts who have created their own forecasts).
Book value of equity (data item 60), price at fiscal year-end (data item 199), and number of
shares outstanding (data item 25) are obtained from the 1999 Compustat annual primary, secondary,
tertiary and full coverage research files.17 Earnings forecasts are derived from the summary 1999
I/B/E/S tape. We determine median forecasts from the available analysts forecasts on the I/B/E/S file
released on the third Thursday of December. We include only firms with December fiscal year- end.18
For observations in 1995, for example, the December 1995 forecasts became available on 21st
December. These data included forecasts for a fiscal year ending just 10 days later (that is, December
31, 1995) and either forecasts for each of the fiscal years ending December 31, 1996, 1997, 1998 and
1999 or the forecast for the fiscal year ending December 31, 1996 and a forecast of growth in earnings
for the subsequent three years (1997 through 1999). That is, in effect we have forecasts for the
subsequent four years. When available, we use the actual forecasts for each subsequent year (in this
example, 1996 through 1999) and when these forecasts are not available, we use the forecast for 1996

17

Observations with negative book value are deleted.

18

We deleted firms with non-December year-end so that the market implied discount rate and
growth rate are estimated at the same point in time for each firm-year observation. We use actual yearend book values (from Compustat) since most of the earnings and dividend activity are known by the
third Thursday of December. The results are robust to using forecasts of year-end book value obtained
via the clean surplus relation. That is, forecasts of book value are determined by taking opening book
value, adding forecasted earnings (for the year ending approximately 10 days after the forecast) and
subtracting net capital contributions.

14

and calculate forecasts for 1997 through 1999 using the I/B/E/S forecasts of growth in earnings.19 Our
sample covers the years 1981 to 1998 and includes a total of 26,546 observations.

4.

EMPIRICAL ANALYSES
Descriptive statistics for the sample firms are provided in Table 1. Since the sample

composition changes over time, we caution the reader that conclusions about changes across years may
reflect changes in sample composition rather than changes in the underlying (stock-market wide)
variables.
The annual sample size increases each year from a low of 756 firms in 1981 to 2,242 firms in
1998. Equation (6a) suggests that the estimate " 0tc of the intercept coefficient will be negative -- the
predicted sign of the numerator is negative since G is expected to be greater than one on average, and
the predicted sign of the denominator is positive since an expected rate of return which is less than the
expected rate of growth in residual income would imply an infinite value. Accordingly, the estimate of
the intercept is negative in every year ranging from a low of
-3.26 in 1998 to a high of -1.30 in 1984. The estimate " 1tc of the slope coefficient ranges from a low
of 3.17 in 1984 to a high of 6.76 in 1998. The pooled adjusted R2 is 42.2 percent.
The annual average r and g are 0.13 and 0.10, respectively. Although there is an obvious timetrend in r from an average of 0.14 for the 1981-1985 time period to 0.13 for the 1995-1998 time
period, we do not draw conclusions from this because of the change in sample composition over time.

19

In this example, firm-year observations with a negative forecast of earnings for 1996 are
deleted because growth from this negative base is not meaningful in our context.

15

The annual estimated g remains relatively constant over the sample period. The difference between the
estimate of r and the estimate of g decreases significantly from 1981 to 1998. Again this may reflect
change in sample composition (we will return to this point).
In view of recent arguments in the popular press that market expectations may be overly
optimistic, we provide some additional statistics to help address the question, are the markets
expectations regarding rates of return reasonable? First, the average return on the Standard and Poors
500 Index over the years 1926 to 1998 was 13 percent which equals the expected rate of return for
1998.20 Second, the Treasury-bond rates for each year are included as the last column of Table 1,
Panel B.21 As expected our estimate of the premium (that is, the difference between r and the T-bond
rate) is positive. The very low estimated premium in the early years of the data is partially explained by
the fact that I/B/E/S concentrated on large firms during this period.22
The estimated premium in recent years is between 6 percent and 8 percent. This is similar to
the 6 - 8 percent premium based on historical data (Ibottson Associates) in contrast to the 3 - 4
percent obtained by prior studies that use analysts forecasts to estimate r (for example, Claus and
Thomas [1998] and Gebhardt, Lee, and Swaminathan [1999]). The relatively low premium obtained in
these studies may be partly attributed to the fact that they assume g rather than estimating it

20

Source: Ibbotson Associates [1998]

21

The validity of comparisons with the T-bond rate is limited by the fact that we are calculating an
r for the infinite horizon future while the T-bond rate is for a finite five-year horizon.
22

We considered repeating the analyses for the sub-sample of firms that survived the entire period
for which we have data. However, time-trends in the estimates of r and g for these data may represent
no more than the changing economic characteristics of this particular group of firms over this portion of
their life-cycle.

16

simultaneously with r.
The implied growth in earnings (Gearn) is also reported in Table 1. These estimates range from
a low of 9.9 percent in 1992 to a high of 12.6 percent in 1998. The average estimate of Gearn for the
entire time period 1981 to 1998 is 11.8 percent which is greater than the actual growth in earnings for
the S&P 500 index of 7.92 percent. We also provide I/B/E/S forecasts of long term growth (ltg).
These estimates are greater than our estimates of Gearn in every year for which we have data. Note,
however, that our growth estimate is for an infinite horizon while the I/B/E/S forecast is for a much
shorter horizon (about five years). The Spearman correlation between the I/B/E/S forecasts of long
term growth in earnings and our estimates of Gearn is 0.606 and significant at least at the 0.01 level.

5.

ANALYSES OF SUB-SAMPLES
We consider several sub-samples of the data in order to demonstrate the application of our

method of estimation.
The annual estimates of r and g for the Dow Jones Industrial firms are detailed in Table 2.23
23

Our estimates are for the firms in the DJIA for which we have complete data. The number of
observations is between 27 and 29 in the 1990s but declines to only 23 observations in 1982 and 18
observations in 1981. There are many reasons for the lack of data. Some of these are as follows. In
1981 I/B/E/S began reporting the long term earnings growth rate. However, in the early years of the
sample this is a missing data item for several firms. For example, in 1981 long term growth rates are
missing for: Allied Signal, American Tobacco Brand, General Foods, International Harvester, JohnsManville, Owens-Illinois, Proctor & Gamble, Sears Roebuck & Co., Standard Oil of California, and
Woolworth. In several years a DJIA firm was excluded as an outlier. These firms had either a price-tobook ratio or an aggregate forecasted earnings-to-book value ratio in the top/bottom 1% of the entire
sample. In the early (later) years of the sample Bethlehem Steel and U. S. Steel (Coke) were eliminated
as outlying observations. Including Bethlehem Steel and U. S. Steel (Coke) in 1981 (1998) changes the
estimates of the cost of equity capital and growth in residual earnings from 0.156 and 0.094 (0.157 and
0.136) to 0.155 and 0.089 (0.170 and 0.150), respectively. We also include DJIA firms with fiscal years
ending in months other than December. This introduces error due to the mis-alignment of prices, book

17

Consistent with the results from Table 1, there is a significant negative relation (correlation coefficient of
-0.86) between r-g and time. These estimates of r and g reflect the apparent bull market this country
has experienced over the last few years. In other words, the current high level of the DJIA can be
explained by the small difference between expected return on equity and expected growth in residual
earnings. Assuming an expected growth of zero (or one that allows a firms return-on-equity to fade to
its industry median) leads Lee, Myers, and Swaminathan [1999] to conclude that the Dow Jones group
of stocks is over-valued. We provide a detailed example of the implications of fading the firms returnon equity to its industry median in Appendix 2.
The following question, however, still remains: Is the markets expected rate of return on the
DJIA realistic? Since the focus of this paper is simply on estimating the markets expectations of cost
of equity capital (rather than using these estimates to comment on the rationality of the market) we only
provide some statistics that may shed some light on this question. We leave conclusions regarding mispricing to the reader. At the end of 1998, the markets expected rate of return on the DJIA was 15.7
percent. This rate is greater than the historic average return on the DJIA. The average annual return
over the entire history of the DJIA (1896 to 1998) is 7.9 percent.24 Over the period during which there
have been thirty stocks in the DJIA (1928 to 1998) the average annual return has been 7 percent.
These rates, coupled with an expected growth in residual income of 13.6 percent in 1998, may suggest
that, perhaps the market was overly optimistic. However, the average annual return on the DJIA over

values and the date of the earnings forecast. However, the estimates of g and r are very similar whether
or not these observations are included.
24

The source of data on the DJIA is the web-site http://averages.dowjones.com/home.html.

18

our sample period (1981 to 1998) is 15.4 percent -- that is, slightly less than the expectation of 15.7
percent -- and the average annual return over the past four years has been much higher (24.6 percent).
Furthermore, the annual return has been greater than 15.7 percent in eleven of the eighteen years of our
sample period.
Table 3 presents the annual estimates of r and g for firms deemed to be in intangible intensive
industries (Panel A) and in the utilities industry (Panel B).25 We chose these industries because we
have priors about expected r and g and a comparison of our estimates of r and g provide a basis for
assessing the validity of these priors. This application at the industry level provides a sense of what the
method may yield if informed market analysts were to estimate the model for a group of similar firms
(that is, firms that have comparable accounting methods and firms of similar operating risk) in order to
estimate the expected rate of return for a firm that, say, is not traded. As expected, r and g are much
higher in the intangible intensive industries than the more stable regulated utilities industry. This is
consistent with the intangible intensive industries being more risky and tending to have higher expected
future earnings as the effects of the earnings from the assets that are not recorded on the Balance Sheet
start to accrue.26 Also, in recent years, r minus g is much lower for the intangible intensive industry

25

Intangible intensive firms are defined as firms in the two-digit SIC codes 48 (electronic
components and accessories), 73 (business services), and 87 (engineering, accounting, R&D, and
management related services); and three digit SIC codes 282 (plastics and synthetics), 283 (drugs), and
357 (computer and office equipment). This classification of intangibles intensive firms is also used by
Collins, Maydew and Weiss [1997] and Easton, Shroff and Taylor [1999]. Utility firms are defined as
firms in the four-digit SIC codes 4900-4953 (electric, gas and sanitary services).
26

Caution must be exercised when comparing our estimates of the cost of equity capital. We
have estimated the cost of equity capital which differs from the cost of capital for the operations of these
industries. For the same level of equity risk, operating risk will be higher if either the debt-to-equity ratio
is higher or if the cost of debt is higher.

19

compared to utilities, reflecting the high valuations placed on technology firms by the bull market in
these years.

6.

OTHER VARIABLES ASSOCIATED WITH EXPECTED RETURNS.

To gain an indication of the empirical relation between our estimates of the expected rate of

return and variables that have been associated with expected returns in the extant literature (see, for
example, Fama and French [1992]), we analyze the correlation between our estimates of the expected
return and these variables.27
We estimate firm-specific expected rates of return by matching each observation to 20 other
firms. Our matching is admittedly crude: arguably, analysts using the model to calculate firm-specific
estimates of cost of equity capital could achieve much better matches. For each observation in each
year we match firms via SIC codes. For each year of the sample we first seek at least 20 matches
based on 4-digit SIC code. We successfully match 31.3 percent of the observations in this way. If at
least 20 matches are not available at the 4-digit level, we match on 3-digit SIC code, then on 2-digit
code, and finally on 1-digit code (we match 12.8, 35.8 and 20.1 percent of the observations,

27

It is tempting to compare our estimates of expected returns with return realizations. However,
as Elton [1999] observes, realized returns are a very poor proxy for expected returns. His observation
regarding the U.S. stock market is particularly pertinent to our sample period. He states: In the recent past,
the United States has had stock market returns of higher than 30 percent per year while the Asian markets have had negative
returns. Does anyone honestly believe that this was the riskiest period in history for the United States and the safest for Asia?

Furthermore, his observation should be underscored in our study where our estimates of expected returns
are for the long-run future rather than for the short-run. The results of analyses of industry portfolios in
Gebhardt, Lee, and Swaminathan [1999] are consistent with Eltons postulate. They find that the
observation of a significant correlation between expected returns and realized returns is due to just one of
the 48 industries in their sample (and this industry includes observations for just two firms on average).

20

respectively). Then, within the sets of firms matched on SIC codes, we select the 20 observations that
have a market capitalization that is closest to the market capitalization of the subject firm.28 We then
calculate the expected rate of return for these 20 observations using the procedure discussed in Section
2 and we assign that estimate to the subject firm. We examine the correlation between our estimate of
expected return rjt and each of the following variables:
sizejt

the natural logarithm of the market capitalization (that is, price per share multiplied by
number of shares outstanding) of firm j at the end of year t,

$jt

the estimate of the riskiness of returns for firm j implied by the capital asset pricing
model. Betas are estimated from the market model using a maximum of 60 and a
minimum of 24 monthly returns up to the month prior to the earnings forecast release
date t using the CRSP Value-Weighted Market Index and the Dimson [1979]
correction for non-synchronous trading (monthly returns on individual stocks are
obtained from the CRSP monthly returns file),

BPjt

is ratio of book value to market value for firm j at the end of year t, and

EPjt

is ratio of earnings per share to price per share for firm j at the end of year t.

Descriptive statistics for each of these indicators of risk and for our estimate of the cost of

28

As a practical matter, when estimating r for a portfolio of stocks, g adjusts for the fact that the
book value and forecasts of earnings are accounting numbers. Differences in accounting methods across
firms are averaged out via the regression. When using the method to calculate the cost of capital for, say,
a non-traded firm by calculating the r and g for a similar set of firms, the same observation applies.
However, we may also wish to apply the estimate of g to the accounting data of the non-traded firm.
This may necessitate some adjustments to the accounting data of the non-traded firm to ensure that they
are comparable with those of the comparable firms. This issue and many others associated with the
practical implementation of the residual income valuation model are discussed in detail in Penman [2000].

21

equity capital (rjt) are provided in Table 4. The estimates of the cost of equity capital range from -8.6
percent to 33.1 percent. Even with our crude matching procedure, only 0.1 percent of these estimates
are negative: no estimates are significantly negative at the 0.05 level (the minimum t-statistic is -1.59).
The median t-statistic for our 21,943 estimates of the cost of equity capital is 7.40. The average ratio
of book-to-price is 0.627 reflecting the period from which we have drawn our sample and the fact that
I/B/E/S tends to cover larger firms (median size of 4,341 million). Estimates of beta are centered
around one and range from -2.499 to 4.925: that is, there is a more or less symmetric distribution of
risk from lower than the market portfolio to higher than the market portfolio. The mean earnings-price
ratio is 0.043, consistent with the higher prices of stocks observed in recent years.
The average of the annual correlations among each of the risk measures are reported in Table
5. The correlations between rjt, book-to-price and earnings-to-price are, consistent with our
expectations, positive and significant (Spearman correlations of 0.128 and 0.164 with t-statistics of
3.312 and 5.351). However, the correlation with size is close to zero (0.028). We offer two
explanations for this result. First, the empirical observation of a relation between firm-size and returns
has been driven primarily by relatively very small firms while I/B/E/S tends to provide forecasts for
larger firms. Second, the relation between firm size and returns that was documented for earlier periods
has not been evident in the decade of the 1990s which is the period from which we draw most of our
sample firms. The Spearman correlation between the expected cost of equity capital and beta is
negative -0.083 with a t-statistic of -2.194. In light of Fama and French [1992] we are not surprised
that we do not find a positive correlation between beta and expected return. The negative correlation
is, however, difficult to explain. We simply note that this negative relation was also observed by
22

Gebhardt, Lee, and Swaminathan [1999] for this sample of firms.29

7.

REITERATING THE PRACTICAL IMPLICATIONS

We have developed a method for determining estimates of expected cost of equity capital and

expected growth in residual earnings. Simultaneous estimation of these estimates provides a means of
adjusting for the reliance of our method on book value of equity and forecasts of accounting earnings
for a short horizon.
We repeat the point made in the introduction that the most obvious application of our method is
the estimation of the internal rate of return from an investment in a portfolio of stocks. However, the
method may also be useful in firm valuation. After completing the pro-forma forecasting of earnings (as
described in, Penman [2000], for example) and/or after obtaining analysts forecasts of earnings for a
number of firms with comparable operating activities, our method may be used to estimate the markets
expectation of r and g for these firms. These estimates for comparable firms may be used to determine
the implicit value of an unlisted firm, a division of a firm, or a firm that is believed to be relatively
over/under-valued.

29

Gebhardt, Lee, and Swaminathan [1999] report Spearman correlations among each of their risk
measures. Although we calculate beta, the book-to-price ratio and firm size a little differently, not
surprisingly, our calculations of the correlations among these variables (and the levels of significance of
these correlations) are very similar to theirs. Since they do not report correlations between their firmspecific estimates of cost of equity capital and each of the risk factors, we cannot draw comparisons.
Gebahardt, Lee, and Swaminathan [1999] observe a negative relation between beta and their estimate of
the cost of equity capital in comparisons across quintiles of observations where the quintiles are based on
estimated beta.

23

Appendix 1
Derivation of an estimate of the growth in earnings Gearn from our estimate of growth in
residual income g

Although our estimate of growth in abnormal earnings is a fundamental ingredient in the

implementation of the residual income model, the more common estimate of growth in the academic
literature and in the business press is growth in earnings. For example, analysts (such as I/B/E/S,
Standard and Poors, Value Line and Zacks) provide forecasts of earnings rather than abnormal
earnings and management frequently give growth forecasts for their firms earnings.
The derivation of our estimate of growth in earnings is provided in this Appendix. This estimate
(denoted Gearn) is a function of our estimates of r and g. We provide estimates of Gearn in our
empirical analyses in Section 4. The derivation of our estimate of growth in earnings, Gearn, is as
follows. From the definition of growth in residual income, g, we can write:

(a1)

Assume E0(dt) = d0 for all t, then this equation becomes:30

(a2)

30

An assumption about expected dividends is a necessary part of the derivation of a formula for
growth in earnings. Note that the purpose of this Appendix is to show how growth in earnings is related
to growth in residual income and to provide a formula for estimating a growth variable with which
academics and investors are more familiar.

24

where:

(a3)

Since, under clean-surplus accounting:

(a4)

it follows that:

(a5)

If we define Gearn as the geometric average annual growth in earnings from the first four years to the
second four years, then:

(a6)

Re-arranging yields our formula for four year growth in earnings:

(a7)

25

Appendix 2
Implications of Fading the Return-on-Equity to the Industry Median Return-on-Equity
A motivation for our paper is the observation that estimates of implicit price (as in Lee, Myers,
and Swaminathan [1999]) and estimates of the cost of equity capital (as in Gebhardt, Lee, and
Swaminathan [1999]) are very sensitive to the assumed rate of growth in residual income beyond the
forecast horizon. In short, if g is artificial then the corresponding implicit price or the corresponding r is
suspect. The extant literature uses economic arguments as a basis for assigning gs and the consequent
error is parsed to the estimate of r. In contrast, our method estimates the unique r and the unique g that
are implied by market prices, book values and analysts earnings forecasts. The aim of this appendix is
to illustrate the issues associated with assuming the g that is implied by fading the firms return-on-equity
to the historic median industry return-on-equity. This procedure is used by Frankel and Lee [1999],
Gebhardt, Lee and Swaminathan [1999], and Lee, Myers, and Swaminathan [1999]. The illustration is
based on a detailed analysis of Merck as at December 31, 1998.
A small change in the assumed g can cause a considerable change in the estimate of implicit
value. For example, at the end of 1998, Mercks price per share was $147.50 with a corresponding
expected g of 0.063. Ceteris paribus, an assumed growth rate of 0.068 would suggest an implicit
price of $172.60 an assumed growth rate of 0.058 would suggest an implicit price of $129.65.
Although an error of half of one percentage point in the estimate of the rate of growth may seem large,
close examination of the data suggests that the use of fade rates as a means of dealing with future
growth, may introduce a great deal of error.
Mercks return-on-equity in 1998 was 0.417 and its forecasted return-on-equity at the end of
26

the I/B/E/S forecast horizon (2002) was 0.344. The method used by Lee, Myers, and Swaminathan
[1999] and Gebhardt, Lee, and Swaminathan [1999] is to assume that this return-on-equity will fade
over the years 2003 to 2010 from 0.344 to the industry median return-on-equity which was 0.004.
This very low return-on-equity, which seems unlikely, amounts to assuming the following rates of
growth in residual income for the years (2003 to 2010) beyond the forecast horizon -- 0.07, 0.03, 0.01, -0.05, -0.11. -0.18, -0.28, -0.46, -0.94.31 The rates of growth in earnings are very similar to the
rates of growth in residual earnings (0.07, 0.03, -0.01, -0.05, -0.11, -0.18, -0.28, -0.45, -0.91). In
other words, the assumed g is too low.
Inverting the residual income valuation model (as in Gebhardt, Lee, and Swaminathan [1999])
using the market price of $147.50 and this (implicit) assumption of a very low g leads to a very low
estimate of the cost of equity capital (0.001). Lee, Myers, and Swaminathan [1999], use this assumed
g and historical estimates of the cost of equity capital to estimate intrinsic value. More realistic
estimates of r coupled with these low estimates of g lead to estimates of intrinsic value that are much
lower than market prices. For example, estimates of r of 5, 10, and 15 percent imply intrinsic values of
$22.27, $16.73, and $13.41, respectively.
We have chosen Merck because it provides a vivid illustration of the problems associated with
imposing an assumed g rather than using the data to estimate g. The problems, however, will be
present whenever the assumption about the future return-on-equity differs from the markets
expectation. In order to give an indication of the pervasiveness of problems associated with the use of

31

Gebhardt, Lee, and Swaminathan [1999] and Lee, Myers and Swaminathan [1999] assume that
g changes abruptly from -0.94 in 2012 to 0.00 in 2013.

27

fades in return-on-equity as a means of implicitly assuming g, Table A2 provides summary statistics for
the analyses of the 30 stocks comprising the Dow Jones Industrial Average as of December 31, 1998.
For each of the DJIA 30 stocks the estimate of the cost of equity capital is lower than the
return-on equity. In other words, if the historic median industry return-on equity is low, the estimate of
the cost of equity capital is low. This observation underscores the sensitivity of the estimate of r in
Gebhardt, Lee, and Swaminathan [1999] to their assumption about the rate of growth g. Further, we
observe that the assumed rate of growth in earnings in 2010 is negative for eleven of the 30 DJIA firms.
Although this is possible, this seems to suggest rates of growth that are low particularly in view of the
fact that these rates are assumed rates of growth in real -- not nominal -- earnings.
The 2001 expected return-on-equity based on I/B/E/S forecasts is, on average, 78 percent
higher than the historic industry median return-on-equity. In other words, assuming that the firms
return-on-equity fades to the industry median return-on-equity, suggests a considerable decline in the
future profitability of these firms. The implied cost of equity capital (based on this assumption) for an
equally-weighted portfolio of DJIA firms is 7 percent which is only slightly above the five-year Treasury
Bond rate of 5.1 percent. Considering the pessimistic forecast of future profitability, the implied
riskiness of these firms is remarkably low.
The role of both r and g is evident from equation (5) -- if the assumed g is similar to (different
from) the estimated cost of equity capital, r-g will be small (large) and the perpetual residual income will
be divided by a very small (large) number. The importance of estimating both r and g rather than
assuming g is also emphasized by the estimates of implicit price in the last three columns of table A2.
These prices are formed by taking the g that is assumed by Lee, Myers, and Swaminathan [1999] and
28

applying costs of equity capital r of 5, 10, and 15 percent. These estimates are all less than the estimate
of the rate of return on the DJIA of 15.7 percent which we estimate for 1998 and report in Table 3. As
in the Merck illustration these implicit prices vary considerably as we change the cost of equity capital.

29

Table A2:
Analyses of the firms in the Dow Jones Industrial Average as at 31 December, 1998
Return-on-Equity

Forecasted growth in:

Implicit Price

DJIA Company

2001
I/B/E/S
forecast

Industry
Median

Implied
r

Residual
Income in
2010

Earnings
in 2010

Market
Price

r=0.05

r=0.10

r=0.15

AT&T

0.302

0.128

0.066

-0.174

-0.061

75.75

106.64

43.21

24.63

Alcoa

0.184

0.113

0.077

-0.114

0.013

74.56

139.54

50.19

25.54

Allied-Signal

0.255

0.151

0.086

-0.046

0.042

44.31

99.47

34.30

16.74

American-Express

0.224

0.154

0.085

0.024

0.084

102.50

236.41

76.26

34.41

Boeing

0.177

0.151

0.086

0.034

0.060

32.63

68.14

26.06

14.10

Caterpillar

0.277

0.146

0.084

-0.094

0.017

71.56

154.21

53.83

26.61

Chevron

0.171

0.078

0.035

-0.194

-0.117

71.56

53.19

30.58

22.12

Citigroup

0.191

0.154

0.143

-0.180

0.094

25.38

145.80

48.87

23.10

Coca Cola

0.378

0.216

0.051

0.024

0.049

67.00

68.16

24.91

12.87

Disney

0.137

0.098

0.058

-0.025

0.035

25.38

31.53

10.83

5.28

Dupont

0.305

0.152

0.047

-0.131

-0.091

53.06

50.39

27.48

19.26

Eastman Kodak

0.338

0.148

0.083

-0.169

-0.038

72.00

141.41

54.78

30.04

Exxon

0.211

0.078

0.036

-0.232

-0.130

71.56

50.22

24.42

15.98

General Electric

0.261

0.163

0.050

0.001

0.029

102.00

102.48

39.10

21.14

General Motors

0.313

0.161

0.120

-0.223

-0.010

71.56

233.22

92.35

51.67

Goodyear

0.181

0.161

0.108

0.075

0.105

50.43

170.90

57.64

27.34

Hewlett-Packard

0.212

0.135

0.074

-0.023

0.047

71.56

128.13

43.38

20.75

I BM

0.284

0.153

0.074

-0.031

0.047

184.38

336.72

108.11

48.49

International Paper

0.259

0.136

0.104

-.300

-0.091

44.81

88.79

46.60

31.48

Johnson & Johnson

0.268

0.004

0.002

-0.924

-0.878

83.88

12.31

9.19

7.38

McDonalds

0.191

0.152

0.068

0.077

0.101

76.81

123.44

39.96

18.12

Merck

0.362

0.004

0.001

-0.937

-0.906

147.50

22.27

16.73

13.41

3M

0.284

0.136

0.052

-0.129

-0.071

71.13

73.67

35.70

23.34

Morgan JP

0.127

0.155

0.090

0.089

0.058

105.06

214.99

91.49

54.21

30

Philip Morris

0.515

0.387

0.102

0.044

0.058

53.50

131.07

54.70

31.81

Proctor & Gamble

0.314

0.148

0.053

-0.075

-0.018

91.06

98.07

36.77

19.54

Sears, Roebuck

0.225

0.137

0.053

-0.017

0.024

42.50

46.31

16.91

8.77

Union Carbide

0.138

0.152

0.089

0.132

0.116

42.50

102.17

35.18

17.12

UTD Tech

0.257

0.151

0.075

-0.034

0.034

108.75

194.40

68.85

34.59

Wal-Mart

0.182

0.137

0.039

0.054

0.069

86.00

62.10

21.22

10.26

Average

0.251

0.141

0.070

-0.117

-0.044

Notes to Table A2:

I/B/E/S forecast is the forecast of return-on-equity based on I/B/E/S forecast of earnings as at 17 December 1998 and
forecasts of dividends using the assumption of a constant dividend payout ratio as in Gebhardt, Lee and
Swaninathan [1999],
Industry Median is the historic median return-on-equity for the firms industry calculated using up to ten years of prior
data and the Fama and French [1997] industry classifications (following the method in Gebhardt, Lee and
Swaninathan [1999]),
Implied r is the implied cost of equity capital obtained by inverting the residual income model after fading the 2001
forecasted return-on-equity to the industry median return-on-equity in 2010. Spreadsheets used to calculate
these implied costs of equity capital using the data for each of the DJIA firms are available from the authors,
Forecasted growth in residual income in 2010 is the implied rate of growth is residual income from 2009 to 2010.
Forecasted growth in earnings in 2010 is the implied rate of growth is earnings from 2009 to 2010.
Market price is the price at the close of trade on December 30, 1998,
Implicit price is the price implied by the forecasted patterns of return-on-equity and the imputed estimate of the cost of
equity capital (r=0.05, r=0.10, and r=0.15).

31

REFERENCES
Claus, J., and J. Thomas, The Equity Risk Premium is much Lower than you think it is: Empirical
Estimates from a New Approach, Working paper, Columbia University, [1998].
Collins, D., E. Maydew, and I. Weiss, Changes in the Value Relevance of Earnings and Book Values
Over the Past Forty Years, Journal of Accounting and Economics, (1997): 39-68.
Dimson, E., Risk Measurement when Shares are Subject to Infrequent Trading, Journal of
Financial Economics, (1979): 197-226.
Easton, P., T. Harris and J. Ohlson, Aggregate Accounting Earnings Can Explain Most of Security
Returns, Journal of Accounting and Economics, (1992): 119-142.
Easton, P., P. Shroff and G. Taylor, The Effect of Accounting Treatment of Unexpected Returns and
Accounting Recognition Lag on Estimates of the Earnings Coefficient, Working paper, the
Ohio State University, (1999).
Elton, E., Expected Return, Realized Return, and Asset Pricing Tests, Journal of Finance, (1999):
1199-1220.
Fama, E., and K. French, The Cross-Section of Expected Returns, Journal of Finance,
(1992):427-465.
Fama, E., and K. French, Industry Costs of Equity, Journal of Financial Economics, (1997):153193.
Gebhardt. W., C. Lee, and B. Swaminathan, Towards an Ex Ante Cost-of-Capital, Working paper,
Cornell University, (1999).
Ibbotson, R., Stocks Bonds Bills and Inflation, 1998 Yearbook. (1998) Ibbotson Associates,
Chicago.
Lee, C., J. Myers, and B. Swaminathan, What is the Intrinsic Value of the Dow? Journal of
Finance, (1999): 1693-1742.
Miller, M., and F. Modigliani, Dividend Policy, Growth and the Valuation of Shares, Journal of
Business, (1961): 411-433.
Ohlson, J., Earnings, Book Values and Dividends in Security Valuation, Contemporary Accounting
Research, (1995): 661-687.
32

Penman, S., Financial Statement Analysis and Valuation Manuscript, Columbia University (2000).
Ryan, S., and P. Zarowin, On the Ability of the Classical Errors in Variables Approach to Explain
Earnings Response Coefficients and R2 in Alternative Valuation Models, Journal of
Accounting, Auditing and Finance, (1991): 768-785.
Rubinstein, M., The Valuation of Uncertain Income Streams and the Pricing of Options. Bell Journal
of Economics, (1976): 407-425.
Standard and Poors, Security Price Index Record Statistical Service, 1998.
Warfield, T. and J. Wild, Accounting Measurement Error and the Pricing of Interim Earnings,
Accounting Review, (1992): 821-842.
Watts, R. And R. Leftwich, The Time Series of Annual Accounting Earnings, Journal of Accounting
Research, (1977):253-271.
Zhang, X., "Conservative Accounting and Equity Valuation", Working paper, University of California, (1999).

33

Table 1
Annual Estimates of the Expected Rate of Return (r) and the Expected Long Term Growth in Residual Income (g)
n

"0c

"1c

R2

r-g

Gearn

ltg

26546

-2.58

5.51

0.422

0.13

0.10

0.03

0.118

0.135

1981

756

-1.97

3.62

0.543

0.16

0.11

0.05

0.125

0.139

0.143

1982

986

-1.66

3.95

0.544

0.14

0.09

0.05

0.105

0.132

0.130

1983

1098

-1.81

4.21

0.466

0.14

0.10

0.04

0.115

0.150

0.108

1984

1158

-1.30

3.17

0.569

0.15

0.09

0.06

0.122

0.150

0.123

1985

1137

-1.58

4.20

0.480

0.13

0.08

0.05

0.105

0.140

0.101

1986

1151

-1.60

4.35

0.489

0.12

0.08

0.04

0.097

0.135

0.073

1987

1153

-2.12

4.45

0.363

0.14

0.10

0.04

0.112

0.133

0.079

1988

1173

-1.38

3.62

0.491

0.13

0.08

0.05

0.109

0.133

0.085

1989

1221

-1.83

4.42

0.444

0.13

0.09

0.04

0.112

0.131

0.085

1990

1169

-1.86

4.32

0.419

0.14

0.09

0.05

0.106

0.133

0.084

1991

1284

-2.51

6.01

0.489

0.12

0.09

0.03

0.102

0.137

0.074

1992

1460

-1.40

4.47

0.585

0.11

0.07

0.04

0.099

0.146

0.062

1993

1717

-1.59

4.72

0.533

0.12

0.08

0.04

0.107

0.152

0.051

1994

1907

-2.20

4.84

0.434

0.14

0.10

0.04

0.125

0.159

0.067

1995

2042

-2.68

5.75

0.425

0.13

0.10

0.03

0.124

0.165

0.064

1996

2402

-2.37

5.63

0.435

0.12

0.09

0.03

0.119

0.187

0.062

1997

2508

-2.49

5.97

0.414

0.12

0.09

0.03

0.121

0.190

0.062

1998

2242

-3.26

6.76

0.387

0.13

0.10

0.03

0.126

0.187

0.051

Year

34

T-Bond

Notes to Table 1:
"0c and "1c (see equations (8) and (9)) are consistent estimates of (0 and (1 which are the coefficients in
the linear relation between the ratio of price-to-book value and the ratio of aggregate earnings-to-book
value:

where:
Pj 0
Bj 0
XjcT

is the price of shares of firm j at time 0;

is the per share book value of equity of firm j at time 0;
is the aggregate expected cum-dividend earnings for the four-year period t =1 to t =4 that
would obtain if all observations had the same R and G,
R2
is the adjusted R2;
agb
is the average ratio of four-year aggregate cum-dividend I/B/E/S analyst earnings forecasts (for
periods t to t+4) to book value of equity at time t;
P/B
is the price-to-book ratio at time t;
agprc is the average ratio of four-year aggregate cum-dividend I/B/E/S analyst earnings forecasts (for
periods t to t+4) to price at time t;
r
is the estimated expected cost of equity capital;
g
is the estimated long-term growth in residual income;
Gearn is the estimated long-term growth in earnings;
ltg
is the average of the I/B/E/S estimates of long term growth in earnings;
T-Bond is the five-year T-bond interest rate;
All estimates of r and g are significantly different from zero at the 0.001 level.

35

Table 2
Annual Estimates of the Expected Rate of Return (r)and the Expected Long Term Growth in
Residual Income (g) for Firms in the Dow Jones Industrial Average
Year

r-g

1981

18

0.156

0.094

0.062

1982

23

0.130

0.058

0.072

1983

23

0.130

0.065

0.065

1984

25

0.147

0.081

0.066

1985

26

0.124

0.059

0.065

1986

25

0.119

0.066

0.053

1987

26

0.140

0.082

0.058

1988

25

0.149

0.093

0.056

1989

25

0.134

0.081

0.053

1990

25

0.146

0.098

0.048

1991

29

0.144

0.112

0.032

1992

29

0.150

0.110

0.050

1993

29

0.121

0.065

0.056

1994

29

0.162

0.126

0.036

1995

28

0.136

0.097

0.039

1996

27

0.111

0.060

0.051

1997

29

0.147

0.115

0.032

1998

29

0.157

0.136

0.021

0.16

0.57*

-0.86**

Notes to Table2:
Annual portfolios only consist of those firms that make up the Dow Jones Industrial Average for each
year.
r
is the estimated expected cost of equity capital,
g
is the estimated long term growth in residual income,
D
is the Spearman correlation between the respective variable (in the column above) and time,
All estimates of r and g are significantly different from zero at the 0.001 level.

36

Table 3
Annual Estimates of the Expected Rate of Return (r)and the Expected Long Term Growth in
Residual Income (g) for Various Industries
Panel A
Intangible Intensive Industries

Panel B
Utilities

Year

r-g

r-g

1981

78

0.16

0.11

0.05

100

0.15

0.06

0.09

1982

110

0.14

0.11

0.03

117

0.15

0.06

0.09

1983

133

0.15

0.13

0.02

118

0.14

0.09

0.05

1984

154

0.15

0.11

0.04

117

0.14

0.07

0.07

1985

144

0.13

0.09

0.04

119

0.13

0.08

0.05

1986

151

0.12

0.08

0.04

117

0.12

0.08

0.04

1987

160

0.15

0.12

0.03

108

0.12

0.07

0.05

1988

159

0.15

0.11

0.04

109

0.12

0.07

0.05

1989

162

0.13

0.09

0.04

112

0.11

0.08

0.03

1990

154

0.14

0.11

0.03

124

0.12

0.09

0.03

1991

181

0.13

0.10

0.03

122

0.12

0.10

0.02

1992

212

0.13

0.09

0.04

127

0.11

0.07

0.04

1993

236

0.13

0.09

0.04

122

0.10

0.05

0.05

1994

279

0.13

0.09

0.04

133

0.11

0.06

0.05

1995

326

0.15

0.13

0.02

133

0.10

0.06

0.04

1996

439

0.14

0.11

0.03

139

0.10

0.06

0.04

1997

482

0.13

0.11

0.02

125

0.10

0.05

0.05

1998

460

0.15

0.13

0.02

116

0.09

0.04

0.05

Intangible intensive firms are defined as firms in the two-digit SIC codes 48 (electronic components and
accessories), 73 (business services), and 87 (engineering, accounting, R&D, and management related
services); and three digit SIC codes 282 (plastics and synthetics), 283 (drugs), and 357 (computer and
office equipment). Utility firms are defined as firms in the four-digit SIC codes 4900-4953 (electric, gas
and sanitary services). All estimates of r and g are significantly different from zero at the 0.001 level.

37

Table 4
Descriptive Statistics for 21,943* observations over the years 1981 to 1998
rjt

MC jt

$jt

BPjt

EPjt

Max

0.331

333,672.00

4.925

2.221

1.256

Min

-0.086

2.63

-2.499

0.064

-15.620

Mean

0.122

2270.74

1.048

0.627

0.043

Median

0.123

4341.12

1.003

0.559

0.061

SD

0.029

7775.56

0.561

0.371

0.175

Notes to Table 4:
is the estimated cost of equity capital for firm j at time t;
is the market capitalization (price per share multiplied by shares outstanding) for firm j at time t
(expressed in millions of dollars);
is the beta for firm j at time t derived from the capital asset pricing model;
$jt
is the ratio of book value of equity to market value for firm j at time t;
BP jt
is the ratio of earnings per share to price per share for firm j at time t
EPjt
* this table includes observation for which we have sufficient data to calculate an estimate of beta.
rjt
MCjt

38

Table 5
Average of the annual correlations between the stated variables for 21,943 observations over
the years 1981 to 1998 (t-statistics in parentheses). Pearson (Spearman) correlations are
reported above (below) the diagonal.
rjt
rjt

MC jt

$jt

BPjt

EPjt

0.023
(0.989)

-0.085
(-2.734**)

0.102
(4.053**)

0.101
(4.055**)

-0.134
(-6.927**)

-0.165
(-8.310**)

0.107
(7.023**)

-0.175
(-9.903**)

-0.209
(-7.430**)

MC jt

0.028
(1.071)

$jt

-0.083
(-2.194*)

-0.108
(-5.060**)

BPjt

0.128
(3.312**)

-0.180
(-7.438**)

-0.162
(-8.115**)

EPjt

0.164
(5.351**)

0.075
(3.848**)

-0.261
(-9.028**)

Notes to Table 5:
See Notes to Table 4:
*
**

significant at the 0.05 level

significant at the 0.01 level.

39

0.164
(6.654**)
0.362
(15.779**)