2. Description of the Article
■ Title: Dividends and Stock Prices
■ Authors: Irwin Friend, Professor of Economics and Finance, University of
Pennsylvania,
Marshall Puckett, Economist, Federal Reserve Bank of New York
■ Journal: The American Economic Review, Vol. 54, No. 5 (Sep., 1964), pp. 656-682
3. Background
■ Empirical evidences indicate market preference for strong dividend effect in
relation to retained earnings creating considerable controversy and confusion over
the relative importance of dividends and retained earnings in determining the
price-earnings ratios of common stocks.
4. Research Gap/Motivation
■ The empirical evidences show the existence of strong dividend effect as compared
to retained earnings in relation to valuing price-earnings ratios.
■ Such empirical findings are not relatable to theoretical preconceptions, creating a
gap between empirical evidences and theory.
5. Purpose of the study
■ To check the validity of previous statistical studies purporting to confirm the
existence of strong dividend effect on stock prices by way of analyzing the
limitations of those findings,
■ To describe various approaches to avoiding these limitations, and
■ To present new results that seem more in accord with theoretical preconceptions.
6. Data
■ Five industry samples, viz., chemicals, electronics, electric utilities, foods, and
steels, in each of two years, 1956 and 1958 were included in the study.
■ The sample included growth and non-growth, cyclical and non-cyclical industries.
■ The years were selected to include substantial stock price rise (year 1956) and
somewhat-depressed economy (year 1958).
Methodology and Econometric Issues
7. Models and variables used
■ The following equation mainly applied to cross-section data is the base model to
describe the price per share (P), Dividends (D) and Retained earnings (R) for
company “i” in an industry measured in the time period “t”.
Pit = a + b Dit + c Rit + eit
■ Theory would suggest that regardless of the optimum payout for any individual
company, at that optimum $1 of dividends would on the average have the same
effect on stock price as $1 of retained earnings. Any difference between the values
of b and c therefore represents either a disequilibrium payout position or a
statistical limitation of the analysis employed, including most notably a correlation
of dividends or retained earnings with omitted factors affecting price.
Methodology and Econometric Issues
Hypothesis
8. ■ However, the following possible biases may be present:
– Omitted variables (Risk variable and per share earnings growth)
– Regression weights (Extreme values more important in regression outcomes
than those clustered around average.)
– Random variations in income (short-run disturbances in reported earnings
may bias regression outcomes in favor of dividend payout influences)
– Income measurement errors (diversity of according procedures employed can
bias regression results in favor of dividends)
– Least-Squares bias (The standard regression equation will yield results biased
in favor of dividend payout because it assumes one-way causality between
dividends and prices.)
Methodology and Econometric Issues
9. ■ Hence, the given model should be modified to limit the effect of such omissions. In
furtherance, the separate effects on price of all omitted variables were aggregated
and the variable Fi was introduced, modifying the model as:
Pit = at + btDit + ctRit + Fi + eit
■ Identical equation for time t-1 was subtracted, to obtain the following model:
Pit - Pi(t-l) = at - at-1 + btDit - bt-1Di(t-l) + ctRit – Ct-1Ri(t-l) + et -ei(t-1)
■ Doing so, the Fi was freed from the equation, so the coefficients on dividends and
retained earnings were made free of the bias due to firm effects.
■ Multiplicative relationship for the firm effects was assumed and Fi was calculated
as,
Fit = fiEit
■ Where, fi = the firm-effect "multiplier" and E = per-share earnings.
Methodology and Econometric Issues
10. ■ Firm effects were held constant by introducing a variable [(P/E)'i(t-1) which
measures individual deviations from the sample average price earnings ratio in the
previous time periods,
Pit = a + bDit + cRit + d(P/E)'(t-l) + eit
■ The problem of least-squares bias was handled by specifying a complete model
including a dividend supply function as well as the customary price relation:
Pit = a + bDit + cRit + d(P/E)'i(t_l)
■ The problem of random income movements was solved by the below equation
where the coefficients b and c measure the extent of short run price adjustment to
short run changes in dividends and earnings retentions respectively:
Pit = a + bDit + cRit + dPi(t-1)
Methodology and Econometric Issues
11. ■ Earnings normalization was done by employing market estimates. The dividend-
price ratio is assumed to be always normal but the earnings-price ratio is subject to
short-run fluctuations. Normalized value of earnings-price ratio was obtained as,
(E/P)n
it = [ai + bit] (E/P)kt
■ Finally, the influence of dividend payout on price was subjected to time-series
analysis.
Methodology and Econometric Issues
12. ■ The usual simple linear
relationships between average
prices and dividends and retained
earnings showed customary strong
dividend and relatively weak
retained earnings effect in three of
the five industries namely,
chemicals, foods, and steels (Table
1).
■ In growth industries (chemicals,
electronics and electric utilities)
more weight relatively was given to
retained earnings than in nongrowth
industries (steels and foods), but the
evidence was not uniform
(chemicals) and for one of the two
remaining industries (electric
utilities) the results depended partly
on the mathematical form of the
regression used (primary form or
logarithmic transformation).
Major Findings
Regression results
13. ■ Adding a lagged earnings-price
ratio to the equations in Table 1
to hold firm effects constant, it
was found that dividends have a
predominant influence on stock
prices in the same 3 out of 5
industries but the differences
between independent variables
were not quite marked as in first
regression (Table 2).
Regression results
14. ■ Inclusion of identity (Eit) in the
equation showed that price
effects on dividend supply
were probably not a serious
source of bias in the
customary derivation of
dividend and retained
earnings effects on stock
prices (Table 3).
Regression results
No major changes from previous
table
Some notable changes from previous table
Eliminating Electronics from study
not explained (limitations)
15. ■ Lagged price variable was added
to standard liner equation in
Table 1 to provide some direct
evidence on potential bias arising
from short run income
movements, which showed that
retained earnings receive greater
relative weight than dividends in
the majority of cases (Table 4).
Regression results
In growth industries (chemicals,
electronics, and electric utilities) the
retained earnings effect was larger
than the dividend effect for both years
whereas for the other two industries
(steels and foods) no any significant
systematic differences between the
retained earnings and dividend
coefficients were noted.
16. ■ Normalization procedure was
performed for chemicals,
foods, and steels companies
in 1956 and 1958 (Table 5)
and the prior year's
normalized earnings-price
variable was also added to
hold firm effects constant
(Table 6) which showed the
significant role of normalized
price-earnings ratio in
eliminating part of the usual
understatement of the
relative importance of
retained earnings.
Regression results
17. ■ Table 6 showed that major
differences between dividend
and retained earnings
coefficients disappear when
earnings are normalized and
firm effects held constant.
Regression results
Compare with table 5 to
understand the rise in value of
coefficient c due to introduction
of lagged E/P ratio in regression
(Normalization of earnings)
18. ■ A more detailed
examination of chemicals
sample disclosed that the
results obtained largely
reflected the undue
regression weighting given
the three firms with prices
deviating most from the
average price in the sample
of 20 firms. Omitting those
three firms, the results
showed that retained
earnings now became
somewhat more important
than dividends as a price
determinant (Table 7).
Regression results
Coefficient of retained earnings
increased exceeding values of
coefficient of dividends
19. ■ Lastly, to eliminate the
potential bias in favor of
retained earnings due to
normalization procedure, the
time-series behavior of the
relative earnings yield ratios
were analyzed for chemicals
sample (Table 8) which
showed that in 12 out of 20
cases, the time-slope
coefficients for the relative
earnings yield and relative
payout regressions had the
same sign, while in 8 cases
they were of opposite sign.
■ The correlation between the
two slope coefficients were
not very high but was
significant, strongly
evidencing that the
customary results were
invalid.
Regression results
20. Limitations
■ Linear regression estimate was used in the study. An adequate study of the size of
optimal payout ratios under varying circumstances may require a more complex
statistical analysis which has been left out in the study.
21. Conclusion and Implications
■ It was concluded that there was no significant basis for customary view explaining
a dollar of dividends has several times impact on price of a dollar of retained
earnings, except in case of nongrowth stocks.
■ It is possible that management might be able to increase stock prices in
nongrowth industries by raising dividends, and in growth industries by greater
retention.
22. Reflection from the article
■ Analyzing the regression equation in depth with understanding of the variables’
nature of inclusion/non-inclusion of various factors is very critically explained in
the study giving insights for beginners like us to consider such instances in our
study.