You are on page 1of 34

The Pricing of Dividends in Equity Valuation

John R. M. Hand*
Wayne R. Landsman
(UNC Chapel Hill)

Abstract

This paper employs Ohlson’s (1995, 1998) accounting based equity valuation model to structure
an empirical assessment of the pricing of dividends in stock prices. We address two questions.
First, to what extent does the pricing of dividends reflect Modigliani and Miller’s (1958, 1961)
one-to-one displacement property? Second, what explains the direction and magnitude of any
divergence from dividend displacement? Using annual cross-sections of NYSE, AMEX and
NASDAQ firms over the period 1974-1996, we find robust evidence that dividends are
materially positively priced, sharply contrasting with the negative relation predicted by dividend
displacement. We also find that the positive pricing of dividends is at least three times larger for
loss firms than for profit firms. Our explanation for these results is that managers of loss firms
use dividends to signal future profitability, while to a lesser degree managers of profit firms use
dividends to alleviate concerns about the misuse of free cash flow. We conclude that dividends
are a component of, and rich proxy for, other information about future abnormal earnings that is
reflected in price but is not yet captured by current financial statements.

This version: June 24, 1999

* Corresponding author (hand@unc.edu). This work is supported by the Center for Finance &
Accounting Research at UNC, and NationsBank Research Fellowships. We are grateful for the
comments of Jeff Abarbanell, Mary Barth, Bill Beaver, Amy Hutton, Rick Lambert, Charles Lee,
Steve Monahan, Krishna Palepu, Steve Ryan, and participants at Columbia, Cornell, Harvard, the
1998 Stanford Summer Camp, Washington University, St. Louis, and William & Mary on an
earlier version of the paper entitled “Testing the Ohlson model: v or not v, that is the question.”
I/B/E/S kindly provided the analyst forecast data.
1. Introduction

Despite having received considerable attention over the past 40 years, no consensus
exists regarding the role of dividends in equity valuation (e.g., Modigliani and Miller, 1958,
1961; Bhattacharya, 1979; Litzenberger and Ramsawamy, 1979; Miller and Scholes, 1982;
Miller and Rock, 1985; Fama and French, 1998). The goal of our paper is to shed new light on
this key issue by utilizing Ohlson’s (1995, 1998) accounting-based equity valuation model.
The Ohlson model links dividends, earnings, equity book values and other relevant
information to intrinsic value, and therefore stock price in an efficient market, by integrating the
dividend discount model with clean surplus and linear information dynamics. The formal nature
of the pricing relation gives the Ohlson model an advantage over more ad hoc approaches of
specifying in theory or in empirical tests the relation between dividends and stock prices.
Moreover, since the standard dividend discount model permits no information asymmetry,
neither does the Ohlson model. As a result, the Ohlson model cleanly reflects Modigliani and
Miller’s (1958, 1961) dividend displacement property, viz. equity value and dividends are
negatively related in that a dollar of dividends reduces equity value by a dollar.
In contrast to this prediction, we find that annual cross-sectional regressions of the equity
value of NYSE, AMEX and NASDAQ firms over the period 1974-1996 on the independent
variables comprising the Ohlson model yield strong and robust evidence that dividends are
materially positively priced. Not only is the multiple on dividends significantly positive, but the
estimated partial derivative of the market value of equity with respect to dividends is also
significantly positive. The partial derivative of the market value of equity with respect to
dividends is a linear combination of the coefficients on book equity, dividends, forecasted next
period earnings, and the discount rate, and will be –1 if dividend displacement property holds.
We explore two hypotheses for the positive relation between dividends and equity value:
profitability signaling and free cash flow mitigation. Both hypotheses relax the assumption of no
information asymmetry, thereby permitting dividends to be part of the set of information v that
predicts future abnormal earnings and is reflected in price, but is not yet captured by current
financial statements. The profitability signaling hypothesis posits that information asymmetry
exists between managers and investors regarding the future profitability of the firm, and that
managers use dividends as a costly, credible signal of their private information. The cash flow
mitigation hypothesis posits that managers have private information about the agency costs of

2
free cash flow generated by the firm, and that by paying dividends can credibly signal that they
are high quality managers who are not going to waste free cash flow. While both hypotheses
predict a positive relation between stock prices and dividends, each makes different predictions
given the sign of current period’s earnings. Under free cash flow mitigation, dividends are more
highly valued when free cash flow is high, which we propose is more likely when current
earnings are positive. In contrast, the profitability-signaling hypothesis predicts that dividends
are valued less when current earnings are positive, because dividends are then a less costly and
less credible signal of future profitability relative to when current earnings are negative.
Regression tests indicate that dividends are positively priced for both profit and loss
firms, but that the positive relation is at least three times larger for loss firms than for profit
firms. The latter finding is consistent with the profitability signaling hypothesis, i.e., dividends
signal future profitability more for loss firms. However, finding a positive pricing multiple for
profit firms is evidence consistent with managers of profit firms using dividends to communicate
credibly that they are not exploiting the firm’s positive free cash flows. Consistent with these
explanations, we also provide evidence that dividends help predict future abnormal earnings,
incremental to current and lagged abnormal earnings, and that the multiple on dividends is over
three times larger for loss firms than for profit firms.
The Ohlson model is frequently applied by regressing stock price on current book value
and current net income. Our findings suggest that failing to include dividends is both
informationally restrictive as well as potentially econometrically risky as a result of the high
positive correlations between dividends, current book value and current net income. By
including dividends in applications of the Ohlson model, the justifiable criticism found in Ohlson
(1998) that v cannot be simply set to zero can be partially addressed, since our evidence indicates
that dividends are positively correlated with v. That is, dividends are a component of, and a rich
proxy for, other information about future abnormal earnings that is reflected in price but is not
yet captured by current financial statements. Finally, by illustrating a key area in which
observed stock prices do not fit the Ohlson model, our findings suggest that explicitly extending
the Ohlson model to include information asymmetry may be fruitful.
The paper continues as follows. Section 2 overviews the Ohlson model, particularly the
role of dividends. Section 3 describes our sample selection criteria and data. Section 4 outlines
the estimating equations and reports regression results, while section 5 concludes.

3
2. Ohlson’s accounting-based equity valuation model

2.1 Model structure

Ohlson’s (1995, 1998) accounting-based equity valuation model links dividends,


earnings, book values and other relevant information to intrinsic value, and therefore stock price
in an efficient market, by integrating the dividend discount model with clean surplus and linear
information dynamics. In an economy with risk neutrality, no information asymmetry (i.e.,
homogeneous beliefs), non-stochastic interest rates, and a flat term structure, combining the
dividend discount model with clean surplus accounting yields the residual income model:


Pt = bt + ∑ R −τ Et [ ~
xta+τ ] , (1)
τ =1

where Pt is equity market value at time t, bt is equity book value at t, xt is accounting earnings for
period t, residual income or “abnormal” earnings is given by xta ≡ xt – rbt-1, r is the one-period
risk-free return, and R = 1 + r.1 Frankel and Lee (1998) and Lee, Myers, and Swaminathan
(1999) use the residual income model to obtain a measure of fundamental value that is then used
in identifying mispriced securities.
Ohlson transforms equation (1) into a linear function of equity book value, current net
income, net dividends, and a scalar v representing other information about future abnormal
earnings that is reflected in price but is not yet captured by current financial statements. This is
accomplished using the following modified AR(1) information dynamics:

xta+1 = ωxta + vt + ε~1,t +1


~ (2)
v~t +1 = γv t + ε~2 ,t +1 . (3)

The parameters 0 ≤ ω ≤ 1 and 0 ≤ γ < 1 are fixed at t and determined by the firm’s economic
environment and accounting principles. Equation (2) models the reasonable economic condition
that abnormal profits in expectation dissipate to zero over time. Since accounting in the Ohlson
model is unbiased, the evolution of abnormal earnings is unaffected by biases induced by
conservative (aggressive) accounting that would result in understated (overstated) equity book

~ τ
1
Equation (1) assumes that the no-infinite-growth condition E t [ b t +τ ] / R → 0 as τ → ∞ is satisfied.

4
values.2 Equation (3) imposes an AR(1) dynamic on v.3 Combining equations (1)−(3) yields:

Pt = (1 − k )bt + k (ϕxt − d t ) + α 2 vt , (4)

where net dividends dt = common dividendst + capital outflowst – capital inflowst, and

ϕ = R /( R − 1) > 0 (4.1)
k = ( R − 1)ω /( R − ω ) , with 0 ≤ k ≤ 1 (4.2)
α 2 = R /( R − ω )( R − γ ) > 0 . (4.3)

While equation (4) is frequently applied in empirical settings by assuming that v = 0,


Ohlson (1998) shows from the linear information dynamics that although v is not directly
observable, it can be inferred from its influence on expectations. Taking rational expectations at
t of the first linear information dynamic equation (2) yields:

vt = Et [ ~
xta+1 ] − ωxta . (5)

v is therefore next period’s full information expectation of abnormal earnings less the purely
autoregressive forecast of next period’s abnormal earnings. When equation (5) is substituted into
equation (4), equity market value can be expressed as:

Pt = β 1bt + β 2 (ζxt − d t ) + β 3 r −1Et [ ~


xt +1 ] (6)

where:

ζ ≡ Rr −1 (6.1)
∆ ≡ ( R − ω )( R − γ ) (6.2)
β 1 = R(1 − ω )(1 − γ )∆−1 ≥ 0 (6.3)
β 2 = −rωγ∆−1 ≤ 0 (6.4)
β 3 = Rr∆−1 > 0 . (6.5)

2
Unbiased accounting obtains if unrecognized goodwill (the difference between equity market value and equity
book value) is expected to tend to zero as time tends to infinity. Feltham and Ohlson (1995) lift this restriction.
3
Ohlson (1995, 1998) notes that the modified AR(1) information dynamics described in equations (2) and (3) can be
deduced from three properties associated with accounting measurements, rather than just being assumed outright.
These properties are that dividends reduce current book value but not current net income; that current dividends
reduce future expected net income with a marginal effect of R2 – 1 for two periods; and that the evolution of other
information v is dividend independent.

5
2.2 Predictions concerning the pricing of dividends

2.2.1 Dividend displacement

Because the standard dividend discount model permits no information asymmetry, the
Ohlson model cleanly reflects Modigliani and Miller’s (1958, 1961) dividend displacement
property. That is, equity value and dividends are negatively related in that a dollar of dividends
reduces equity value by exactly a dollar: viz. ∂Pt/∂dt = –1. This can be seen in equation (4) by
noting that ∂bt/∂dt = –1 and ∂xt/∂dt = ∂vt/∂dt = 0, and in equation (6) by noting that ∂Et[xt+1]/∂dt =
–r, and β1 + β2 + β3 = 1. As a result, the extent to which dividend displacement holds can be
assessed by comparing the signs and magnitudes of coefficient estimates obtained from empirical
regressions of equation (4) and/or equation (6) with their theoretical values in two ways.
First, the observed coefficient on dividends alone can be compared with its predicted
value, which depends on the information dynamics parameters ω in equation (4.2) or ω and γ in
equation (6.4). Thus, in a regression corresponding to equation (4) in which the standard
assumption is made that v = 0, the coefficient estimated on dividends d is predicted to equal –k =
–(R – 1)ω /(r – ω) < 0 if dividend displacement holds. Or, if v follows the modified AR(1)
dynamics detailed in equations (2) and (3), then in a regression corresponding to equation (6),
the coefficient on dividends should equal –β2 = rωγ∆-1 > 0.
Second, and more broadly, the partial derivative of the market value of equity with
respect to dividends ∂Pt/∂dt can be calculated and compared with its predicted value of –1 under
dividend displacement. The partial derivative of market value of equity with respect to
dividends is a linear combination of the coefficients on book equity, dividends, forecasted next
period earnings, and the discount rate, depending on the assumptions made concerning v. For
example, in a regression corresponding to equation (4) where the standard assumption is made
that v = 0, the partial derivative of the market value of equity with respect to dividends is the
coefficient on dividends less the coefficient on book equity. Or, if v follows the modified AR(1)
dynamics detailed in equations (2) and (3), then in a regression corresponding to equation (6),
the partial derivative of the market value of equity with respect to dividends is the coefficient on
dividends less the coefficient on book equity, less the coefficient on forecasted next period
earnings multiplied by R – 1.

6
2.2.2 Dividends and taxes

The dividend discount model accounts for corporate level taxes, but not shareholder level
taxes. The question of whether or not the dividend taxes faced by shareholders are capitalized
into share prices received much attention in the 1970s and 1980s (Miller and Scholes, 1978,
1982; Auerbach, 1979; Poterba and Summers, 1984, 1985; Bernheim, 1991). A resurgence of
interest in this question is reflected in recent work by Fama and French (1998), Harris and
Kemsley (1999), Harris, Hubbard and Kemsley (1999), and Collins and Kemsley (1999). We
briefly discuss these papers so as to delineate our contribution to the literature on the pricing of
dividends in equity valuation.
A key feature of our examination of the pricing of dividends is predicating our estimating
equations on the Ohlson model, which provides a rigorous link between dividends and share
price. This contrasts with the empirical work of Fama and French (1998), who adopt a more ad
hoc method of examining the relation between dividends and stock prices. Fama and French are
interested in isolating and measuring the effects of taxes in the pricing of dividends and debt.
However, rather than appealing to an explicit valuation model, they use a wide range of past,
current and future factors such as earnings, investment and R&D expenditures to serve as proxies
for expected net cash flows of the firm. These factors are included as independent control
variables in cross-sectional regressions of firm value under the assumption that as a group they
represent an unbiased proxy for expected net cash flows.
Consistent with our results, Fama and French find that dividends have a positive
estimated marginal relation with firm value. They view this as puzzling since they hypothesize
that the presence of tax effects in the pricing of dividends should lead to a negative marginal
relation. However, the lack of an explicit valuation model makes it hard to rule out the
possibility that their results stem from economic misspecification rather than dividends
conveying incremental information about firm value. For example, their set of independent
variables excludes equity book value. From the standpoint of the Ohlson model, this is
equivalent to assuming that abnormal earnings persist indefinitely, rather than decay to zero over
time as seems economically much more plausible. Moreover, because dividends, earnings and
book equity are highly positively correlated, omitting book equity suggests that the coefficient
estimates found on Fama and French’s set of independent variables are likely upward biased.
In work undertaken concurrently with our own, Collins and Kemsley (1999), Harris,

7
Hubbard and Kemsley (1999), and Harris and Kemsley (1999) examine the effects of dividend
and capital gains taxes on equity valuation by modifying the Ohlson model to reflect these taxes.
Most particularly, Collins and Kemsley (1999) predict that when capital gains taxes are taken
into account, firms can reduce or even eliminate their shareholders’ capital gains taxes on
earnings by paying dividends. This creates a positive role for dividends in firm valuation that is
a function of the capital gains tax rate and which is in addition to any signaling or agency role
for dividends (our emphasis). They predict and find a positive intertemporal relation between
investors’ valuation of dividends and the tax rate on capital gains over the period 1975-1994 in
Ohlson-type regressions. However, they are silent as to the sign of the unconditional coefficient
mapping dividends into stock prices, and cross-sectional variation in that coefficient. In contrast,
our work particularly addresses these latter two issues.

3. Sample selection and descriptive statistics

Data for the period 1974−1996 were extracted from the 1997 Compustat Primary,
Secondary, and Tertiary, Full Coverage, and Research Annual Industrial Files for each year that
any of the data items were available. We start our sample period in 1974 because the data items
comprising net capital outflows are unavailable from Compustat prior to 1971. We set aside
− 73 so as to limit our tests to a single standard setting regime, the FASB. Certain logical
1971−
checks were then applied. For example, price per share at fiscal year-end and number of shares
had to be positive, and cash dividends had to be non-negative. If a logical check was not met,
the data item was set missing.
Our main regressions are run with variables expressed in undeflated millions of dollars.4
In the spirit of Kothari and Zimmerman (1995), Collins, Maydew and Weiss (1997) and Fama
and French (1998), for each variable we mitigate the effects of outliers by setting observations to
missing that were in their extreme top and bottom one percentile. This procedure is applied on a
year-by-year basis separately for four groups of firms comprising dividend and non-dividend
paying firms and firms with positive and negative core earnings. We adopt this trimming
procedure rather than simply trimming year-by-year because our primary empirical tests
comparing the predictions of the profitability signaling and free cash flow mitigation hypotheses
partition the data set into these four subgroups. The resulting data set was then “shrunk” by

8
requiring core earnings, equity book value, dividends and other net capital outflows all to be
non-missing. Because Compustat does not report the data items comprising net capital outflows
for banks, utilities, life insurance, or property and casualty companies, such firms are excluded
from our sample. The final data set consists of 103,932 firm-year observations for NYSE,
AMEX and NASDAQ firms over 23 years, an average of 4,519 per year.
Table 1 defines the primary variables used in our study in terms of their Compustat data
items. Following Collins, Maydew and Weiss (1997), our tests use core net income, CORENI,
as the measure of earnings. CORENI aims to capture firms’ earnings before one-time, non-
recurring items on an after-tax basis, ONE, and is defined as income before extraordinary items
(Compustat item 18). The variable ONE is bottom line net income, NI, less CORENI. Our
motivation for using CORENI instead of NI is that we expect one-time items to be transitory and
hence have a smaller pricing multiple than CORENI (Ohlson, 1999). The market value of
common equity is computed at the fiscal year-end.5 The measure of net capital contributions,
NETCAP, is not exact because per Compustat’s data definition it unavoidably includes the
issuance and repurchase of preferred in addition to common stock. However, it is important to
note that NETCAP excludes DIV. That is, the sum of NETCAP and DIV equals net dividends.
Table 2, panels A and B, reports descriptive statistics for these variables. Panel A reports
year-by-year means together with certain percentages. The number of usable observations per
firm-year doubles over the 23-year sample period, rising from 3,235 in 1975 to 6,330 in 1996.
Other trends include an almost ten-fold increase in mean nominal firm size, and steadily
increasing mean core earnings CORENI, equity book value BV, and dividends DIV. The
dominance of negative means for NETCAP indicates that firms have tended to issue more
common and preferred stock than they have repurchased. Dirty surplus items DIRT have on
average become reductions in equity (debits) in recent years, and very common.
Panel A also shows that the percentage of firms paying dividends has plummeted from a
high of 72% in 1977 to only 31% in 1996, while between 14% and 29% of firms have one-time
items in any given year. As brought to light by Hayn (1995), the percentage of firms with losses,
as defined by negative core (not bottom line) net income, rose dramatically from a low of 9% in
1978 to the mid-30% range by 1985 and has remained around that level ever since. A similar

4
In section 4.5, we demonstrate that inferences from undeflated regressions are robust to deflating by shares.

9
pattern over time exists for firms with negative common equity.6 The percentage of firms with
dirty surplus items DIRT has increased steadily from 48% in 1974 to 81% in 1996. This is partly
attributable to the increasing incidence of stock option exercise.
Panel B of table 2 reports the means of year-by-year distributional statistics. I/B/E/S year
ahead earnings forecasts are more positive because I/B/E/S forecasts are more common for larger
firms. Panel C reports the mean year-by-year Pearson and Spearman correlations among the
same variables. Equity market value MVE is highly positively correlated with BV, CORENI and
DIV, each of which is strongly positively correlated with each other. The high positive
correlations among DIV, BV and CORENI suggests that omitting dividends from accounting-
based valuation regressions will likely result in upward biased coefficients on BV and CORENI.
Panel D of table 2 reports the industry composition of our sample. We employ the same
industry classifications as Barth, Beaver and Landsman (1998), based on the primary SIC code
reported by Compustat. Since Compustat only reports the most recent primary SIC code, not
historical SIC codes, the accuracy of industry classifications likely deteriorates as one goes back
in time. As would be expected, the industries with the largest concentrations of firm-year
observations are durable manufacturers and retailers. Reflecting the fact that Compustat does not
report the data items comprising net capital outflows for banks, utilities, life insurance, or
property and casualty companies, the fraction of firm-year observations represented by financial
institutions (#12) and insurance companies (#13) is lower than in the population.

4. Estimating equations

In this section we report the results of estimating the equations describing Ohlson’s linear
information dynamics and the pricing of common equity. Almost all equations are estimated on
a year-by-year basis using industry fixed effects, so the intercept term actually denotes a vector
of industry intercepts.7 The number of year-by-year regressions varies depending on the number
of lags or leads in the dependent and/or independent variables, and whether the regression

5
Based on other studies, we would not expect that using market value computed three months after the fiscal year-
end would materially affect our results or inferences (e.g., Barth, Beaver and Landsman, 1992, 1996).
6
We differ slightly from Collins, Maydew and Weiss (1997) in that we do not exclude firm-year observations if
equity book value is negative. Also, the year-by-year means reported in panel A of table 1 likely understate the true
percentage of firms with negative equity book value since in each year the top and bottom 1% of each variable have
been set missing.
7
Although inferences do not materially change when industry fixed effects are omitted, their inclusion generally
yields lower standard errors on slope variables.

10
directly or indirectly contains forecasted one year ahead CORENI. In particular, I/B/E/S
forecasts are only available beginning in 1984, limiting regressions directly or indirectly
containing forecasted one year ahead CORENI to a maximum of 12 years.
Our regressions use unscaled data expressed in millions of dollars. Later tests assess the
robustness of our inferences to deflating by the number of common shares outstanding at fiscal
year-end. Although unscaled data are heteroscedastic, year-by-year estimation should yield a
series of unbiased regression coefficient estimates (Barth and Kallapur, 1996). Following Fama
and MacBeth (1973), standard errors for the mean slope parameter estimates are taken from the
time-series standard deviations of the yearly slopes. Since the error terms from the annual
regressions may be autocorrelated, significance levels based on the Fama-MacBeth approach
may be overstated. To err on the side of caution, we adopt the Fama and French (1998)
approach of allowing for a 50% understatement of standard errors by only taking a mean t-
statistic exceeding 3.0 in absolute magnitude as strong evidence of statistical significance.

4.1 Linear information dynamics

Since the predicted coefficients on dividends in equations (4) and (6) depend on the linear
information dynamics parameters ω and γ, we estimate those first using empirical analogues to
equations (2) and (3). The results are shown in table 3. Panel A reports the Pearson and
Spearman correlations among the dependent and independent variables; panels B and C report
regressions estimating ω and γ separately for dividend and non-dividend paying firms. For
dividend and non-dividend paying firm-year observations, other information v in equation (5) is
computed using the estimates ω1 = 0.61 and ω1 = 0.46 obtained in the first and sixth regressions
in panel C, respectively. Following Dechow, Hutton and Sloan (1998) abnormal earnings
ACORENI is computed for all firm-year observations using r = 0.12.8
Of particular note in panel A is that v and DIV are positively correlated, in contrast to the
assumption of the Ohlson model that ∂v/∂DIV = 0. In panel A, the mean year-by-year Pearson
(Spearman) correlation between v and DIV is 0.33. This opens the door to dividends potentially
being a component of, or proxy for, v. Panel B indicates that v is more positively first-order

8
We recognize that per equations (2) and (3), v depends on ω1, and ω1 depends on v. We are currently exploring
ways of dealing with this simultaneity, but do not expect it to materially affect our inferences, as panel A of table 3
reveals the correlation between ACORENI and v is small.

11
autocorrelated for dividend paying than for non-dividend paying firms, with γ1 = 0.45 for the
former versus γ1 = 0.34 for the latter in the simplest regression.9 Also, for dividend paying firms,
dividends are significantly predictive of one year ahead v, with only a small impact on the
magnitude or significance of γ1. The ability of dividends to predict one year ahead v remains
even after controlling for lagged book equity, BV-1, which per Feltham and Ohlson (1995) and
Dechow, Hutton and Sloan (1998) is our empirical proxy for conservative accounting.
Table 3, panel C presents parameter estimates for the equation governing the evolution of
one year ahead abnormal earnings, ACORENI+1. We note the following results. First, similar
to Dechow, Hutton and Sloan (1998), abnormal earnings are first-order autoregressive, with ω1 ≈
0.60 for dividend and non-dividend paying firms.10 AR(2) parameter estimates ω2 are small and
insignificant once v is included as a control. Second, the second and third regressions indicate
that current dividends are incrementally relevant in predicting ACORENI+1, particularly after
controlling for conservative accounting via lagged book equity BV−1. Third, the coefficient on
BV−1 is significantly negative for non-dividend paying firms. This may be because many non-
dividend paying firms are young firms that are growing by investing heavily in R&D. Fourth,
although the coefficient on other information v is always reliably positive, it is also always
significantly less than the +1 value assumed by the linear dynamics in equation (2). This may be
attributable to measurement error in v induced by I/B/E/S forecasts of expected one period ahead
earnings being noisy and/or biased. Finally, v soaks up all the explanatory power in dividends:
controlling for v yields an insignificant ω4 coefficient estimate on DIV in the fifth regression.
Taken together, the results thus far indicate that dividends are correlated with v and can
help predict next period abnormal earnings. As a result, they should be able to predict cross-
sectional variation in equity values over and above a dividend displacement role. However, we
expect this ability to be most pronounced where other information v is (set to) zero than where v
follows an AR(1) dynamic, or where one or more of the assumptions underlying the Ohlson
model, such as no information asymmetry and market efficiency, are violated.

4.2 Basic tests on the pricing of dividends into the market value of common equity

Our tests on the pricing of dividends into the market value of common equity, MVE, are

9
For all firms pooled across time, Dechow, Hutton and Sloan (1998) report the estimate γ1 = 0.32.

12
structured around the following OLS regressions:

MVEit = a0t + a1t BVit + a2t CORENIit + a3t DIVit + a4t NETCAPit + eit (7a)

MVEit = c0t + c1t BVit + c2t CORENIit + c3t DIVit + c4t NETCAPit + c5t E[ NI + 1]it + eit (7b)

Equations (7a) and (7b) are the empirical analogues to equations (4) and (6), respectively, with
the additional restriction in equation (7a) that other information v = 0. We make this restriction
for three reasons. First, the results in table 3 lead us to expect that the pricing of dividends over
and above dividend displacement will be strongest in regressions where v is set to zero. Second,
a comparison of the adjusted R2 statistics across equations (7a) and (7b) readily provides one
measure of the statistical relevance of v. Third, most empirical work based on the Ohlson model
has set v = 0 and has ignored the pricing of dividends by implicitly assuming that a3 = a4 = 0.
Thus, a comparison of the adjusted R2 statistics and parameter estimates on BV and CORENI in
equation (7a) as shown versus equation (7a) with the restriction that a3 = a4 = 0 provides one
assessment of the impact of omitting net dividends in prior studies. So as to focus on dividends,
we break net dividends d into two parts: common equity dividends, DIV, and all other net
common equity capital contributions, NETCAP.11 As with the information dynamics, equations
related to (7a) and (7b) are estimated on a year-by-year basis using industry fixed effects, so the
intercept term actually denotes a vector of industry intercepts.
The results of estimating the pricing of DIV into MVE via equations (7a) and (7b)
separately for dividend and non-dividend paying firms are reported in tables 4 and 5. For each
independent variable, we provide both coefficient sign and magnitude predictions. The latter are
obtained by substituting from table 3 the parameter estimates ω1 = 0.61 and γ1 = 0.45 for
dividend paying firms and ω1 = 0.46 and γ1 = 0.34 for non-dividend paying firms into equations
(4.1)−(4.2) and (6.1)−(6.5). In addition to individual coefficient estimates, each table also
reports the linear combination of coefficients comprising the empirical estimate of ∂MVE/∂DIV,
the partial derivative of the market value of equity with respect to dividends. In equation (7a)
where v = 0, ∂MVE/∂DIV = −a1 + a3, while in equation (7b) where v follows an AR(1) dynamic,

10
The sole exception is the first regression for non-dividend paying firms where ω1 = 0.46.
11
As noted above, NETCAP is not exact because per Compustat’s data definition it unavoidably includes the
issuance and repurchase of preferred in addition to common stock. However, because preferred is much less

13
∂MVE/∂DIV = −[c1 − c3 + (c5 x r)].
From table 4 we begin by noting that although table 3 indicates BV, CORENI and DIV are
highly positively correlated, omitting DIV for dividend paying firms has a relatively small impact
on a1 and a2, the coefficients on BV and CORENI. This suggests that prior studies that have
omitted DIV and NETCAP are unlikely to result in significant bias in inferences drawn from the
coefficients on BV and CORENI.
More importantly for our study, table 4 indicates that the dividend displacement property
is firmly rejected. First, the coefficient a3 on DIV is a reliably positive 3.86 (t-statistic = 6.2),
and clearly significantly greater than its predicted value of –0.14. Second, ∂MVE/∂DIV = 3.28,
with a t-statistic of 6.8 relative to its predicted value of –1. The magnitudes of the deviations
from coefficient predictions suggests that the stock prices of dividend paying firms reflect
economically as well as statistically significant departures from dividend displacement. It is also
noteworthy that although the mean adjusted R2 statistics of dividend and non-dividend paying
firms indicate good model fits (exceeding 50% and 80%, respectively), coefficient predictions
are typically rejected for both dividend and non-dividend paying firms. This may mean that part
of the rejection of the dividend displacement property comes from violations of Ohlson model
other than those leading to the rejection of dividend displacement.12
Similar inferences concerning dividend displacement to those from table 4 are found in
table 5 where measures of v are integrated into the regressions reported in table 4. Using I/B/E/S
forecasts of one year ahead earnings, the coefficient c3 on DIV is a reliably positive 1.88 (t-
statistic = 5.8), and also significantly greater than its predicted value of 0.08 (t-statistic relative to
a null of 0.08 = 5.6). If rational expectations are invoked instead and actual one year ahead
earnings are used as proxy for forecasted one year ahead earnings, the coefficient c3 on DIV
increases to 2.49 (t-statistic = 4.8) and is also reliably greater than predicted (t-statistic relative to
a null of 0.08 = 4.7). Turning to the partial derivative of the market value of equity with respect
to dividends, the evidence indicates the null of ∂MVE/∂DIV = −1 under dividend displacement is
soundly rejected using either I/B/E/S forecasts of one year ahead earnings (∂MVE/∂DIV = 0.65;

common than common stock, this source of measurement error in NETCAP is likely to have negligible effects on
NETCAP’s coefficient and those of the other included variables.
12
Similar coefficient estimates are obtained if a pooled time-series cross-section approach with year and industry
dummies is used. However, t-statistics on coefficient estimates are much larger, probably as a result of lack of
independence in the residuals.

14
t-statistic relative to the predicted value of –1 = 5.6) or actual one year ahead earnings
(∂MVE/∂DIV = 1.60; t-statistic relative to a null of –1 = 4.9).
Consistent with our expectation based on table 3 that the ability of dividends to explain
cross-sectional variation in equity values over and above dividend displacement will be most
pronounced when v = 0 than when v follows an AR(1) dynamic, ∂MVE/∂DIV is at least twice as
large in table 4 as it is in table 5 (3.28 versus 0.65 and 1.60). However, even when v is fully
included per equation (7b) via E[NI+1], DIV still has significant explanatory power above its
predicted dividend displacement role. This may indicate that the I/B/E/S proxy of a full
information forecast of one year ahead, F[NI+1], is systematically less than entirely rational. If
F[NI+1] is rational and the other assumptions in the Ohlson model are met, then the coefficient
c2 on CORENI in equation (7b) should be nonpositive, in contrast to the nonnegative prediction
for a2 in equation (7a). Although a2 in table 4 is indeed significantly positive, so is c2 in table 5,
and more so for dividend paying than for non-dividend paying firms. The finding of a positive c2
in table 5 would not be expected if our proxies for E[NI+1] were simply noisy: noisy proxies
would lead to estimates of c2 that were less negative than predicted, but not positive. At the
same time, it is likely that our proxies for E[NI+1] are noisy, since the increase in adjusted R2
statistics going from table 4 to table 5 are modest. For dividend paying (non-dividend paying)
firms, adjusted R2 statistics increase at most by 4% (5%).

4.3 Hypotheses for the pricing of dividends in equity valuation beyond dividend displacement

In this section we propose and test hypotheses for why dividends are positively priced in
the market value of common equity, diverging both statistically and economically from the one-
to-one negative relation predicted by dividend displacement. Both hypotheses capitalize on the
possibility that dividends signal private information held by managers. A significant body of
research exists in finance around the proposition that because of information asymmetries
between managers and investors, both the level of and change in a firm’s dividend may be
positively related to its equity value. In contrast, the Ohlson (1995, 1998) and Feltham and
Ohlson (1995, 1996) models assume that there are no information asymmetries, and therefore no
role for dividends and/or financial structure as signaling devices.13

13
Feltham and Ohlson (1995), Ohlson (1995), Lundholm (1995, p. 744-745), and the Coopers & Lybrand Academic
Advisory Committee (1997, p.11) are all well aware of the restriction that the no-information-asymmetry

15
Finance research has proposed and tested two alternative, although not necessarily
mutually exclusive, hypotheses that yield a positive relation between the level of and change in a
firm’s dividend and its equity value, each of which can be viewed in the context of information
asymmetry between managers and investors. The first is Jensen’s (1986) free cash flow theory,
which proposes that managers with cash flow over that needed to fund normal investing
activities (“free cash flow”) may waste it by investing it in negative NPV projects or spending it
on themselves rather than distributing it to shareholders. The model assumes that managers hold
private information about their willingness to impose such agency costs, and that there are
sufficient frictions to prevent “high-quality” managers who waste low amounts of free cash flow
from costlessly distinguishing themselves from “low-quality” managers who waste high amounts
of free cash flow. The second hypothesis that yields a positive relation between dividends and
equity value is the profitability signaling argument developed by Bhattacharya (1979), Kalay
(1980), and Miller and Rock (1985), in which managers use dividends to signal private
information about current and/or future profitability to investors.14

4.4 Tests of the free cash flow and profitability signaling hypotheses

We test the predictions of the free cash flow and profitability signaling hypotheses by
partitioning our sample into four mutually exclusive groups: DIV > 0 and CORENI > 0, DIV > 0
and CORENI < 0, DIV = 0 and CORENI > 0, and DIV = 0 and CORENI < 0. We partition in this
manner because although both hypotheses predict a positive relation between stock prices and
dividends, the free cash flow hypothesis predicts that the pricing of dividends will be smaller for
firms with negative current period earnings than for firms with positive current period earnings.
Under the free cash flow explanation, dividends are more highly valued when free cash flow is
greatest, which we propose is more likely to be the case when current period earnings are
positive, and less likely to arise when firms are losing money. In sharp contrast, the profitability-
signaling hypothesis makes the opposite prediction. Under profitability signaling, dividends are
less (more) valued when current period earnings are positive (negative), because then paying
dividends is less (more) costly and thus a less (more) credible signal of good future earnings.

assumption places on the model, as well as the fact that analytically the model could at some cost be expanded to
account for this reality.
14
In addition to dividends, it is also possible that firms use one or both components of NETCAP (stock repurchases
and stock sales) as signaling devices. To keep the paper focused, we limit our analysis to dividends alone.

16
Tables 6 and 7 present findings from tests of these differential predictions in which the
same regressions as in tables 4 and 5 are estimated but on the partitioned data. In panel B of
table 7 we diverge from tables 4−6 by adopting a pooled time-series cross-section method (with
both year and industry dummies included but not reported). This is because the number of
available observations per year in the partition where DIV > 0 and CORENI < 0, and I/B/E/S
forecasts of one year ahead earnings are available is small. Although there are other notable
differences in coefficients across firms with positive and negative CORENI, particularly on
CORENI itself, we focus on the coefficients on DIV, noting the following.
First, regardless of whether v is set to zero (table 6) or allowed to follow an AR(1)
dynamic (table 7), the estimated coefficients on DIV are always reliably positive, and with the
possible exception of the partition where CORENI > 0 and E[NI+1] = F[NI+1], reliably larger
than their predicted values. Second, the estimated coefficients on DIV when CORENI < 0 are
always at least 2½ times larger than those for which CORENI > 0. Third, ∂MVE/∂DIV is always
significantly greater than –1, and always at least 3 times larger when CORENI < 0 than when
CORENI > 0. The direction of the second and third set of findings are consistent with the
profitability signaling hypothesis, and inconsistent with the free cash flow mitigation hypothesis.
However, the fact that when CORENI is positive we observe reliably positive mean coefficient
estimates on DIV, as well as ∂MVE/∂DIV that are significantly greater than –1, suggests that for
firms with CORENI > 0 investors may be rewarding managers for wasting free cash flow less.
If dividends mitigate information asymmetry by being a credible signal of future
profitability when current earnings are negative, they can be viewed as a component of, or a
proxy for, other information v in the Ohlson model. If so, then the role of dividends in
augmenting the linear information dynamics should be more pronounced when current earnings
are negative than when they are positive. Table 8 reports the results of re-estimating the linear
dynamics regressions shown in panels B and C of table 3 when the sample is limited to dividend
paying firms only, but partitioned into firm-years when CORENI > 0 versus firm-years when
CORENI < 0. The findings in table 8 are consistent with the prediction of the profitability
signaling hypothesis. For example, in panel A the coefficient γ2 on DIV is more than 4 times
larger when CORENI > 0 as compared to when CORENI < 0. In panel B, prior to controlling for
v, the coefficient ω4 on DIV is almost 4 times larger when CORENI > 0 as compared to when
CORENI < 0. Similarly to panel C of table 3, however, controlling for v drives out the

17
explanatory power of DIV regardless of the sign of CORENI.

4.5 Robustness tests

Table 9 reports two types of robustness tests, where the data are limited to dividend
paying firms only. Panel A examines the sensitivity of the basic results in tables 4 and 5 to
adding two and three year ahead CORENI; to adding CORENI lagged one year (Subramanyam
and Venkatachalam, 1998); and to including current year one-time and dirty surplus income
statement items. The latter two regressions have no effect on the inference that dividends is
priced into equity value more positively than predicted under dividend displacement. The first
regression indicates that adding two and three year ahead CORENI has a marginal impact, in that
the magnitudes of both the coefficient estimate and associated t-statistic on DIV decline from
2.49 (t-statistic = 4.8) in panel A of table 5 with E[NI+1] = CORENI+1 to 2.11 (t-statistic = 2.9).
Given that we allow for possible inflation of t-statistics arising from autocorrelation in the year-
by-year parameter estimates by only taking a mean t-statistic exceeding 3.0 in absolute
magnitude as strong evidence of statistical significance, the addition of CORENI+2 and
CORENI+3 yields only a marginally significant positive coefficient estimate on DIV in panel A
of table 9. However, this is quite consistent with the profitability signaling hypothesis: if current
period dividends signal future profitability, then controlling for future profitability should
dampen the predictive role of DIV.
Panel B reports a sample of regressions where all variables (except the intercepts) are
deflated by the number of common shares outstanding at the firm’s fiscal year-end. The concern
is sometimes raised that undeflated regressions in particular may merely reflect scale bias, rather
than economic substance (Barth and Clinch, 1998; Easton, 1998). Although still a matter of
some controversy, to address this concern we apply the deflator most commonly used in
empirical studies based on the Ohlson model, namely shares outstanding. The results in panel B
indicate that relative to tables 4 and 5, per-share regressions yield more positive coefficient
estimates on DIV as well as more positive partial derivatives of the market value of equity with
respect to dividends.

18
6. Summary and concluding remarks

In this paper we have employed Ohlson’s (1995, 1998) accounting-based equity valuation
model to structure an empirical assessment of the pricing of dividends in stock prices. We
addressed two questions. First, to what extent does the pricing of dividends reflect Modigliani
and Miller’s (1958, 1961) one-to-one displacement property? Second, what explains the
direction and magnitude of any divergence from dividend displacement? Using annual cross-
sections of NYSE, AMEX and NASDAQ firms over the period 1974-1996, we found robust
evidence that dividends are materially positively priced, sharply contrasting with the negative
relation predicted by dividend displacement. We also found that the positive pricing of
dividends is at least three times larger for loss firms than for profit firms. Our explanation for
these results is that managers of loss firms use dividends to signal future profitability, while to a
lesser degree managers of profit firms use dividends to alleviate concerns about the misuse of
free cash flow. It seems that dividends are a component of, and rich proxy for, other information
about future abnormal earnings that is reflected in price but is not yet captured by current
financial statements.

19
References

Amir, E., and B. Lev, 1996, Value-relevance of nonfinancial information: The wireless
communication industry, Journal of Accounting and Economics 22: 3-30.
Auerbach, A., 1979, Share valuation and corporate equity policy, Journal of Public Economics
11, 291-305.
Barth, M.E., W.H. Beaver and W.R. Landsman, 1992, The market valuation implications of net
periodic pension cost, Journal of Accounting and Economics 15, 27–62.
Barth, M.E., W.H. Beaver and W.R. Landsman, 1996, Value-relevance of banks’ fair value
disclosures under SFAS No. 107, The Accounting Review 71, 513-537.
Barth, M.E., W.H. Beaver and W.R. Landsman 1998, Relative valuation roles of equity book
value and net income as a function of financial health, Journal of Accounting and Economics
25; 1-34.
Barth, M.E, and G. Clinch, 1998, Revalued financial, tangible, and intangible assets:
Associations with share prices and non market-based value estimates, Journal of Accounting
Research (forthcoming).
Barth, M.E, and S. Kallapur, 1996, Effects of cross-sectional scale differences on regression
results in empirical accounting research, Contemporary Accounting Research 13, 527-567.
Bernheim, B.D., 1991, Tax policy and the dividend puzzle, Rand Journal of Economics 22, 455-
476.
Bhattacharya, S., 1979, Imperfect information, dividend policy, and the ‘Bird in the hand’
fallacy, Bell Journal of Economics, 10, 259-270.
Collins, D.W., Maydew, E.L., and I.S. Weiss, 1997, Changes in the value-relevance of earnings &
equity book values over the past forty years, Journal of Accounting and Economics 24, 39-67.
Collins, J.H., and D. Kemsley, 1999, Capital gains taxes in firm valuation and corporate financial
policy, working paper, University of North Carolina.
Coopers & Lybrand Academic Advisory Committee, 1997, Evaluating financial reporting
standards, working paper.
Dechow, P.M., Hutton, A.P., and R.G. Sloan, 1999, An empirical assessment of the residual
income valuation model, Journal of Accounting and Economics 26, 1-34.
Easton, P.D., 1998, Discussion of Revalued financial, tangible, and intangible assets: association
with share prices and non-market-based value estimates. Journal of Accounting Research 36,
Supplement, 235-247.
Fama, E.F., and K.R. French, 1998, Taxes, financing decisions, and firm value, Journal of
Finance 53, 819-843.
Fama, E.F., and J.D. MacBeth, 1973, Risk, return, and equilibrium: empirical tests, Journal of
Political Economy 81, 607-636.

20
Feltham, G.A., and J.A. Ohlson, 1995, Valuation and clean surplus accounting for operating and
financial activities, Contemporary Accounting Research 11, 689-732.
Feltham, G.A., and J.A. Ohlson, 1996, Uncertainty resolution and the theory of depreciation
measurement, Journal of Accounting Research 34, 2, 209-234.
Frankel, R., and C. Lee, 1998, Accounting valuation, market expectation, and the book-to-market
effect, Journal of Accounting and Economics 25, 3, 283-319.
Harris, T.S., and D. Kemsley, 1999, Dividend taxation in firm valuation, Journal of Accounting
Research, forthcoming.
Harris, T.S., R.G. Hubbard, and D. Kemsley, 1998, The share price effects of dividend taxes and
tax imputation credits, working paper, Columbia University.
Hayn, C., 1995, The information content of losses, Journal of Accounting and Economics 20,
125-133.
Jensen, M., 1986, Agency cost of free cash flow, corporate finance, and the market for takeovers,
American Economic Review, 323-329.
Kalay, A., 1980, Signaling, information content, and the reluctance to cut dividends, Journal of
Financial and Quantitative Analysis, 15, 855-869.
Kothari, S.P., and J. Zimmerman, 1995, Price and return models, Journal of Accounting and
Economics 20, 155-192.
Lee, C., J. Myers, and B. Swaminathan, 1999, What is the intrinsic value of the Dow?, Journal of
Finance, forthcoming.
Litzenberger, R., and K. Ramaswamy, 1979, The effects of personal taxes and dividends on
capital asset prices: theory and empirical evidence, Journal of Financial Economics 7; 163-
195.
Lundholm, R.J., 1995, A tutorial on the Ohlson and Feltham/Ohlson models: Answers to some
frequently asked questions, Contemporary Accounting Research, 749-761.
Miller, M.H., and K. Rock, 1985, Dividend policy under asymmetric information, Journal of
Finance, 40, 1031-1051.
Miller, M.H., and M. Scholes, 1978, Dividends and taxes, Journal of Financial Economics 6,
333-364.
Miller, M.H., and M. Scholes, 1982, Dividends and taxes: some empirical evidence, Journal of
Political Economy 90, 1108-1142.
Miller, M., and Modigliani, F., 1961, Dividend policy, growth and the valuation of shares,
Journal of Business 34, 411-433.
Modigliani, F., and M. Miller, 1958, The cost of capital, corporation finance and the theory of
investment, American Economic Review 48, 261-297.
Ohlson, J.A., 1995, Earnings, equity book values, and dividends in equity valuation,

21
Contemporary Accounting Research, 661-687.
Ohlson, J.A., 1998, Earnings, equity book values, and dividends in equity valuation: an empirical
perspective, working paper, Columbia University.
Ohlson, J.A., 1999, On transitory earnings, Review of Accounting Studies, forthcoming.
Poterba, J.M., and L.H. Summers, 1984, New evidence that taxes affect the valuation of
dividends, Journal of Finance 39, 1397-1415.
Poterba, J.M., and L.H. Summers, 1985, The economic effects of dividend taxation, In E. Altman
and M. Subrahmanyan, eds., Recent advances in corporate finance. Homewood: R.D. Irwin,
1985.
Subramanyam, K.R., and M. Venkatachalam, 1998, The role of book value in equity valuation:
Does the stock variable merely proxy for relevant past flows?, working paper, USC.

22
Table 1

Data definitions

Our Ohlson
Variable (in $ millions) label notation Compustat annual data items, or computation details
Market value common equity MVE P 199 (fiscal year-end closing price) x 25 (common shares
outstanding at fiscal year-end).
Net income NI x 172 (net income or loss).
Forecasted one year ahead NI F[NI+1] Et[xt+1] [From I/B/E/S] Forecast of primary EPS before extraordinary
items for year t+1 x common shares at the time the forecast was
made. The latter is the date of the forecast made closest to before
the date that annual earnings for year t were reported. Thus, if year
t ended 941231 and earnings for year t were reported on 950121,
our forecast would be prior to 950121, e.g., 950114. If there was
no report date for annual earnings in year t, we took the forecast
closest to 3 months after the fiscal year end for year t, e.g., 950320.
Core earnings CORENI -- 18 (income before extraordinary items).
One year ahead core earnings CORENI+1 -- Per CORENI, if available.
One-time items ONE -- NI – CORENI.
Book value of common equity BV b 60 (total common equity).
Common dividends DIV 1st part of da 21 (common dividends declared).
nd
Net capital contributions NETCAP 2 part of d 115 (purchase of common + preferred stock) – 108 (sale of
common and preferred stock). Each component of NETCAP is
available beginning 1971.
Dirty surplus items DIRT -- BV – BV{one year lagged} – NI + DIV + NETCAP. To allow for
rounding errors, DIRT = 0 if abs{DIRT} < $10,000.

a
d = DIV + NETCAP.
25
Table 2

Descriptive statistics on key variablesa over the sample period 1974-1996

Panel A: Yearly means (in $ millions) and percentages

. % with % with % with %with %with


Year # obs. MVE CORENIb F[NI+1] ONE BV DIV NETCAP DIRT DIV>0 ONE≠0 CORENI<0 BV<0 DIRT≠0
1974 3,235 $ 90 $ 12 n.a. $ -0.0 $ 99 $ 4.3 $ -1.1 $ -1.0 64% 21% 11% 0.6% 48%
1975 3,234 124 13 n.a. 0.0 110 4.7 -2.2 -1.6 65 22 14 0.6 49
1976 3,297 149 15 n.a. 0.1 122 5.2 -2.5 -0.7 69 19 11 1.2 49
1977 3,301 149 17 n.a. 0.0 135 6.2 -2.3 -1.1 72 20 10 1.2 51
1978 3,422 151 19 n.a. 0.1 141 6.6 -2.5 -1.5 70 20 9 1.5 55
1979 3,619 167 21 n.a. 0.3 145 7.2 -2.9 -1.9 64 21 13 1.4 52
1980 3,800 197 20 n.a. -0.0 146 7.3 -4.6 -1.1 59 21 17 2.1 50
1981 4,351 172 19 n.a. 0.1 144 7.4 -4.0 -2.6 52 21 21 2.8 56
1982 4,313 196 16 n.a. 0.0 150 7.7 -3.7 -3.3 49 21 29 3.7 55
1983 4,589 246 17 n.a. 0.0 160 8.0 -7.5 -2.4 44 21 28 3.3 58
1984 4,698 248 21 55 0.0 177 8.8 -1.3 -3.8 43 22 28 3.4 59
1985 4,637 302 18 50 -0.2 179 9.5 -2.7 1.7 41 23 34 4.4 59
1986 4,829 351 18 51 0.3 183 10.1 -3.4 2.0 38 26 36 4.3 59
1987 4,956 357 25 62 0.5 203 11.2 -1.4 2.3 37 28 35 4.5 60
1988 4,783 397 30 69 0.7 224 12.9 1.8 -1.1 37 27 36 5.2 64
1989 4,654 500 33 73 0.5 250 14.8 -1.0 -4.8 38 23 37 5.7 66
1990 4,586 464 30 70 0.2 266 15.2 1.7 4.4 37 21 37 6.0 67
1991 4,723 560 23 65 -0.6 280 15.6 -5.6 0.5 36 22 37 5.5 69
1992 5,019 564 24 60 -5.3 271 14.6 -7.0 -4.6 35 27 35 4.5 74
1993 5,439 589 23 58 -1.5 259 13.0 -11.1 0.5 34 29 34 3.2 76
1994 5,804 579 32 69 -0.3 273 14.0 -4.7 3.2 33 17 31 2.9 79
1995 6,293 708 37 76 -0.3 298 14.5 -4.0 7.9 32 14 34 4.5 80
1996 6,330 785 36 n.a. -0.1 306 15.2 -5.2 3.1 31 14 34 4.0 81

26
Table 2 (continued)

Panel B: Distributional statistics

Each number is the mean of 23 year-by-year statistics (except F[NI+1], which is mean of 12)

Lower Upper
Variable (in $ millions) Label # obs. Min. quartile Med. Mean quartile Max.
Market value of equity MVE 4,519 0. 11 41 350 179 15,597
Core earnings CORENI 4,519 -620 -0. 2 23 12 1,126
I/B/E/S year t+1 forecast F[NI+1] 1,219 -140 4 12 63 45 1,731
Year t+1 CORENI CORENI+1 3,853 -428 -0. 2 26 14 1,192
One-time items ONE 4,451 -216 0 0 -0. 0 116
Book value of equity BV 4,519 -125 7 26 197 105 8,405
Dividends declared DIV 4,519 0 0 0. 10 3 581
Net capital contributions NETCAP 4,519 -344 -2 -0. -3 0. 287
Dirty surplus items DIRT 3,974 -451 -0. 0. -0. 0. 344

Panel C: Correlations among independent and dependent variablesc

Each correlation below is the mean of 23 individual year-by-year correlationsd

Variable MVE CORENI F[NI+1] CORENI+1 ONE BV DIV NETCAP DIRT


MVE 0.86 0.91 0.88 0.00 0.86 0.82 -0.08 -0.08
CORENI 0.76 0.89 0.90 0.01 0.84 0.83 -0.07 -0.11
F[NI+1] 0.90 0.87 0.90 -0.02 0.87 0.83 0.10 0.02
CORENI+1 0.73 0.82 0.83 0.00 0.86 0.83 -0.12 -0.12
ONE -0.06 -0.03 -0.05 -0.05 0.02 0.00 0.03 0.01
BV 0.90 0.77 0.88 0.75 -0.06 0.85 -0.14 -0.13
DIV 0.71 0.69 0.75 0.68 -0.09 0.75 -0.16 -0.19
NETCAP -0.21 -0.07 0.05 -0.07 0.02 -0.09 -0.01 0.35
DIRT 0.02 -0.02 0.06 -0.02 0.01 -0.02 -0.06 0.08

a
Variable definitions are per table 1.
b
Positive values of CORENI, F[NI+1], CORENI+1, ONE, BV and DIRT are credits. Positive values of DIV and
NETCAP are debits.
c
For most correlations, the average number of observations is 4,519 per year. Pearson (Spearman) correlations are
above (below) the diagonal.
d
For correlations involving F[NI+1], there are 12 year-by-year correlations (1985-1996). For correlations
involving CORENI+1, there are 22 year-by-year correlations (1974-1995).

27
Table 2 (continued)

Panel D: Industry composition

# Industry Primary SIC codes # obs. % obs.


1. Mining and construction 1000 − 1999 3,634 3.5%
2. Food 2000 − 2111 2,996 2.9
3. Textiles, printing/publishing 2200 – 2780 7,537 7.3
4. Chemicals 2800 – 2824, 2840 – 2899 2,763 2.7
5. Pharmaceuticals 2830 – 2836 2,785 2.7
6. Extractive industries 2900 – 2999, 1300 – 1399 5,747 5.5
7. Durable manufacturers 3000 – 3999, excluding 3570 – 3579
and 3670 – 3679. 30,090 29.0
8. Computers 7370 – 7379, 3570 – 3579, 3670 – 3679 6,160 5.9
9. Transportation 4000 – 4899 4,817 4.6
10. Utilities 4900 – 4999 5,144 4.9
11. Retail 5000 – 5999 13,021 12.5
12. Financial institutions 6000 – 6411 4,488 4.3
13. Insurance and real estate 6500 – 6999 5,237 5.0
14. Services 7000 – 8999, excluding 7370 – 7379 9,513 9.0
TOTAL 103,932 100.0%

28
Table 3
Mean coefficients and associated t-statistics from annual cross-sectional OLS
regressions testing Ohlson’s (1995) modified AR(1) linear information dynamics.
Industry dummies are included but not reported. Time period is 1974-1996.

Panel A: Correlations among independent and dependent variablesa


Each correlation below is the mean of individual year-by-year correlationsb
Variable ACORENI+1 ACORENI ACORENI-1 v+1 v DIV BV-1
ACORENI+1 0.64 0.48 0.12 0.26 0.33 0.19
ACORENI 0.64 0.64 0.16 -0.09 0.34 0.18
ACORENI-1 0.48 0.65 0.19 0.14 0.42 0.29
v+1 0.18 0.26 0.24 0.47 0.35 0.36
v 0.32 0.16 0.22 0.44 0.33 0.34
DIV 0.27 0.32 0.37 0.21 0.21 0.86
BV-1 0.16 0.19 0.27 0.21 0.21 0.77

Panel B: Other information dynamics


vi,t+1 = γ0t + γ 1t vit + γ 2t DIVit + γ 3t BVi,t-1 + eit
# annual
Partition γ1 γ2 γ3 Mean adj.R2 regressions
DIV > 0 0.45c 0.25 11e
(10.4)d
DIV > 0 0.39 0.13 0.28 11
(8.6) (9.3)
DIV > 0 0.38 0.08 0.004 0.29 11
(8.5) (3.6) (1.9)

DIV = 0 0.34 0.15 11


(7.1)
DIV = 0 0.33 0.012 0.16 11
(6.8) (4.4)
a
Variable definitions are per table 1. In addition, abnormal earnings ACORENIt = CORENIt – (r x BVt-1), where r =
0.12. Other information vt = (F[NI+1] – {r x BVt}) – (ω1 x ACORENIt). If DIV > 0 (< 0), ω1 is set to 0.61 (0.46),
per results from panel C. The term “v+1” is used to denoted vt+1, etc.
b
Pearson (Spearman) correlations are above (below) the diagonal. For correlations involving v, there are at most 12
year-by-year correlations (1985-1996). For correlations involving CORENI, there are at most 22 year-by-year
correlations (1974-1995).
c
Mean year-by-year coefficient estimate.
d
t-statistic based on the standard error of the year-by-year coefficient estimates.
e
# annual regressions: 11 = 1985− −1995, 12 = 1985− −1996, 22 = 1974− − 1995.

29
Table 3 (continued)

Panel C: Abnormal earnings dynamics

ACORENIi,t+1 = ω0t + ω1t ACORENIit + ω2t ACORENIi,t-1 + ω3t vit + ω4t DIVit + ω5t BVi,t-1 + εit

Mean # annual
ω1 ω2 ω3 ω4 ω5 adj.R2 regressions
DIV > 0 0.61 0.14 0.45 22e
(10.0) (3.0)
DIV > 0 0.60 0.08 0.15 0.49 22
(9.8) (1.9) (2.8)
DIV > 0 0.58 0.10 0.21 -0.006 0.50 22
(9.9) (2.4) (5.0) (-1.4)
DIV > 0 0.65 -0.01 0.70 0.51 12
(15.6) (-0.1) (14.7)
DIV > 0 0.64 0.02 0.73 -0.001 -0.006 0.53 12
(18.7) (0.4) (19.4) (-0.0) (-0.8)

DIV = 0 0.46 0.09 0.30 22e


(19.9) (4.2)
DIV = 0 0.61 0.06 0.59 0.38 12
(20.3) (2.3) (10.0)
DIV = 0 0.60 0.06 0.61 -0.03 0.39 12
(21.5) (2.7) (11.2) (-3.4)

30
Table 4

Mean coefficients and associated t-statistics from partitioned annual cross sectional OLS regressions of the market value of
common equity MVE on independent variables in Ohlson’s (1995) model when other information v is assumed to be zero.
Industry dummies are included in the regressions but not reported. Time period is 1974− − 1996.

MVEit = a0t + a1t BVit + a2t CORENIit + a3t DIVit + a4t NETCAPit + eit

Independent variablesa, and coefficient sign {magnitude} predictions


Partitions BV CORENI DIV NETCAP Mean # annual ∂MVE/∂DIV
0 ≤ a1 = 1 − k ≤ 1 a 2 = kϕ ≥ 0 − 1 ≤ [a 3 = a 4 = − k ] ≤ 0 adj.R2 regressions

DIV > 0 {0.86}b {1.34} {–0.14} {–0.14} {–1}b


0.69c 5.90 0.81 23
(8.7)d (15.9)
0.58 5.21 3.86 0.51 0.82 23 3.28
(7.8) (15.7) (6.2) (2.5) [6.8]e

DIV = 0 {0.92} {0.78} {–0.08}


1.24 1.98 0.55 23
(11.2) (8.8)
1.16 1.95 -3.01 0.59 23
(11.4) (8.8) (-11.8)
a
Variable definitions are per table 1.
b
Magnitude predictions for BV, CORENI, DIV and NETCAP come from substituting the coefficient estimates ω1 = 0.61 and ω1 = 0.46 for dividend and non-
dividend paying firms, respectively into equation (4.2) (see table 3, panel C). The dividend displacement prediction follows from ∂MVE/∂DIV = -a1 + a3.
c
Mean year-by-year coefficient estimate.
d
t-statistic versus a null of zero, based on the simple standard error of the year-by-year coefficient estimates.
e
t-statistic versus a null of –1, based on the simple standard error of the year-by-year coefficient estimates.

31
Table 5
Mean coefficients and associated t-statistics from partitioned annual cross sectional OLS regressions of the market value of
common equity MVE on independent variables in Ohlson’s (1995) model when other information v follows Ohlson’s modified
AR(1) dynamic. Industry dummies are included in the regressions but not reported. Time period is 1974− − 1996.

MVEit = c0t + c1t BVit + c2t CORENIit + c3t DIVit + c4t NETCAPit + c5t E[NI+1]it + eit

Independent variablesa, and coefficient sign {magnitude} predictions


Partitions and BV CORENI DIV NETCAP E[NI+1] Mean # annual ∂MVE/∂DIV
E[NI+1] proxy used c1 = β 1 ≥ 0 c 2 = β 2ζ ≤ 0 [ c 3 = c 4 = − β 2 ] ≥ 0 c 5 = β 3 r −1 > 0 adj.R2 regressions

Panel A: DIV > 0 {0.70}b {–0.90} {0.08} {0.08} {3.28} {–1}b


E[NI+1] = F[NI+1] 0.43c 1.94 1.88 0.39 6.58 0.86 12e 0.65
(7.3)d (9.6) (5.8) (1.4) (13.6) [5.6]f
E[NI+1] = CORENI+1 0.40 2.72 2.49 0.58 4.00 0.84 22 1.60
(6.1) (9.6) (4.8) (3.5) (10.2) [4.9]f

Panel B: DIV = 0 {0.78} {–0.34} {0.04} {2.18}


E[NI+1] = F[NI+1] 0.91 0.39 -2.16 6.45 0.64 12
(15.3) (2.5) (-10.7) (13.7)
E[NI+1] = CORENI+1 1.05 1.15 -3.01 1.82 0.61 22
(11.8) (5.8) (-11.0) (9.8)
a
Variable definitions are per table 1, with the exception of E[NI+1]. E[NI+1] denotes the expectation of next year’s earnings, and is proxied by either next
period’s actual CORENI, denoted CORENI+1, or forecasted next period core earnings from I/B/E/S, denoted F[NI+1].
b
Magnitude predictions for BV, CORENI, DIV , NETCAP and E[NI+1] come from substituting γ1 = 0.45 and ω1 = 0.61 for dividend paying firms, and γ1 = 0.34
and ω1 = 0.46 for non-dividend paying firms, into equations (6.2)–(6.5), per panels B and C of table 3. The dividend displacement prediction follows from
∂MVE/∂DIV = – [c1 – c3 + (c5 x r)].
c
Mean year-by-year coefficient estimate.
d
t-statistic versus a null of zero, based on the simple standard error of the year-by-year coefficient estimates.
e
# annual regressions: 12 = 1985–1996, 22 = 1974–1995.
f
t-statistic versus a null of –1, based on the simple standard error of the year-by-year coefficient estimates.

32
Table 6

Mean coefficients and associated t-statistics from partitioned annual cross sectional OLS regressions of the market value of
common equity MVE on independent variables in Ohlson’s (1995) model when other information v is assumed to be zero.
Industry dummies are included in the regressions but not reported. Time period is 1974− − 1996.

MVEit = a0t + a1t BVit + a2t CORENIit + a3t DIVit + a4t NETCAPit + eit

Independent variablesa, and coefficient sign {magnitude} predictions


BV CORENI DIV NETCAP Mean # annual ∂MVE/∂DIV
Partitions 0 ≤ a1 = 1 − k ≤ 1 a 2 = kϕ ≥ 0 − 1 ≤ [a 3 = a 4 = − k ] ≤ 0 adj.R2 regressions

Panel A: DIV > 0 {0.86}b {1.34} {–0.14} {–0.14} {–1}b


CORENI > 0 0.37c 7.42 2.39 0.46 0.83 23 2.02
(5.5)d (23.4) (4.0) (2.4) [5.0]
CORENI < 0 0.71 -0.06 6.82 -0.39 0.85 23 6.11
(7.7) (-0.3) (6.6) (-1.1) [6.9]e

Panel B: DIV = 0 {0.92}b {0.78} {–0.08}


CORENI > 0 0.62 7.92 -2.78 0.67 23
(9.5) (10.8) (-10.7)
CORENI < 0 0.74 -0.68 -2.07 0.45 23
(9.8) (-8.4) (-12.0)
a
Variable definitions are per table 1.
b
Magnitude predictions for BV, CORENI, DIV and NETCAP come from substituting the coefficient estimates ω1 = 0.61 and ω1 = 0.46 for dividend and non-
dividend paying firms, respectively into equation (4.2) (see table 3, panel C). The dividend displacement prediction follows from ∂MVE/∂DIV = -a1 + a3.
c
Mean year-by-year coefficient estimate.
d
t-statistic versus a null of zero, based on the simple standard error of the year-by-year coefficient estimates.
e
t-statistic versus a null of –1, based on the simple standard error of the year-by-year coefficient estimates.

33
Table 7
Mean coefficients and associated t-statistics from partitioned cross sectional OLS regressions of the market value of common
equity MVE on independent variables in Ohlson’s (1995) model when other information v is non-zero and follows Ohlson’s
modified AR(1) dynamic. Industry dummies are included but not reported. Time period is 1974− − 1996.

MVEit = c0t + c1t BVit + c2t CORENIit + c3t DIVit + c4t NETCAPit + c5t E[NI+1]it + eit

Panel A: Annual regressions (industry dummies included by not reported)


Independent variables, and coefficient sign {magnitude} predictions
BVa CORENI DIV NETCAP E[NI+1] Mean # annual ∂MVE/∂DIV
Partitions and c1 = β 1 ≥ 0 c 2 = β 2ζ ≤ 0 [ c 3 = c 4 = − β 2 ] ≥ 0 c 5 = β 3 r −1 > 0 adj.R2 regressions
proxies used {0.68}b {–1.16} {0.12} {0.12} {3.72} {–1}b

DIV > 0, CORENI > 0, 0.27c 4.23 1.50 0.56 3.96 0.85 22e 0.76
E[NI+1] = CORENI+1 (4.3)d (13.6) (2.9) (3.3) (10.4) [3.3]f
DIV > 0, CORENI < 0, 0.61 -0.39 5.54 -0.89 1.09 0.87 22 4.80
E[NI+1] = CORENI+1 (9.3) (-1.6) (5.1) (-2.2) (3.0) [5.3]f

Panel B: Pooled time-series cross-sectional regressions (year and industry dummies included but not reported)
DIV > 0, CORENI > 0, 0.34 3.36 1.97 -0.37 6.58 0.84 13,867 0.85
E[NI+1] = F[NI+1] (19.9) (21.6) (10.1) (-2.3) (47.8) [<0.001]g
DIV > 0, CORENI < 0, 0.87 -0.02 4.69 -0.01 3.01 0.81 1,066 3.46
E[NI+1] = F[NI+1] (27.8) (-0.1) (6.3) (-0.0) (9.5) [<0.001]g

a
Variable definitions are per table 1, with the exception of E[NI+1]. E[NI+1] denotes the expectation of next year’s earnings, and is proxied by either next
period’s actual CORENI, denoted CORENI+1, or forecasted next period core earnings from I/B/E/S, denoted F[NI+1].
b
Magnitude predictions for BV, CORENI, DIV , NETCAP and E[NI+1] come from substituting γ1 = 0.45 and ω1 = 0.61 for dividend paying firms, and γ1 = 0.34
and ω1 = 0.46 for non-dividend paying firms, into equations (6.2)–(6.5), per panels B and C of table 3. The dividend displacement prediction follows from
∂MVE/∂DIV = – [c1 – c3 + (c5 x r)].
c
Mean year-by-year coefficient estimate.
d
t-statistic versus a null of zero, based on the simple standard error of the year-by-year coefficient estimates.
e
# annual regressions: 22 = 1985− −1996. Pooled time-series cross-sections: n = 13,867 and n = 1,066 are both from 1985−
− 1996.
f
t-statistic versus a null of –1, based on the simple standard error of the year-by-year coefficient estimates.
g
p-value on F-statistic testing the null that ∂MVE/∂DIV = –1.
34
Table 8

Mean coefficients and associated t-statistics from pooled time-series cross-sectional OLS
regressions testing Ohlson’s (1995) modified AR(1) linear information dynamics.
Regressions are restricted to dividend paying firms, and partitioned separately for firms
with positive and non-positive core earnings. Industry and year dummies are included but
not reported. Time period is 1985-1996 in panel A, and 1974-1996 in panel B.

Panel A: Other information dynamicsa


vi,t+1 = γ0t + γ 1t vit + γ 2t DIVit + γ 3t BV-1it + eit
Partitions γ1 γ2 γ3 Adj.R2 # obs.

DIV > 0, CORENI > 0 0.55b 0.10 0.33 9,958


(53.5)c (17.6)
DIV > 0, CORENI > 0 0.55 0.05 0.004 0.34 9,943
(52.3) (5.4) (4.9)

DIV > 0, CORENI < 0 0.13 0.45 0.18 663d


(3.9) (5.8)
DIV > 0, CORENI < 0 0.11 0.28 0.011 0.18 663
(3.2) (2.7) (2.6)

Panel B: Abnormal earnings dynamics


ACORENIi,t+1 = ω0t + ω1t ACORENIit + ω2t ACORENIi,t-1 + ω3t vit + ω4t DIVit + ω5t BV-1it + εit
Partitions ω1 ω2 ω3 ω4 ω5 Adj.R2 # obs.

DIV > 0, CORENI > 0 0.69 0.08 0.22 -0.01 0.42 34,660
(99.9) (12.4) (22.3) (-19.2)
DIV > 0, CORENI > 0 0.71 -0.02 0.68 0.01 -0.01 0.50 12,048
(62.1) (-1.7) (38.7) (0.7) (-9.1)

DIV > 0, CORENI < 0 0.19 -0.06 0.77 -0.07 0.33 2,589
(7.0) (-2.1) (7.3) (-10.4)
DIV > 0, CORENI < 0 0.60 -0.16 0.74 -0.25 -0.01 0.35 955
(9.0) (-4.1) (9.8) (-1.4) (-0.7)

a
Variable definitions are per table 1. In addition, abnormal earnings ACORENIt = CORENIt – (r x BVt-1), where r =
0.12. Other information vt = (F[NI+1] – {r x BVt}) – (ω1 x ACORENIt). If DIV > 0 (< 0), ω1 is set to 0.61 (0.46),
per results from panel C. The term “v+1” is used to denoted vt+1, etc.
b
From pooled time-series cross-sectional regression, where industry and year dummies are included but not reported.
c
t-statistic versus a null of zero.
d
Regressions where DIV > 0 and CORENI < 0 exclude 4 significant outliers.

35
Table 9
Robustness tests: Mean coefficients and associated t-statistics from cross sectional OLS regressions of the market value of
common equity MVE on independent variables in Ohlson’s (1998) model for dividend-paying firm-year observations only.
Maximum time period is 1974− − 1996.

Panel A: All variables undeflated in $ millions


Independent variablesa
Mean # annual
BV CORENI DIV NETCAP CORENI+1 CORENI+2 CORENI+3 CORENI-1 ONE DIRT adj.R2 regressions
0.31b 2.70 2.11 1.73 2.31 0.80 1.38 0.85 20d
(4.5)c (7.3) (2.9) (3.2) (6.6) (2.9) (6.5)
0.32 4.41 1.89 0.20 3.21 0.83 23
(5.4) (11.5) (3.6) (1.1) (7.5)
0.54 5.68 3.77 -0.05 -1.18 2.00 0.83 23
(7.4) (15.9) (5.5) (-0.2) (-0.7) (4.6)

Panel B: All variables deflated by number of common shares at fiscal year-end


Proxy for Mean # annual
E[NI+1] BV CORENI DIV NETCAP E[NI+1] ∂MVE/∂DIV adj.R2 regns.
not applicable 0.37 2.92 5.80 -0.39 5.43 0.58 23
(6.9) (26.7) (16.2) (-2.9) [17.3]e
CORENI+1 0.29 1.78 5.25 -0.36 1.97 4.73 0.61 22
(6.0) (21.0) (12.7) (-2.7) (10.3) [13.1]
F[NI+1] 0.28 1.04 3.55 0.14 5.46 2.62 0.67 12
(11.4) (9.9) (12.5) (1.6) (18.0) [12.8]

a
Variable definitions are per table 1.
b
Mean year-by-year coefficient estimate.
c
t-statistic versus a null of zero, based on the simple standard error of the year-by-year coefficient estimates.
d
# annual regressions: 12 = 1985− −1996; 20 = 1974− −1993, 23 = 1974− − 1996.
e
t-statistic versus a null of –1, based on the simple standard error of the year-by-year coefficient estimates
36

You might also like