Louisiana State University LSU Digital Commons
LSU Historical Dissertations and Theses Graduate School
1992 Three Essays in Dividend Policy. Jaisik Gong Louisiana State University and Agricultural & Mechanical College
Follow this and additional works at: https://digitalcommons.lsu.edu/gradschool_disstheses
Recommended Citation Gong, Jaisik, "Three Essays in Dividend Policy." (1992). LSU Historical Dissertations and Theses. 5382. https://digitalcommons.lsu.edu/gradschool_disstheses/5382
This Dissertation is brought to you for free and open access by the Graduate School at LSU Digital Commons. It has been accepted for inclusion in LSU Historical Dissertations and Theses by an authorized administrator of LSU Digital Commons. For more information, please contact [email protected]. INFORMATION TO USERS
This manuscript has been reproduced from the microfilm master. U M I films the text directly from the original or copy submitted. Thus, some thesis and dissertation copies are in typewriter face, while others may be from any type of computer printer.
The quality of this reproduction is dependent upon the quaiity of the copy submitted. Broken or indistinct print, colored or poor quality illustrations and photographs, print bleedthrough, substandard margins, and improper alignment can adversely affect reproduction.
In the unlikely event that the author did not send UM I a complete manuscript and there are missing pages, these will be noted. Also, if unauthorized copyright material had to be removed, a note will indicate the deletion.
Oversize materials (e.g., maps, drawings, charts) are reproduced by sectioning the original, beginning at the upper left-hand comer and continuing from left to right in equal sections with small overlaps. Each original is also photographed in one exposure and is included in reduced form at the back of the book.
Photographs included in the original manuscript have been reproduced xerographically in this copy. Higher quality 6" x 9" black and white photographic prints are available for any photographs or illustrations appearing in this copy for an additional charge. Contact U M I directly to order.
UMI University Microfilms International A Beil & Howell Information Company 300 North Zeeb Road. Ann Arbor. Ml 48106-1346 USA 313.'761-4700 800.'521-0600
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Order Number 8S02900
Three essays in dividend policy
Gong, Jaisik, Ph.D.
The Louisiana State University and Agricultural and Mechanical Col., 1992
Copyright ©1998 by Gong, Jaisik. All rights reserved.
UMI 300 N. Zeeb Rd. Ann Arbor, MI 48106
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. THREE ESSAYS IN DIVIDEND POLICY
A Dissertation
Submitted to the Graduate Faculty of Louisiana State University and Agricultural and Mechanical College in partial fulfillment of the requirements for the degree of Doctor of Philosophy
in
The Interdepartmental Programs in Business Administration
by Jaisik GONG B.A., Seoul National University, 1981 M.B.A., Seoul National University, 1984 August 1992
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. ACKNOWLEDGMENTS
I would like to thank Professor George M. Frankfurter, the chairman of my
dissertation committee, for his guidance and encouragement. I am equally grateful to the
other members of the committee: Professors G. Geoffrey Booth, William R. Lane, Joh ' S.
Howe, and Robert E. Martin.
I am also indebted to many other people for their assistance. I appreciate the help
of Professors Carter Hill, Douglas McMillin, and Tae-Hwy Lee. I also wish to thank Joan
Payne, Shirley Young, Bessie Avera, Yvonne Day, Quang Do, Minbo Kim, and my colleagues
in the Department of Finance.
My most heartfelt thanks go to my family. The support and sacrifice of my mother
was crucial to my completion of this dissertation. The love and encouragement of my wife,
Gilwon, kept me on the right track. My two sons, David and Richard, deserve my deepest
appreciation for their tolerance and understanding.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. TABLE OF CONTENTS
Page
Acknowledgments List of Tables V List of Figures vii Abstract vlli
CHAPTER 1: INTRODUCTION
CHAPTER 2: REVIEW OF LITERATURE
A. Dividend Policy 5 B. Dividends and Taxes 12 C. Dividends and Stock Price 17 D. Dividend Signalling 18
CHAPTER 3: THE DETERMINANTS OF DIVIDEND POLICY 23
A. Introduction 23 B. Data 25 1. Sample Selection 25 2. Variable Definitions 26 C. Methodology 28 1. Time-Series Cross-Sectional Analysis 28 2. Vector Autoregressive Model 29 D. Empirical Results 34 1. Sample Characteristics 34 2. Time-Series Cross-Sectional Regression Results 37 3. Vector Autoregression Results 44 4. The Lagged Dividend Model 57 5. Interpretation of Results 64 E. Chapter Summary and Conclusions 66
CHAPTER 4: DIVIDENDS. TAXES, and PORTFOLIO CHOICES 68
A. Introduction 68 B. The Tax Reform Acts of 1984 and 1986 70 1. The Tax Reform Act of 1984 70 2. The Tax Reform Act of 1986 73 C. Models and Testable Hypotheses 74 1. The Tax-Effect Model 74 2. Short-Term Trading Model 77 3. The Portfolio Model 79
III
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. D. Data and Methodology 82 1. Data 82 2. Methodology 84 E. Empirical Results 85 1. Descriptive Results of Consistency 85 2. Test of Short-Term Trading Hypothesis 89 3. Test of Tax-Effect Hypothesis 96 4. Portfolio Test Results 99 F. Chapter Summary and Conclusions 110
CHAPTER 5: EMPIRICAL TESTS OF DIVIDEND SIGNALLING 113
A. Introduction 113 B. The John and Williams Model 115 1. Narrative 115 2. Testable Analytical Models 116 C. Methodology 118 D. Data and Proxy Variables 120 1. Liquidity 123 2. Firm Size 124 3. Dividend Policy 124 4. Accounting Variables 125 E. Testing 127 1. Sample Statistics 127 2. Hypothesis #1 : Demand for Liquidity Causes Dividend Payments. 129 3. Hypothesis #2: The Equilibrium Price Is Monotonie in the Dividend Signal (or Private Information). 139 4. Hypothesis 43: Announcement Effects (Excess Returns during the Event Period) Are an Endogenous Function of Dividend Levels, the Demand for Liquidity, Market Value, and Firm Size. 152 F. Chapter Summary and Conclusions 160
CHAPTER 6: SUMMARY, CONCLUSIONS, AND FUTURE RESEARCH 163
BIBLIOGRAPHY 168
APPENDIX A: THE RELATIONSHIP BETWEEN DIV AND FCASH 178
APPENDIX B: THE EFFECTS OF 1986 TRA 180
APPENDIX C: SIGNALUNG MODEL 182
VITA 188
IV
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. LIST OF TABLES
TabI# Page
3.1 Sample Characteristics 35
3.2 Time-Series Cross-Sectional Estimates from Regressing Fuller and Battese Model for 446 Firms in the Sample from 1979:11 to 1990:IV 38
3.3 Time-Series Cross-Sectional Regression Analysis Using Alternative Model Specifications 41
3.4 Tests of Stationarity 45
3.5 Decompositions of Forecast Error Variance 48
3.6 Sensitivity Analysis for Dividend Variance Decompositions Using Randomly Selected Samples 53
3.7 Time-Series Cross-Sectional Regressions with Lagged Dependent Variables 59
3.8 Robustness Tests of Lagged Dividend Model 62
4.1 Summary of Changes in the Tax Reform Acts of 1984 and 1986 71
4.2 Number of Sample Firms and Dividend Payment Events from January 1980 throughDecember 1989 83
4.3 Sample Descriptive Statistics on Ex-Dividend Day Excess Returns and Dividend Yields 86
4.4 Non-Parametric Tests of Equality of Ex-Dividend Day Excess Returns Between the two periods 90
4.5 Average Excess Returns Around the Ex-Dividend Day for the Period January 1, 1980, to July 17, 1984 93
4.6 Average Excess Returns Around the Ex-Dividend Day for the Period July 18, 1984, to December 31, 1986 94
4.7 Average Excess Returns Around the Ex-Dividend Day for the Period January 1, 1987, to December 31,1989 95
4.8 Cross-Sectional Regression of Ex-Dividend Day Excess Returns on Dividend Yields 98
4.9 Comparison of Cum-Dividend and Ex-Dividend Efficient Portfolios Between the Pre-1984 Tax Reform Period and the Post-1984 Tax Reform Period 100
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table Page
4.10 Comparison of Cum-Dividend Efficient Portfolios Between the Pre-1986 Tax Reform Period and the Post-1986 Tax Reform Period 107
5.1 Number of Quarterly Dividend Payments in the Sample, by Group 122
5.2 COMPUSTAT Variable Names and Definitions 126
5.3 Quarterly Summary Statistics of the Final Sample, 1986 1/4 - 1990 4/4 128
5.4 Correlation Matrix of Proxy Variables 130
5.5 Average Volume Prediction Error for Quarterly Dividend Announcements 1986/1 - 1990/4 132
5.6 Regressions of Dividend Yield on Liquidity Variables 138
5.7 Average Return Prediction Error for Quarterly Dividend Announcements 1986/1 - 1990/4 140
5.8 Tests of Firm Size Effects on the Relationship Between Signal and Price 145
5.9 Tests of the Polynomial Pricing Functions 147
5.10 Likelihood Ratio Tests of the JW Model: Constrained and Unconstrained Estimation of Equation (5.21) 151
5.11 Tests of Cross-Sectional Relations by Dividend for Three Windows, Equation (5.23) 154
5.12 Tests of Liquidity Related Variables by Firm Size, Equation (5.23) 156
5.13 Other Liquidity Tests 159
VI
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. LIST OF FIGURES
Figure Page
3.1 impulse Response Functions of Dividends 55
4.1 Comparison of Efficient Frontiers (Before 1984 TRA) 104
4.2 Comparison of Efficient Frontiers (After 1984 TRA) 105
4.3 Comparison of Efficient Frontiers (Before 1986 TRA Versus After 1986 TRA) 109
5.1 Information-Motivated Trading Volume Volume APEs and CAPEs 134
5.2 Liquidity Trading Around the Dividend Announcement Day 136
5.3 The Relationship Between Signal and Price Return APEs and CAPEs 143
VII
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. ABSTRACT
This dissertation focuses on three leading theoretical and empirical issues in the
anomaly of dividend policy. The first essay analyzes the explanatory power of the various
theories of dividend determinants, many of which have not been tested yet. In this essay,
time-series cross sectional tests are undertaken using individual firm data, while the
structural vector autoregressive (VAR) methodology is applied to the aggregate data.
The second essay proposes an alternative approach to the ex-dividend anomalies
that fully incorporates both the tax-effect hypothesis and the short-term trading hypothesis.
This approach draws on normative portfolio selection models to examine the relationships
between the ex-dividend anomalies and the ex-post portfolio choice in the context of the
1984 and 1986 federal tax reforms.
The third essay tests signalling equilibrium models empirically. In this essay, tests
are conducted to provide statistical evidence of the empirical validity of dividend signalling
models. Empirical validation of the notion of a signalling equilibrium is examined for a
sample of dividend-paying firms selected from the NYSE and AMEX.
VIII
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Chapter 1. INTRODUCTION
The objective of thii dissertation is to extend empirical work on dividend policy
in three ways. The first essay investigates the determinants of dividend policy. The
extant dividend theories suggest that dividends should be distributed depending on the
firm's attributes such as earnings, investments, free cash flow, tax, firm size, industry
classification, etc. Empirical work on these dividend determinants has lagged behind
theoretical research, and the econometric techniques used thus far leave something to
be desired.
The first essay examines a wide range of theoretical determinants of dividend
policy, including free cash flows. In this essay, time-series cross-sectional tests are
undertaken using individual firm data, while the vector autoregressive (VAR)
methodology is applied to the aggregate data. These methodologies are desirable
because previous studies, which employed cross-sectional analysis of either firm-specific
or aggregate data, failed to capture important information explaining differences in
dividend policy over time. This essay, furthermore, analyzes the explanatory power of
the various theories of dividend policy.
The second assay examines the relationships between ex-dividend anomalies and
ex post portfolio choice in the context of the 1984 and 1986 federal tax reforms. Ex-
dividend anomalies are addressed through competing hypotheses of tax effects and
short-term trading. The two tax reforms are expected to have a significant effect on
dividend policy. The 1984 Tax Reform Act extended the minimum holding period for the
dividend tax deduction, exposing tax-induced short-term traders to more risks. The
1986 Tax Reform Act equalized tax rates on dividends and capital gains, reducing the
dissipative costs of dividends. Transition periods between the two tax reforms provide a
valuable opportunity to analyze the validity of each hypothesis.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 2
The second essay draws on normative portfolio selection models and
incorporates both tax effects and short-term trading around the ex-dividend day. The
underlying conjecture of this essay is that any change in the risk-expected return trade
off induced by the two tax reforms should be manifested in the portfolio choices. This
portfolio approach seems more appropriate since it is an equilibrium model with proper
measures of risk and return, rather than a model of behavior.
The third essay tests signalling equilibrium models empirically. Signalling
equilibrium models imply that firms distribute dividends in order to convey favorable
insider information and thus achieve a higher stock price. For example, corporate
insiders are proposed to have incentives to signal more valuable future cash flows by
paying larger dividends if the firm's current shareholders require more liquidity than
internally generated funds.' This signal would result in a bidding up of stock price, thus
benefiting the firm or current shareholders selling stocks. The third essay tests the
explicit relations among the announcement effect, the dividends, and the cum-dividend
market values, suggested by signalling models.
Dividend issues are addressed in a vast body of theoretical and empirical
literature. The literature can be categorized into several broad groups. The first group
has sought to identify the determining factors of dividend policy. The seminal works of
Lintner (19561 and Miller and Modigliani (1961) have motivated many people (see, for
example, Dhrymes and Kurz (1964, 1967), Fama and Babiak (1968), Higgins (1972),
Fama (1974), McCabe (1979), Smirlock and Marshall (1983), and Partington (1985)1 to
investigate empirically the determinants of dividend policy with mixed results. Rozeff
(1982), Easterbrook (1984), and Jensen (1986) draw on agency theory to explain
dividend payments, whereas Myers (1984) and Myers and Majluf (1984) view dividend
policy in terms of pecking order theory. In addition, many authors (see, for example,
Michel (1979), Feldstein and Green (1983), Michel and Shaked (1986), and Dyl and
^This argument ia originally attributable to John and Williams (1985).
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 3
Hoffmeister (1986)] have extended theoretical and empirical works on other dividend
determinants such as industry classifications, firm size, and beta coefficients.
The second group focuses on the ex-dividend anomaly. In order to explain
excess ex-dividend returns, some authors [see, for example, Elton and Gruber (1970),
Booth and Johnston (1984), Elton, Gruber and Rentzler (1984), Barclay (1987), and
Michaely (1991)] follow the tax-effect hypothesis, which is challenged by Eades, Hess
and Kim (1984), and Grinblatt, Masulis and Trtman (1984). Others [see, for example,
Miller and Scholes (1978, 1982), Kalay (1982, 1984), Lakonishok and Vermaelen
(1983, 1986), Karpoff and Walkling (1988, 1990), Grammatikos (1989), and Fedenia
and Grammatikos (1991)] relate this ex-dividend anomaly to the short-term trading
hypothesis, by allowing for transaction costs around the ex-dividend day.
The third strand of literature began with studies dealing with dividend
announcement effects (see, for example, Pettit (1972), Watts (1973), Aharony and
Swary (1980), Asquith and Mullins (1983), Penman (1983), Patell and Wolfson (1984),
and Kalay and Loewenstein (1985)]. These studies, which document the informational
contents of dividends, have evolved into recent dividend signalling models [see, for
example, Bhattacharya (1979), Miller and Rock (1985), John and Williams (1985),
Ambarish, John and Williams (1987), Ofer and Thakor (1987), Williams (1988), and
Kumar (1988)]. Even though empirical testing of signalling is in its infancy, these
dividend-signalling models have emerged as one of the most appealing theories that
seemingly explain the enigma of dividend policy.
The remaining body of this dissertation is organized as follows. Chapter 2
presents a more detailed discussion and review of the extensive literature relating to the
three essays comprised in this dissertation. Chapter 3 encompasses various
econometric analyses of dividend determinants using time-series cross-sectional tests
and the vector autoregressive (VAR) methodology. In Chapter 4, a portfolio approach is
proposed that can explain the ex-dividend anomalies. In Chapter 5, some empirical tests
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 4
of dividend signalling are attempted. A summary of this dissertation and some
suggestions for future research are presented in Chapter 6.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Chapter 2. REVIEW OF LITERATURE
The vast body of literature on dividends includes numerous theoretical and
empirical papers. This chapter reviews several bodies of literature that are directly
related to the dissertation. The first body of literature, focusing on dividend policy,
investigates the determining factors of dividends. The second body encompasses
studies associated with two hypotheses tnat attempt to explain the ex-dividend
anomalies: the tax-effect and short-term trading hypotheses. The other two bodies of
literature cover studies dealing with dividend announcement effects and discuss many
recent theoretical studies on dividend signalling.
A. Dividend Policy
Dividend policy has long been an issue of interest in financial literature. The
seminal paper on dividend policy is that of Lintner (19561, which was based on field
interviews with managers at 28 companies. The results of these interviews indicate that
a company's dividend decisions depend on current earnings and previous dividends,
where the relationship between the two determinants constitutes a target pay-out ratio.
Lintner (1956) encapsulizes these observations in a theoretical model, as follows:
DIV/ . 0,EARN h (2.1)
DIV, - DIV,,., « a, + d, ( D IV / ■ DIV,.,., ) + 12.2)
where DIV,* = target dividends in the current year, 0, = target pay-out ratio, EARN, > current earnings, and DIV,, DIV,,., = dividends paid in the current and previous years. (Lintner (1956), p. 107]
Equation (2.2) can be rearranged as
DIV, . o, + A EARN, + K, DIV,.,., + f,. (2.3)
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 6
Lintner fitted equation (2.3) to the aggregate data that he secured from the national
income accounts. The dividend prediction equations that he obtained are
DIV - 352.3 + 0.15 EARN^ + 0.70 DIV^„ (2.4)
when earnings are adjusted for inventory gains, or
DIV . 106.0 + 0.145 EARN„„^ + 0.788 DIV,.,, (2.5)
when earnings are not adjusted.' {ibid., p. 109]
This Lintner model has drawn two criticisms. First, as Dhrymes and Kurz (1964)
point out, the Lintner model gives a satisfactory prediction of short run dividend policy,
but it does not explain intertemporal variations in dividend policy. Second, Miller (1986)
argues that the Lintner model remains a behavioral model because no authorhas been
able to solve it as a maximization problem.
The "two variable" Lintner model is supported by Fama and Babiak (1968). They
examine the predictive powers of various dividend models including the Lintner model by
using data for individual firms. Fama and Babiak conclude that the Lintner model
performs well, but the best is the model deleting the constant term and adding the
lagged term for earnings.
Miller and Modigliani (1961) present a strong challenge to the conventional view
of dividend preference. They are concerned with the effect of a firm's dividend policy
on stock price in an ideal economy where perfect capital markets, rational behavior, and
perfect certainty are assumed. Miller and Modigliani develop the irrelevance proposition
on dividend policy in the sense that the value of a firm is not affected by dividend policy.
Miller and Modigliani derive a firm's value as the following expression:
VAL, * [EARN,,, • INV„, + VAL,., ) / (I + p,„ ) (2.6)
where VAL,,VAL,,, « the values of a firm at t and t+ 1 , EARN,,, » the firm's earnings, INV„, = the firm's investment, and p „ , - the rate of return. (Miller and Modigliani (1961), p. 414]
^The units dsnotad are billions of dollars.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 7
Miller and Modigliani show that •• since the firm's earnings, investment, and the value at
t+ 1 are all independent of dividend in equation (2.6) ~ the dividend decisions are
irrelevant for the value of the firm. They also argue that dividend decisions are still
irrelevant for a growing firm with debt and taxes or under uncertainty. This "dividend
irrelevance” proposition is possible because investors can create "homemade dividends"
by selling off portions of the stock of the non dividend paying firm or reinvesting the
dividends paid by the dividend paying firm.
Another major thrust of their paper on dividend policy is an implication of the
Fisherian model of the firm, in which the value of a firm is independent of its method of
financing. Miller and Modigliani propose that dividend decisions are independent of
investments because the higher dividends would induce the more external financing
through the issue of new shares to keep the investment level unchanged. They argue
that the firm undertakes the investment opportunities that maximize its current value,
which is independent of dividend decisions. These imply that there are no "financial
illusions" in a rational and perfect environment. In a world of well-functioning capital
markets, a firm's dividend policy is essentially an exercise in financial packaging. The
value of the shares is determined solely by the underlying real, economic factors, and
not by "mere financial packaging."
This separation proposition on dividends and investments has motivated many
researchers to undertake empirical tests. Higgins (1972) designed a dividend model for
a firm maximizing stockholders' wealth and conducted a two-stage least squares
estimation as follows:
INVRATIO - 0.0280 + 0.7194 SALRATIO + 0.0437 DIVRATIO (2.77) (11.68) (0.19) and DIVRATIO . 0.0033 + 0.5903 EARRATIO - 0.0625 INVRATIO, (2.16) (30.20) (-5.25) (2.7) where INVRATIO " investments/assets, DIVRATIO * dividends/assets, SALRATIO - sales/assets. EARRATIO ” eamings/assets, and t-values are in parentheses. (Higgins (1972), p. 15391
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 8
Higgins concludes that dividends are considered a residual in the corporate decision
nexus and that investments are not significantly influenced by dividends.
In a more rigorous study, Fama (1974) applies simultaneous equations models to
the data for individual firms. His findings are that dividend decisions are not determined
by investment, and, furthermore, that investment decisions are not affected by
dividends. These results are very consistent with the separation proposition of Miller
and Modigliani (1961) on dividends and investments, even though the proposition
assumes a perfect capital market.
Smirlock and Marshall (1983) provide an extension of Fame's work on the
separation proposition of Miller and Modigliani (1961) by using the Granger-causality
tests. Their use of causality tests has some appeal because "they were developed to
test for statistical exogeneity without specification of structural models" [Smirlock and
Marshall (1983), p. 1660). Smirlock and Marshall do not find any Granger-causality
between dividends and investments. These results, consistent with Fame's (1974), fully
support the separation proposition on dividends and investments.
Another test in support of the separation proposition is offered by Partington
(1985). He conducts a questionnaire survey with 93 large Australian companies to
analyze the relationship among dividend, investment, and financing decisions. His
results indicate non-residual determination of dividends. He concludes that firms would
usually make dividend and investment decisions independently, and that external
financing is residually determined.
On the other hand, Dhrymes and Kurz (1964, 1967) go against the separation
proposition. The motivation for their study (1964) stems from arguments that Untner's
(1956) hypothesis is vulnerable to the explanation of long-run dividend policy. In order
to account for the observed variety of dividend payment practices in firms, Dhrymes and
Kurz (1964) examine several explanatory variables such as size, investment,
indebtedness, liquidity position, control, and income variability in electric utility firms.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 9
They find that a firm's dividend policy is affected by investment, indebtedness, size, and
the regulatory status of the firm.
Dhrymes and Kurz (1967) provide an interesting argument and empirical
evidence of dividend policy. They note that, contrary to the assumption of Miller and
Modigliani (1961), capital markets are characterized by imperfections, and that external
financing, including new equity and bond issues, is a more expensive vehicle of
financing than internal funds. They argue that dividends and investments are
competitive outlays, both of which rely on limited internal funds. Their hypothesis
relates that dividends, investments, and external financing are interrelated. Consistent
with the hypothesis, their results indicate that a firm's dividend decisions are
significantly affected by its investment requirements, while investment decisions are
impeded by the rigid dividend policy, and that investments and dividends induce external
financing.
McCabe (1979) examines Dhrymes and Kurz's (1967) study and Fame's (1974)
study, and stresses the role of new debt as a determinant of dividend policy. McCabe
argues that Fama (1974) failed to consider all the relevant variables, which led to a
biased result supporting the Miller and Modigliani separation proposition. His hypothesis
follows arguments of Dhrymes and Kurz (1967) that funds raised from profits and
external financing are allocated between investments and dividends, but his contribution
is that new debts have significant effects on dividend decisions. He finds strong
interdependence among dividends, investments, and new debts, which is against the
separation proposition. This implies that dividend decisions are influenced by new debt
decisions as well as investment decisions.
Agency theory explanations of dividend payments build on Ross (1973), Jensen
and Meckiing (1976), Fama (1980), Rozeff (1982), Fama and Jensen (1983),
Easterbrook (1984), and Jensen (1986). Of these authors, Ross (1973), Fama (1980),
and Fama and Jensen (1983) are dealing with the general agency problems that arise
when the agents do not act in the best interests of the principal. They argue that these
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 10
agency problems developing In any princlpal-agent relationship are characterized by the
aberrant activities of the agent from the principal's viewpoint. They also suggest that,
since the purpose of a firm Is to maximize the utility of the principals, control of agency
problems Is necessary for the survival of a firm. In this regard, the contribution of
Jensen and Meckiing (1976) Is In viewing agency problems as quantifiable costs Incurred
In agency relationships; monitoring costs, bonding costs, and the residual loss.
Rozeff (1982) Is among the first authors who attempt to explain dividend
payment In light of agency theory. He argues that "a wealth maximizing firm adopts an
optimal "monitoring/bonding" package which acts to reduce agency costs. — the
payment of a dividend Is a device, like a bonding cost or an auditing cost, which Is
employed to reduce agency cost of equity* (Rozeff (1982), p. 2501. Assuming that the
firm Is raising new funds to finance the payment of dividends, Rozeff posits that a firm
seeks an optimum dividend policy minimizing the sum of dividend agency costs and
external financing costs. He suggests that a higher dividend payout ratio associated
with smaller agency costs should be linked to lower Inside ownership and a large number
of stockholders. Consistent with his hypotheses, Rozeff finds that the dividend payout
ratios are negatively related to the percentage of stock held by Insiders, whereas they
are positively related to the firm's number of common stockholders.
Pointing out that Rozeff (1982) does not show any dividend mechanism of
mitigating agency problems, Easterbrook (1984) attempts to explain simultaneous
payment of dividends and raising of new funds In view of agency theory. Easterbrook
defines two forms of agency cost as monitoring cost and risk aversion of managers, and
he presumes that dividend payment forces firms to tap new funds from the capital
market. Easterbrook argues that agency problems characterized by monitoring and risk-
aversion problems are reduced when firms enter capital markets for external financing,
because those firms are efficiently reviewed and monitored by Investment banks and
Intermediaries In the market. His conjecture Is that dividends are paid In order to keep
firms In the market subjecting managers to consistent monitoring. Since other financial
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 11
devices can be supposed to keep firms in the capital market, his argument is considered
a naive explanation of dividends that does not explain the dividend itself. Easterbrook
himself acknowledges this criticism by saying that "dividends exist because they
influence the firms' financing policies, because they dissipate cash and induce firms to
float new securities" (Easterbrook (1984), p. 652).
Jensen (1986), elaborating on takeovers, emphasizes the role of dividends in
reducing free cash flows at managers' disposal. Defining free cash flows as cash flows
exceeding the funds required for investments. Jensen argues that a firm's substantial
free cash flow aggravates agency problems between stockholders and managers. Under
this theory, managers have strong incentives to expand the resources under their control
and they are likely to waste funds on inefficient projects. Jensen suggests that
dividends should be paid out in ways that instigate managers to gorge the cash beyond
the optimal amount. This implies that free cash flow positively determines dividend
payments.
Pecking order theory, suggested by Myers (1984) and Myers and Majluf (1984),
is at variance with agency theory. Although both theories imply that dividends are paid
to reduce asymmetric information problems, pecking order explanation considers
dividend payments when managers have superior information, whereas agency theory
explanation is concerned with how the interests of agents can be aligned with those of
shareholders.
According to Myers and Majluf (1984), a firm's issue invest decisions are
affected when managers know more about the true value of the firm than outsiders do.
The managers having access to true information are assumed to act in the interest of the
shareholders. If the stock is undervalued, the firm would be reluctantly forced to either
issue stock at a low price for a good investment opportunity, or pass it up. If the stock
is overvalued, the firm would have incentives to take advantage of investors by issuing
stock at a high price. These adverse selection problems can be averted if the firm has
ample financial slack. With adequate financial slack, the firm can avoid having to issue
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 12
stock at a low price when it needs funds for investments. Financial slack also prevents
the firm from exploiting investors by sending a strong negative signal, if the firm
attempts to finance investments by issuing risky stock rather than using internal cash
flows or safe debts.
Myers and Majluf (19841 therefore argue that, since the financing pecking order
ranks internally generated cash flows first, followed by debt and finally equity, firms are
interested in building up financial slack by restricting dividends. In pecking order theory,
a firm's free cash flow representing financial slack is negatively related to dividend
payments.
B. Dividends and Taxes
Elton and Gruber (1970) originally proposed the tax-effect hypothesis, which
suggests that the differential tax rates of dividends and capital gains affect the ex-
dividend behavior of a stock. They assume long-term investors who have already
decided to sell a stock around the ex-dividend day. The only concern of these investors
is the timing decision of whether to sell before or after the ex-dividend day. In the case
of no transaction costs, equilibrium market ex-dividend prices will be determined such
that a stockholder with different tax rates on dividends and capital gains will be
indifferent as to selling the stock between, before, or after the ex-dividend day. This
indicates that the equilibrium market price on the ex-dividend day should reflect the
value of dividends vis-a-vis capital gains to the marginal shareholders.
Elton and Gruber argue that price change on ex-dividend days implies the
marginal shareholders' tax rates. Observing the relationship between this implied
stockholders' tax rates and the firm's payout ratio, they support a clientele effect
hypothesized by Miller and Modigliani (1961). The clientele effect suggests that
stockholders in lower tax brackets prefer high-yield stock, whereas those in higher tax
brackets prefer low-yield stock. Elton and Gruber find that, with the exception of the
first and eighth decile out of ten deciles, the price change-to-dividend ratio increases up
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 13
to 1 as the dividend yield rises. Since the price change to dividend ratios less than 1
indicate low dividend yields and high implied tax rates, they argue that low dividend
yields should attract stockholders in relatively higher tax brackets.
Criticism of the tax-effect hypothesis dates to Black and Scholes (1973). They
document unusual returns for several days on each side of the ex-dividend day. This
phenomenon is referred to as the ex-dividend period anomaly because the tax-effect
hypothesis is unable to interpret this behavior. The tax-effect hypothesis is later
exposed to serious criticism by Eades, Hess and Kim (1984) and Grinblatt, Masulis and
Titman (1984). They examine the ex-dividend day returns for non-taxable cash
dividends, stock dividends, and splits and find that the returns behave as if dividends are
taxable. These findings may seriously weaken the validity of the tax-effect hypothesis.
Miller and Scholes (1982), on the other hand, offer an alternative explanation,
arguing that the excess returns on the ex-dividend day, if any, are expected to be
eliminated by short-term traders. Every investor, taxable or not, has a strong incentive
to take advantage of profit opportunities on the ex-dividend day. Individual investors are
likely to be constrained from cum-ex trading by regulatory provisions because of their
tax status. But corporate traders such as brokers and dealers are induced to exploit
profit opportunities on ex-dividend days, because they are eligible for the 85% corporate
dividend tax deduction and taxable at the same rates on dividends and capital gains.
These traders have to face the round-trip costs in their trading. Transaction costs may
well provide a short-term trading equilibrium, securing above-normal returns on the ex-
dividend day.
Kalay (1982) reviews past studies evidencing the ex-dividend day anomalies and
finds potential biases inherent in those studies. He argues that these biases may result
from improper use of closing prices on ex-dividend day or correlation techniques with
some of dependent observations. He adjusts for these biases, but obtains similar results
consistent with the tax-effect hypothesis. However, the striking insight of Kalay (1982)
is in showing that, because of short-term profit elimination on ex-dividend day.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. u
stockholders' marginal tax rates cannot be inferred from those ex dividend price drops
less than dividend per share.
Elton, Gruber and Rentzler 11984) dispute Kalay (1982) by arguing that Kalay
has underestimated transaction costs to the extent that short term trading is profitable
to investors. Their argument is that the costs associated with bid and ask spread,
clearance, and transfer taxes should be included in transaction costs; then, short-term
trading around ex-dividend days would no longer be profitable. Responding to Elton,
Gruber and Rentzler (1984), Kalay (1984) provides evidence that transaction costs are
not actually large enough to constrain short-term trading. Kalay concludes that short
term trading, as well as long-term tax rates, is a determining factor of equilibrium prices
around the ex-dividend day.
Lakonishok and Vermaelen (1983) and Booth and Johnston (1984) extend the
Elton and Gruber technique to test a Canadian tax reform. They note that, under the
Canadian 1971 tax reform, dividends and capital gains are not differentially treated for
tax purposes. Examining the ex-dividend stock prices during the post-Canadian tax
reform period, Lakonishok and Vermaelen (1983) find the price changes smaller than the
dividend per share on the ex-dividend day. This is interpreted as the result of short-term
trading activities, which is inconsistent with the tax-effect hypothesis.
The concern of Booth and Johnston (1984) is to investigate whether marginal
tax rates can be inferred from the price changes on ex-dividend days in Canada. They
do not obtain much evidence consistent with dividend tax clienteles. Booth and
Johnston also find ex-dividend day price ratios between zero and one. This shows that
the market prefers capital gains to dividends. Booth and Johnston thus argue that
capital gains may still be treated more favorably than dividends because individual
investors can maintain a tax timing option on the realization of capital gains.
Barclay (1987) supports the tax-effect hypothesis by testing the ex-dividend
stock prices before the enforcement of federal income tax. His findings are that in a
world without taxes, dividends are perfect substitutes for capital gains, and stock price
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 15
drop on ex-dividend days is always equal to the full amount of dividends. Barclay argues
that, in a world with taxes, investors receive a compensating premium for the tax
penalization of dividends relative to capital gains on the ex-dividend day. Positive excess
returns are observed on ex-dividend day, which reflects the higher dividend taxes
compared to taxes on capital gains. This means that the relative taxation of dividends
and capital gains alone determines the ex-dividend prices. It is thus predicted that the
higher the effective dividend tax is than capital gains tax, the greater is dividend tax
penalization and the larger are positive excess returns on ex-dividend day.
Lakonishok and Vermaelen (1986) analyze trading volume around ex-dividend
days to test for tax-induced trading. The original contribution of Lakonishok and
Vermaelen is in realizing the importance of trading activities by short-term traders and in
investigating trading volumes around the ex-dividend day. Their hypothesis is that an
observed increase in trading volume will be followed by short-term trading around ex-
dividend days. They also predict that abnormal trading volume caused by short-term
trading has a negative relationship with transaction costs.
Consistent with their hypothesis, Lakonishok and Vermaelen (1986) find that
trading volume increases significantly around ex-dividend days. They ascribe this result
to the existence of short-term traders who are tempted to capitalize on profit
opportunities around ex-dividend days. It is interpreted that pressures by the short-term
traders result in abnormal trading volume on ex-dividend day.
Another test in support of short-term trading is undertaken by Karpoff and
Walkling (1988). They argue that short-term trading complements the dividend tax
penalty explanation of positive ex-day returns. Karpoff and Walkling propose that ex
day returns positively related to transaction costs are eliminated up to marginal
transaction costs. Using four proxies for transaction costs, they find positive
relationships between transaction costs and ex-day returns, which turn out to be
apparent in high-yield stocks after the enactment of negotiated commissions.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 16
Green (1980) and Grundy (1985) suggest the delay and acceleration hypothesis.
Their arguments state that trading activities around the ex dividend day will concentrate
on the last cum dividend day and on the first ex dividend day, because investors attempt
to avoid costs resulting from delaying or accelerating transactions relative to their
optimum trading date for tax purposes. High dividend taxed sellers and low dividend-
taxed buyers who want to trade cum-dividend will be induced to transact on the last
cum-dividend day. High dividend-taxed buyers and low dividend-taxed sellers who want
to trade ex-dividend will be induced to transact on the first ex-dividend day.
On the other hand. Long (1977) is the first author to consider dividend and tax
problems in view of portfolio choice. The basic idea of his portfolio approach is that the
ex-dividend value of stock and short-term trading behavior around ex-dividend day
should reflect the trade-off of risk and expected return. Long argues that the market
value of stock is given by a linear function of its expected end-of-period value, the
covariance of its end-of-period value with the end-of-period value of all risk assets, and
its end-of-period dividend. He suggests that "the portfolio dividend yield choice cannot
be made independently of the risk-expected return trade-off, since the dividend yield of
all mean variance efficient portfolios is a linear function of their non-diversifiable risk”
(Kalay (1982), p. 1059).
Long (1977) shows that, with the introduction of income taxation, investors will
demand after-tax efficient portfolios, which may not be before-tax efficient. For
example, under the tax regime that raised dividend tax rate relative to capital gains tax,
investors are induced to revise their portfolio holdings to reflect new tax differentials.
That sort of portfolio revision will lead to an efficiency gain at the new tax rates because
"such a move will both increase the expected after-tax return at the new tax rates and
reduce the after-tax variance of the portfolio" (Long (1977), p. 391. An increase in the
expected return is to be anticipated in the tax-effect hypothesis since the tax premiums
occur with an increase in dividend tax or a decrease in capital gains tax, but its variance
is not considered in the tax-effect hypothesis.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 17
Accordingly, the after-tax efficient portfolio will dominate any given before-tax
efficient portfolio on an after-tax basis. Furthermore, Long predicts that the potential
after-tax efficiency gains due to moving from the before-tax efficient portfolio to the
dominating after-tax frontier will be smaller if the correlations between two portfolios are
large.
C. Dividends and Stock Price
It has been accepted by many authors [Graham and Dodd 11951); Walters
(1956): Gordon (1963); and Asquith and Mullins (1983)1 that investors prefer dividends
to capital gains, and that firms could increase the market value of their shares by
choosing a generous dividend policy. These arguments belong to the "bird in the hand"
hypothesis and are perhaps the most popular and durable arguments for dividends. This
notion implies that, because stock prices are highly variable, dividends represent a more
reliable form of return than capital gains.
Among the first authors in favor of this dividend preference are Graham and
Dodd (1951). Graham and Dodd state, "The considered and continuous verdict of the
stock market is overwhelmingly in favor of liberal dividends as opposed to niggardly
ones." Elaborating on his "bird in the hand" argument, Gordon (1963) shows that
dividend has an influence on stock price because investors discount the expected stream
of future dividends. According to Frankfurter and Lane (1990), this "bird in the hand"
hypothesis implicitly assumes that there are two rates of return for evaluating future
cash flows: the investors' opportunity rate and the firm's opportunity rate. These two
rates are completely known under symmetric information. Dividend policy is then the
consequence of the relationship of these two rates.
On the other hand. Miller and Modigliani (1961) note that their dividend
irrelevance proposition holds even under uncertainty, but they puzzle over the observed
fact. They observe that in the real world a change in the dividend rate is often followed
by a change in the market price. Miller and Modigliani hint that the "informational
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 18
content of dividends” may be considered as a way of reconciling the dividend irrelevance
theory with this observed fact. But what they had in mind is different from the formal
signalling models that Spence (1973) invented later. Asquith and Mullins (19831
demonstrate that dividend policy affects shareholders' wealth because dividends
provide valuable information to investors. Asquith and Mullins recognize that dividends
may contain information.
The major problem with these dividend preference models - thougn it is very
common to other dividend models ~ is the inability to explain observed behavior
thoroughly. Some firms do not pay dividends, whereas other firms pay dividends
following distinct patterns of dividend policy. For example, one explanation for observed
dividend policies may be "the needs of individual investors.” Feldstein and Green (19831
point out that "there is the desire on the part of small investors, fiduciaries, and
nonprofit organizations for a steady stream of dividends with which to finance
consumption” {ibid., p. 17). This argument germinates a signalling model such as that
of John and Williams (1985).
D. Dividend Signalling
Signalling models were developed under the assumption of asymmetric
information, with corporate insiders being better informed than the market as a whole.
These models seek to explain dividend payment in the context of its informational
content. Signalling models originate in the works of Akerlof (1970), Spence (1973), and
Riley (1979). Bhattacharya (1979) is the first to view a dividend as a signal of
management's private information about future cash flows. Later, Miller and Rock
(1985), John and Williams (1985), and Kumar (1988) have developed different models
of dividend signalling. In dividend signalling models, dividends are assumed to affect
stock price because the market believes that they signal favorable insider information.
Management is motivated to pay dividends in order to convey information and thus
achieve a higher stock price.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 19
As Miller (19871 specified, dividend signalling models are characterized by three
key conditions. First, the dividend is intended to signal the firm's attributes such as its
current and future earnings, which are asymmetrically known to insiders. Insiders act in
the best interest of shareholders. Second, the beneficiaries of signalling must exist. The
higher stock price induced by signalling benefits (11 the firm having easy access to
future public floatations (e.g., Leland and Pyle (19771 and Harris and Raviv (1986)]; (2)
the current shareholders selling their shares [e.g.. Miller and Rock (1985) and John and
Williams (1985)1; and (3) management who is compensated by the rise of stock price
[see, for example, Ross's (1977) incentive'Signailing model]. Third, dividend payments
should incur cost penalties including personal taxes, the firm's transaction costs of
funding liquidity shortfalls, and costs of underinvestments.
In signalling models, these sufficient conditions for signalling equilibrium are
manifested in the form of maximizing shareholders' objective function. This qualitatively
differs from other dividend models dealing with its announcement effects. Thakor
(1989) raises another puzzling question: "Why dividends are chosen as a signal when
less costly signals are apparently available 7 Why management does not convey
information in some other way involving less tax cost to the stockholders than the
payment of dividends?" [Thakor (1989), p. 433].
The first model in dividend signalling is that of Bhattacharya (1979). He
assumes that current shareholders care about their firm's present value and that outside
investors cannot differentiate the quality of projects undertaken by the firm. Outsiders
correctly appreciate the firm's stock and then buy its stocks at the correct price in the
perfect capital market. Bhattacharya develops an equilibrium model that relates the
benefits and costs of dividend signalling. In his signalling model, taxable dividends are
paid in order to signal expected cash flows of the firm. This dividend maximizes the
after tax objective function of shareholders, which incorporates the signalling benefit of
dividends coming from the increase in liquidation value, conditional on dividend payment.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 20
The tax-based cost structure generates feasible signalling equilibria because it is
negatively related to true expected cash flows. An asymmetric transaction cost is also
embodied in his signalling model. Bhattacharya (1979) argues that larger values of
corporate private attribute reduce the extent of outside financing. If the firm has
insufficient cash flows to make up its promised dividend, then it must rely on outside
financing and incur a higher transaction cost relative to the case of sufficient cash flows.
This transaction cost is asymmetric in the sense that larger values of private attribute
known only to insiders decrease the present value of future transaction costs and
thereby decrease the marginal signalling cost. This marginal signalling cost, which is
strictly negatively related to true expected cash flows, weighs the marginal benefits of
dividend signalling and produces a signalling equilibrium.
Miller and Rock's model (1985) works with the similar assumption that outside
investors cannot observe the firm's current cash flows. The investor infers that, since
corporate earnings have a great deal of persistence, current earnings convey information
about the firm's future prospects. Investment opportunities are available to all firms
with diminishing marginal returns, and outside financings are costlessly raised through
sales of corporate bonds. Dividends are perfect substitutes for repurchases of bonds,
and they are not taxable differentially from capital gains. Miller and Rock also assumes
that, unlike Miller and Modigliani (1961), investments are determined as a residual in the
financial nexus.
Miller and Rock (1985) argue that, in the context of information asymmetry,
paying dividends instigates the market to believe higher current earnings in the firm, thus
bidding up the stock price. Corporate insiders then have incentives to signal by
distributing more dividends and cutting investment below the optimal level. Even
insiders with less valuable projects attempt to mimic more valuable firms by paying more
dividends and forgoing profitable projects. Since dividend payments reveal cash inflows
and firms forgo projects with positive net present values, the Fisherian optimum
investment policy breaks down.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 21
Miller and Rock (1985) suggest one possible route for achieving a signalling
equilibrium, as follows. Outsiders understand that more informed insiders are tempted
to take advantage of their superior information. Outsiders are therefore induced to
discount the stock price offered on dividend announcement day, while corporate insiders
are exploiting a departure from the Fisherian optimum. These conflicting pressures lead
to an equilibrium, restoring the consistent dividend and investment policies even at the
sacrifice of efficiency.
John and Williams (1985) develop a signalling equilibrium with dividends and
personal tax on dividends. In their model, capital gains are not taxable, and issuing,
retiring, or trading shares does not incur any cost. According to their arguments, if the
liquidity demand by the firm and its current shareholders exceeds internally generated
funds, corporate insiders are motivated to distribute a taxable cash dividend and to
reveal to outside investors the present value of their firm's future cash inflows. This
signal would result in a raising of stock price and a benefit to current stockholders.
John and Williams (1985) attempt to explain why some firms do not pay dividends,
whereas others do pay dividends and simultaneously sell new shares to investors. In
order to finance investments, a firm needs to either issue new shares or retire fewer
outstanding shares. Similarly, to collect cash for personal use, current stockholders
must sell their shares. In either case, stockholders have to suffer some dilution in their
fractional ownership of the firm. Current shareholders desire to reduce this dilution on
corporate or personal accounts, and insiders, in the best interest of current stockholders,
are induced to convey their favorable information by paying a taxable dividend.
Recognizing the relationship between favorable information and stock price, the market
bids up the stock price and thereby mitigates stockholders' dilution.
The signalling equilibrium of John and Williams (1985) is achieved because the
marginal gains from paying dividends are balanced against the marginal cost incurred by
dividend tax. In the market, stocks with marginally larger dividends are responded with
the premiums, following the reduction in dilution for current shareholders. These
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 22
benefits compensate shareholders for taking proportional costs due to dividend taxes.
For firms paying marginally smaller dividends, the marginal dead-weight costs of
dividends outstrip the marginal benefits of reducing ownership dilution. In John and
Williams' (1985) model, an optimal signalling equilibrium is derived when firms with
favorable inside information distribute higher dividends and receive higher stock prices
for their shareholders.
Following Miller and Rock (1985), Kumar (1988) assumes that the dividend
signal produces endogenous investment. He then argues that dividend changes can be
used as a "coarse" signal kwcause dividends reflect partitioned spaces of all possible
corporate prospects and thereby dividend changes represent "broad" changes in these
prospects. Kumar states that these coarse signalling equilibria attain the simultaneous
explanation of several extant dividend anomalies such as (1) information effects of
dividend changes (e.g.. Petit (1972) and Aharony and Swary (1980)]; (2) dividend
smoothing [e.g., Lintner (1956), Fame and Babiak (1968), and leub (1972)1; and (3)
dividends as a poor predictor of future earnings [e.g.. Penman (1983)1.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Chapter 3. THE DETERMINANTS OF DIVIDEND POLICY
A. Introduction
This essay empirically investigates the determining factors of dividend policy. In
this study, a structural model is specifically designed and tested, encompassing
determinants from previous dividend theories. The motivation for this study is that
many of these previous theories have not been tested, and that the econometric
techniques used thus far are inappropriate.
The vast body of literature dealing with dividend determinants can be grouped
into two distinct categories: symmetric and asymmetric information, in the symmetric
information milieu, the seminal work is that of Lintner (1956). According to Lintner's
model, the current dividends are predicted on the current profits and past dividends.
This "two-variable* model is supported by evidence by Fame and Babiak (1968). On the
other hand, in an idealized world. Miller and Modigliani (1961) show that the dividend
decision must be independent of the investment decision. This is because higher
dividends would induce more external financing through the issue of new shares to keep
the investment level unchanged. Miller and Modigliani (1961) also argue that the
dividend decision is irrelevant for a growing firm under the conditions of taxes and debt
financing. This separation proposition on dividend and investment has motivated many
people to investigate empirically the dividend subject with mixed results. Higgins
(1972), Fame (1974), Smiriock and Marshall (1983), and Partington (1985), for
example, find that corporate dividend and investment decisions are separable and
independent. Ohrymes and Kura (1964, 1967) show that dividend and investment
decisions are strongly interrelated. McCabe (1979) also provides strong evidence of
interdependence among the dividend, investment, and financing decisions, and he
argues that dividends are affected by new long-term debts.
23
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 24
Theories based on the premise of asymmetric Information Includes agency,
pecking order theory, and dividend signalling. Agency theory explanations of dividend
behavior build on the works of Jensen and Meckling (1976), Fama (1980), Rozeff
(1982), Easterbrook (1984), and Jensen (1986). The argument Is that agency costs
associated with free cash flows positively determine dividend payments. Pecking order
theory, suggested by Myers (1984) and Myers and Majluf (1984), Is at variance with
agency theory. In these papers it Is argued that, since the financing pecking order ranks
Internally generated cash flows first, followed by debt and finally equity, firms are
Interested In building up financial slack by restricting dividends. In the pecking order
theory, a firm's free cash flows are negatively related to dividend payments.
In addition, many authors have extended theoretical and empirical works on
other dividend determinants. Michel (1979) and Michel and Shaked (1986) consider
Industry Influence on dividend policy. In Feldstein and Green (1983), size and risk
aversion are assumed to Influence the dividend decision. Dyl and Hoffmeister (1986)
suggest that dividend policy should be reflected In the beta coefficient as a proxy of the
riskiness of a firm .'
The work here differs from earlier studies in three ways. First, in this essay,
time series cross-sectional tests are undertaken using individual firm data, while the
vector autoregressive (VAR) methodology is applied to aggregate data. These
methodologies are more appropriate In the sense that previous studies employed cross-
sectional analysis of either firm-specific or aggregate data and they lost Important
Information explaining differences in dividend policy over time. Second, this study
includes empirical tests of agency theory and pecking order theory associated with free
cash flows. The literature In this area Is limited or non-existent. Third, this study
'Rozeff (1982) also shows that beta coefficients are negatively correlated with dividend payout ratios.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 25
examines a wide scope of dividend determination theories including industry, size, and
beta. Section B is a description of the sample and the variable definitions. In Section C,
the methodology is explained. Section D presents tests, results, and their interpretation.
Section E is the conclusion.
B. Data
The tests are undertaken using both individual firms' data and aggregate data.
Five firm-specific accounting variables - dividends, earnings, investments, long-term
debts, and free cash flows - are analyzed over the 1979-1990 period using quarterly
data because firms usually pay dividends on a quarterly basis. The aggregate data are
obtained by cumulating individual firms’ data.' The effects of firm size, beta, and
industry are also tested. The data sources are the Quarterly COMPUSTAT tapes and the
CRSP files.
1. Sêoipl» SeheHon
The sample used in this study is taken from two COMPUSTAT quarterly tapes
for the period 1979-1990. The 1991 COMPUSTAT tape is used to obtain the data for
1980-1990. The 1990 COMPUSTAT tape is used only for 1979.
To be included in the sample, the data must meet several screening criteria.
First, all data required for this study must be complete for the sample period for each
firm. All data items used to represent or calculate dividends, earnings, investments,
long-term debts, free cash flows, and firm sizes must be in the COMPUSTAT files
throughout the sample period. Second, the firms must keep a December fiscal year-end
for the sample period. This screening allows matching of quarterly data items across
The tests conducted for aggregated data may be affected by the aggregation bias because some of the variables are not necessarily non-negative. This aggregation problem will be explored later by testing for sensitivity to sample sizes.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 26
firms. Third, the firms must be listed on the Center for Research in Security Prices's
(CRSP) monthly file for the sample period. The firms' SIC codes for classification of
industries are also obtained from the CRSP file. Application of these criteria during the
screening process yielded a final sample of 446 firms.
2. Variabh OefhA&vrs
Variable definitions used in the analysis and their sources are as follows. All
accounting variables are seasonally unadjusted quarterly data, it is assumed that
because of budget constraints, firms are more interested in the levels of accounting
variables than in their percentage changes. Since the purpose of this study is to
investigate the relationships among five quarterly variables, flow variables that uniquely
accrue to each quarter are needed for dividends, earnings, investments, long-term debts,
and free cash flows. Investment and long-term debt are stock data obtained from each
firm's balance sheet, whereas earnings and free cash flows are flow data calculated
from the firm's income statement. Accordingly, net quarterly changes are calculated for
investments and long-term debts in order to get the amounts allocated to each quarter.
DIV: Three different proxies of dividend are usually discussed in the literature:
(1) amounts of common stock dividends paid (Lintner (1956), Dhrymes and Kurz (1967), Higgins (1972), McCabe (1979), and Smiriock and MashalK 1983)1; (2) change in dividends (Fama and Babiak (1968), Fama (1974)1; and (3) dividend payout ratio [Rozeff (1982)1.
The dividend proxy used in this study is the amounts of common stock dividends
paid. This measure is calculated by COMPUSTAT data item #15 (common
shares used to calculate earnings per share) * data item *1 6 (dividends per
share).
/NV: Proxies of investments include;
(1) change in net plant and equipment (Higgins (1972), Fama (1974)1; and (2) investment in fixed assets (Ohrymes and Kurz (1967)1.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 27
The measure of investments used In this study is the change in property, plant,
and equipment less depreciation. This measure is given by COMPUSTAT data
item *42 (total property, plant, and equipment) • lagged data item #42.
EARN'. Available earnings after taxes for common stockholders are defined as:
Net Income • Preferred Dividends [Lintner (1956), Higgins (1972), and
Fama (1974)1.
The earnings measure in this study is obtained from COMPUSTAT data item #25
(income before extraordinary items, available for common).
DEBT'. This includes net new long-term debts (McCabe (1979)1.
The measure of debts used in this study is calculated by COMPUSTAT data item
#51 (total long-term debts) - lagged data item #51.
FCASH: A measure of free cash flows is calculated according to Lehn and Poulsen
(1989) as follows.*
CF = INC - TAX - INTEXP - PFDDIV - COMDIV, where INC > operating income before depreciation, TAX “ total income taxes, minus change in deferred taxes from the previous year to the current year, INTEXP « gross interest expense on short- and long-term debt, PFDDIV > total amount of preferred dividend requirement on cumulative stock and dividends paid on noncumulative preferred stock, COMDIV > total dollar amount of dividends declared on common stock (Lehn and Poulsen (1989), p. 7771.
This measure is also used by Lang, Stulz and Walkling (1991). Unlike their
annual measures, the quarterly measure of free cash flows used in this study is
calculated by COMPUSTAT data item #21 - (data item #8 - change in data item
#52) - data item #22 - data item #24 - (data item #15 * data item #18).
'Although this measure of free cash flows is computed by several accounting numbers, including operating income and dividends, FCASH is an another independent variable. Thus, the parameter estimation on FCASH is not influancad axclusivaly by ona or two input variables (e.g., operating income and dividends). The intrinsic problem of interdependencies among accounting numbers would be considered, but simultaneous estimation methods and structural models can produce superior estimations under such conditions (see Appendix A).
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 28
SIZE: Two proxies of firm size are:
(1) the logarithm of book assets [Wansiey and Lane (1987), Friend and Hasbrouck (1988) I; and (2) market value of equity outstanding.
The proxy of firm size used in this study is the logarithm of book assets, which
is given by COMPUSTAT data item *44 (total assets).
INDUSTRY: Four digit SIC industries are obtained from CRSP.
The first two-digit SIC codes are used to determine industry.
BETA: The market model is used to obtain beta coefficients (see Sharpe (1963)1.
C. Methodology
1. Tim»-S9ri9S Cross-Sêctionêl Aniyais
First, a time-series cross-sectional model is applied to individual firms' data. A
generalized linear model is considered, where observations occur at T time periods for
each of N firms. The model at time t for each firm I is designed as
DIVg~9f,* e , INVg * 0 2 EARNg * 8 , DEBTg * 8 4 FCASH, * u, , , 3 ^ , /■i,...,Af N i,...,r
where 9 are the coefficients, and Un is a regression error term.
To estimate 0 , the error components model of Fuller and Battese (1974) is
followed. In the error components model, the regression error u* is assumed to consist
of three independent components.
Wt ■ Y; ♦ */•*•« # • 13 2)
where k, is an unique cross-sectional effect, k ~ (0 ,o /); d, is an unique time effect, d, (0 ,a /); and f* is an error term, f , ~ (0 ,0,').
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 29
Under these assumptions, the covariance matrix of the regression error term u„ is
a[ «/') - oj ♦ 0* <3.3)
where I* is an NxN identity matrix, ® is the Kronecker product matrix, and Jt is a TxT matrix of ones.
Fuller and Battese found that premultiplying equation (3.1) by the covariance
matrix yields the uncorrelated errors with variance a /. The observations are thus
transformed by the covariance matrix, and an OLS regression is run on this transformed
data. Then, unbiased and efficient generalized least squares (GLS) estimates are
obtained.
The estimated 9 values and t statistics for H,: 8 = 0 are used to determine the
relationship between dividends and explanatory variables.
2. V«etof Autongfttsiv* Modt!
Vector autoregressions are employed to analyze the aggregate data (see, for
example, Sims (1980), Lhterman (1986), Lupoletti and Webb (1986), Holtz Eakin,
Newey and Rosen (1988), Blanchard (1989), Keating (1990), and Clements and Mizon
(1991)1. Under the VAR model, a system of dynamic linear equations are constructed
such that a vector of dependent variables is related to lagged vectors of all variables in
the system. For example, the dividend equation takes the form;
M M M (3.4)
where the a, f u, x and « are the coefficients, m is the lag length, and e, is a white noise error term.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 30
Since the model is assumed to consist of a system of five equations, the structural VAR
model has the following form:
X • Am X * C * E, (3.5) aSxl) (5*5) (5x1) (5x1) (5xf)
where X is the vector of variables (i.e., DIV, INV, EARN, DEBT and FCASH), A (LI is a matrix polynomial in the lag operator L (i.e., A(L) -A ,L + AjL* + ... + A^L"), C is a vector of constant terms, and E is a vector of error terms (i.e., eo*v>
Equation (3.5) can be rewritten compactly as
X ~ (I - Am ) X • C * E , 136)
where B(L) is a matrix polynomial in the lag operator L (i.e., B(U-l-A,L-A,L*-...-A„L"’), and I is the identity matrix.
If X is stationary and the roots of B(L) lie outside the unit circle, equation (3.6)
can be inverted as foilows.
X - Bm'^ c ♦ ' E 13 7)
According to Wold's decomposition theorem, the vector process X of equation
(3.7) consists of two components: a deterministic one and an indeterministic one. This
can be represented as a vector moving average.
♦ E • / «r-/ > hO
where \ is a constant, 0 , is the coefficient matrix, and e, is a vector white noise.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 31
Following Sims (1980), the VAR methodology generally uses two statistics
which are derived from the coefficient matrix 0 in equation (3.8). The first VAR
statistic is the variance decomposition of forecast errors for X,, which are constructed to
show the dynamic relationship between the system variables. This measures the
proportion of each variable's contribution to the forecast error variance in the h-step
ahead forecast of X,. Since the h-step ahead forecast of X, is
♦ E ♦ / «►*-/ • hh
the forecast error and its variance matrix are caiculated as foilows:
*-i " E •/ *hh-i . K (3.10) vu h K - - EK •/ ♦/' -E M)
The h-step forecast error variance of the variable, X», is expressed as the sum
of the k"* diagonal elements of each 0 ,0 ,' in equation (3.10), where the k*** diagonal
elements of 0 ,0 ,' is defined as the sum of the squares of elements in the k" row of 0,.
E (ÊC). M Ml m>1
where 0kmj the km*^ element of 0„ and n is the number of variables.
The contribution of the j* variable, X,, to the h-step forecast error variance of X„ is
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 32
*•1
M
where is the kj'” element of 0 ,.
Then, the relative proportion of the j"" variable, X,, to the h-step forecast error variance
of Xk is measured by
(3.13) »-i » E(EO M M>1
Variance decomposition allocates the h-step forecast error variance for each
variable into the components accounted for by its own surprise movements and by
shocks to other variables in the system. Since the error variance of a variable is
identified as the sum of contributions by all variables in the system, variance
decomposition can be used to investigate the influences of a variable on another
variable. For example, if investment makes significant contributions to the forecast error
variance of dividend, then this will imply that investment is a major influence on
dividend.
The second test statistic in VAR is an impulse response function (IRF), which is
meant to trace out the response path of the system variables to an unexpected unit
shock in a variable. The normalized unit shock to a variable is given by a one standard
deviation surprise movement in that variable.*
*This is often called innovation accounting in the sense that it traces out the system's reaction to a shock (innovation) in a variable [see Sims (1980)1.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 33
Since the impulse response of X, to the shock e, is measured as the h*”
moving average coefficient of X, is obtained from equation (3.9).
. (3.14) a«, *
Consequently, this impulse response function can be regarded as dynamic multiplier that
indicates the size and direction of response of system variables to an unit shock in a
variable.* For example, if impulse responses of dividend to a unit shock in investment
are significant and negative, then Miller and Modigliani's proposition of irrelevance can
not hold because dividend and investment would be substitutes for the firm's funds.
Prior to the VAR estimation, the error terms e, in equation (3.8) must be
orthogonalized in order to eliminate any contemporaneous correlation. To obtain the
orthogonalized errors, the Choleski decomposition transforms equation (3.8) by using the
lower triangular matrix, P, from the covariance matrix, E, as follows.
M (3.15)
M
where O,* = (P,P is the transformed coefficient matrix, and 6,* > P‘'e, is the transformed vector white noise.
This VAR model has much appeal over earlier models. In the VAR, all variables
are jointly modeled to represent their dynamic relationships without imposing any «
priori restrictions. The VAR also has no risk of misspecifications such as serially
correlated and heteroscedastic errors.
*See McMillin and Parker (1990).
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 34
D. Empirical Rasults
1. S iflip H »ChênetÊristha
Table 3.1 presents sample characteristics for the 446 firms included in this
study. The cross sectional medians, means, and standard deviations for each variable's
averages from the second quarter of 1979 to the fourth quarter of 1990 are shown in
Panel A of Table 3 .1 . It is interesting to note from the mean column that the sum of
DIV and INV representing outflows of corporate funds approximates the sum of EARN,
DEBT, and FCASH representing inflows of corporate funds. This seems to be consistent
with Dhrymes and Kurz's (1967) suggestion that one of the firm's long-term objectives
is to grow while keeping the cash flow balance. The standard deviation of SIZE is quite
large compared to that of other variables, which may be the cause of heteroscedasticity.
The correlation matrix between the variables is shown in Panel 8 of Table 3.1.
Accounting variables are standardized on SIZE in order to adjust for potential
heteroscedasticity. There are also several other points to note in Panel B. First, the
correlations between DIV and other explanatory variables provide a preliminary picture of
the subsequent tests. The correlations with INV and DEBT are not significant. The
correlation with EARN is positive and significant, whereas the correlations with FCASH
and BETA are negative and significant.
Second, although some correlations between the explanatory variables are
significant, they do not produce serious multicollinearity.* The correlations of INV with
EARN, DEBT, FCASH, and BETA are significant. The correlation between EARN and
"Multicollinearity Is not a problem in this study for several reasons. First, the coefficients of simple correlations are not high. Second, this kind of multicollinearity can be remedied when time-series cross-sectional regressions follow the estimated generalized least-squares (GLS) procedure with more than 20,000 observations. Third, the VAR analysis focuses on forecast error variance decompositions and impulse response functions, which are not affected by multicollinearity.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. ::ü CD ■D I
Q.s Tabl# 3.1# BêmpXm Character
■D CD PaiMl Ai Croaa-aacClonal aaan and atandard daviation of aach variablaa avaragad from 1979:11 to 1990tlV. % O Varlablm varlabla daacrlptlon Madlan Haan Standard daviatlon 3 OIV Common atoek dlvldanda paid $5.658 «21.709 $56.717 (in milllona) CD INV Changa In invaatawnt $7.359 «31.830 $81.490 8 (in milliona) ■D MM Earninga availabla to common «12.145 «40.996 «111.645 atockholdara (in milliona) C5- DEBT Changa in long-tarm dabt «3.854 «22.744 «87.316 3" (in milliona) PCA8 H Eatimatad fraa caah flow «6.746 -$6.158 $306.313 i (in milliona) 3 CD SI» Book valua of aaaata $1, 155.400 $5.735.038 $15,049.108 (in milliona) BETA Bata coafficiant 0.894 0.845 0.358 C 3. 3" Panal 8 i Croaa-aactional correlations batwaan CD from 1979:11 to 1990ilV.
CD ■D Variabla^ DIVINV EAHNDEBTFCASH
CI INV 0.0230 a (0.627) O 3 ■D EAM 0.6831 0.2452 S (0 .0001)* (0 .0001)* DEBT 0.0300 0.5078 0.0133 (0.527) (0 .0001)* (0.778) & FCASH -0.1270 0.4158 0.1023 0.0875 (0.007)* (0 .0001)* (0.030)** (0.064) O C BETA -0.2696 -0.1444 -0.1304 -0.0107 -0.0942 % (0 .0001)* (0.002)* (0.005)* (0.820) (0.046)**
% % O 3 CD ■D O Q. C g Tmbl# 3.1 (cont'd) Q.
P*n#l Cl Smmpl* by Industry ■D CD Distribution of firm sisss Sample (in millions) CO W Industry SIC cods frequency o" Median Mean standard deviation 3 0 Mining snd 1000-1999 28 (6.28%) $633.118 $2,710.202 $7,028.921 construction CD Non-dursblo 2000-2999 93 (20.85%) $1,053.903 $4,195.658 $9,405.505 8 swnu fscturing ■D Dursbis 3000-3999 127 (28.48%) $361.115 $3,441.985 $10,648.941 SMnuCscturing (O' Trsnsportstion 4000-4999 124 (37.80%) $2,112.433 $3,318.134 $3,593.406 3" mnd utilities Mh •isssis snd 5000-5999 11 (2.47%) $330.290 $1,262.387 $1,847.221 1 3 rotsii trsde CD Pinsncs, 6000-6999 52 (11.65%) $10,240.014 $23,533.952 $33,293.750 insurance and real estate 3. Services 7000-8999 11 (2.47%) $384.695 $501.159 $491.321 3" CD 446 (100%)
■DCD O ^Accounting variable# Including DIV, INV, BARN, DBBT and FCASH ara atandardisad on SISE. Q. C a ^Significant at tha 0.01 laval. O 3 ^Significant at tha 0.05 laval. ■D O
CD Q.
■D CD I (/) w W o> o" 37
BETA is negative and significant. Since these correlations are not high, they imply that
controlling for heteroscedasticity yields less multicollinearity.
Panel C of Table 3.1 displays the sample by industry classification. The firms in
each industry are distributed by size of the firm. It would appear that cross industrial
variations in firm size are evident. The finance, insurance, and real estate industry has
the largest mean of firm size. The smallest mean of firm size is found in industries such
as services, wholesale, and retail trade.
2. Tbn»-S9ri0s Crost-Saetionsl Rtgntthn Rasults
Time-series and cross-sectional regressions are estimated using the error
components model of Fuller and Battese (1974). The error components model is an
estimated generalized least-squares method for estimating regression parameters when
time-series and cross-sectional data are simultaneously used. This procedure can take
care of heteroscedasticity as well as serial correlation in the error terms.’
The estimates of time-series cross-sectional regressions are presented in Table
3 .2 , which is divided into four panels. Panel A shows the estimates for the full sample.
The most striking result is the verification of Miller and Modigliani's (1961) dividend
irrelevance proposition; the coefficient estimates on INV and DEBT are not significant.
Equally surprising is the coefficient estimate of FCASH, which is negative and
significant. This seems to imply that pecking order theory explanation holds.* The
coefficient of EARN is positive and significant, consistent with Lintner's (1956)
prediction. The F-test rejects the null hypothesis at probability less than 0.01.
’Since this procedure transforms the variables by the covariance matrix of residuals, it can control heteroscedasticity in a less restrictive way than simply dividing the variables by firm size, while it adjusts for serial correlations due to time-series data.
"As shown in Sections D.3 and D.4, the effect of FCASH should be explored in the context of dynamic interactions between FCASH and DIV.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 38
i i i i o i i i O o d d d d O d d d V V V V V V V V V V V
0 s in 0 M o 0 0 0 m m S s m m n d d
M A^ p# ^0 0 « ^#4 0 0 1» oimoMomp^MQ^ m 0 0 0 0 M P* 0 0 O # » # m » o w . 3 M § §® M , M ' o « o * m # m N o # 5 o N m p# i # M # ^ #0 % »M o7oT
0 P» 0 0 0 0 vn S o o 0 o S P# 0 0 m Otf^Oor^omoowtootdi * ## » ## *o S'îg'îgm o 0 o * • • * M *0 «0 »0 o l ô t OMO ^OMOMO I d V d V d i* m m m ^ m ^ ^ ^ # I"» p* 0 0 0 ^ m 0 m 0 0 0 M mN*^000OP»M M 0 0 pk 0 0 0 0 0m M • O • O •g 0 0 0 e «o « M M 0 o • • «4 t ^ # 0 :: O^O««O^O0O*4• ^ « o * M • • * m O N O ^ O m ô ià i
^ ^ 0 -»r* ^ m m <•» A r» «» # # # O#o^o»OMm0 om^0^p»rif<4O^ n^e^iAMainMio oii d OOONO0O0OM g s g T g : s 3 s ; ooomornomom Il ^ I I m# ^ o N o I O 0 O 0 O 0
-• -• # « « 0 ^ m ^ 0 « A ««n m#r*MmO0N00 000Om0r>Pi0m b m##mvo0o#m00 m000m«4p»0mo mOMI»0O0VM0 0O000P*O0P#m O 0 P# 0 0 I» 0 0 p» 0 p# #4 0 0 M 0 *0 . 0 ## %» 0 %# Pl#40»#m»^0w0^
O 0 O II 0 0 0 s
8 8 S s U w 0 p# m i % 0 0 0 p* % i o O d V V V V V
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 39
I I o o o e o o d V V V V V V V
1 1 1 o 3 «ne >1 o « A 4» o e o 2 SÎ is 8 g fl* 8 M o 3 :: dI V# d #4 *— «•- A III^ # 0 M *» ^ M «» 0% » S ^ “ m » e S g * Soon 8 ; • d • n d m III ? i e ■ e N Old? u n 41
- m •2 nonn « o g s gS8 % 3 §o n n n 1,5 i; 0' #* 0 ' •» X o O # 0 ^ 0 ^ I « n • I I n • I n • n • St 8 Mi?8883 illU «M • :si l i l O n d n î hi A « # ## # MACiAO^no #% m A nooiA«nnniA<«oviAA SgSS88St8S3S3g Î!1 • # * nononm^on ^ n ^ n ^ n »»* m on ^n #4* ^* M e#
II C 04 A n o O d d # *> I ! !i 4» _ • 9 9 U i i n
g 2S 8 8 t:a Z . __ - 1 3 ■? ? 8.3 : 3 8 “ 8 I Îi|ijiîl|ilî I I 8 8 i. J ii i j"
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 40
Panais B, C, and D of Table 3.2 present the estimates for subsamples classified
by size, beta, and industry. Coefficient estimates on EARN and FCASH are quite
consistent with the findings for the full sample. However, the signs and significances of
the INV and DEBT estimates reveal mixed results, compared with those for the full
sample. The estimates for intercept are interesting, since they provide some insight into
the subsequent tests. Coefficient estimates on intercept in the subsamples sorted by
size are significant and larger in magnitude as firm size increases. In contrast, for the
subsamples sorted by beta, the intercept estimates are significant and smaller in
magnitude as beta increases. Coefficient estimates on intercept in the industrial
subsamples vary both in significance and magnitude. This suggests that missing
variables may cause model misspecification collapsing in the intercept.
Table 3.3 reports time series cross sectional regression estimates using six
different model specifications. In Table 3.3, model (1) is the regression in which the
variable BETA is included. The coefficient estimate on BETA is negative and significant.
In model 12), four size variables are included. Of these, two size variables, S* and S„
representing large firm sizes, show positive and significant estimates. When BETA and
size variables are entered at the same time in model (3), the effects of beta and firm size
are the same as those in models (1) and (2).
Alternatively, model (4) includes industry variables. The estimate results show
that none of the industry variables are statistically significant at the 1 % levai. When
BETA is added to model (41, BETA shows a negative and significant estimate in model
(5), but the industry variables do not show any significance. Also shown are the
regression results of model (6) in which four size variables are added to model (4). Two
large size variables have positive and significant coefficient estimates, whereas the 6^
industry variable representing finance, insurance, and real estate reports a negative and
significant estimate of coefficient.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 41
« ^ • « ü Ü Ü iS Ü « A m«V M 9 9 9 9 9 | 2 §5 3 : * + 4 * ♦ i s i f 3 2 2 d 2 * 2d s i d s O M O ^ 7 X ^ I <9 M «m o o O « ♦ ^ *•- a m tn s §5 i s i : 0 3 î 9 I «4 8 tü o ® « 2 * O ^ ?? • ^ « • m « g * * S ST ÎT # 2 § S s " § 3 1 s M ft 2S 2Ü; 2 2 ♦ «0 ®do — 2d e — d ^ ?" S ? iJ • X HI » tn m «% g e « s s « i s 1 2 gi s d 2S d 2. o *» d ^ d2d ? 1 ♦ « I H 9 • D & »•- • S S »m in« S S s i 3 S 3 S 5 S m § : 22 • M 5 Z 2S 2 ! S o ^ e ?" s X 1< " 41 * f m in ^ o a a f** ft 0» w § ; m lO I I I 3 ® 52 ii82 2d 2 S 2d s * f# ft d S d Z {• e 9 • • o ^ d? I • ? « ♦ ♦ ♦ A# AI w «4 W #4A « m m m 2* s g 2S 5 i i M M M a T s g 5 S2 p r* N #4 0 ii is • • • • JS 2d 25 d ® o ^ d 3, 2d s? ♦ ♦ A * A A ?" OO o o • • • • u I ■ ■ ■ I Al *» AI w W WAI «4AI > > > > > > M u • 1 Q O Q O Q o ■d* <•* m tn •H W M *4 #4 #4 I I I I I I X S ;1 II 1
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 42
_ «A SS 8 * - I o 1 §5 S• s t0k 5 s # s# 5 #s ^s s s # :@ s# Ô 7 Î SSL^ 2L« 2 .7 7 ? 1 ? ? V Is41 0 O A 0 • SS SS C!'SS59Mmr*#0%mmp# ^ > ' m M ^m0g0*»o#00MN 33 0 f*«» O 0 N w ' 0 m «0 I ■ O fiT ^ • • «0 • * 0 ^ tp i • ' oo M*O 0 I * o m N o 1! 0 M m# p# ^ I %# m V I I V ^ ^0#»N00#4 lA A ^ «0^000000000 ill 0M000m0NNM0m o «• ••m o ^ • 0• 0# m• #4•« *0««00 • 0 # 4 0 M 0 # 4 # 4 # p# I , Q o -I- ^ M ^ W *# I w | vf»«M V 3 .3 0 M 0 0^«^04»0#»M0p.m0m0##00@e 0 1** mM0p*p%eop####40m 0M##m«m#m000p»p# o V M0 Wr*p# f# 0 M M m0 W ^, O^ M« M d V
m =1: « 0 m 0 0 «M - i l s g 0 m « O 0 0 M 0 0 m m r* M 0 ■ o 3 S 0 r* « 0 0 I I o o «4 d m 1 M 1 m rS 0 M , - » î i 1 1 0 V
« o m 0 M «•- o 0 0 « 0 o m r* 0 0 i'î: 0 « n M m m m 0 M 0 0 0 P» t M •H m f» r 0 1 « w 0 1 1 #4 0 o V Is: a 3 " § I S V 8 S d ".a I ■ V •il:
W i O M w « " S 11 ^ <£* »:§" ÏSS "
In Is5- j = s | * ^ i - y Ii i i J «"3 %
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. CD ■D O Q. C 8 Q. Tabla 3.3 (cont'd)
^Th# Industry dummy variables, I%e, are defined ae follows: 1%-llf mining and construction industry, ■ T3 0 otherwise; Ig » 1 If non-durable awnufacturlng industry, %2 - O otherwise; Ij • 1 if durable manufacturing CD industry, I3 • 0 otherwise; I4 ■ 1 If transportation and utility industry, I4 • 0 otherwise; Ig - 1 if wholeaale and retail trade industry, Ig • 0 otherwise; Ig - 1 if finance, insurance and real estate %(/) industry, Ig * 0 otherwise; ly • 1 if service industry, ly • 0 otherwise.
^fbr the purpose of estimation, each one dummy variables ere lost. The dummy variables not represented in the table are 83 for 81:8 variable and ly for INDU8T8T variable.
*t-etatietice are given in parentheses. 8 ^Significant*8 ignificant at thetha OeOl0.01 level.lavale
^Significant at the 0.05 level.
CD
3. 3" CD
■DCD O Q. C a O 3 "O o
CD Q.
T3 CD
(/) (/) w 44
Finally, in model (7), all the variables are included in the regression. The sets of
coefficient estimates on BETA and size variables are consistent with the estimate results
in other models of Table 3 .3 . While other industry variables are insignificant at the 1 %
level, the industry variable for finance, insurance, and real estate also shows significant
estimate results, similar to those of model (6). This casts doubt on the industry effects
in dividend policy. As noted in Table 3 .1 , the finance, insurance, and real estate
industry has the largest mean of firm size. Also that industry variable has a significant
estimate of coefficient only when size variables enter the model. This evidence, along
with the insignificant estirnates of other industry variables, suggests that industry effect
might not exist and that industrial variations in dividend policy as documented in Michel
(1979) might be only the manifestation of size effects.
The estimate results of INV, EARN, DEBT, and FCASH In model (7) are entirely
consistent with the findings for the full sample in Table 3.2, as well as those for other
models in Table 3.3. The coefficient estimates on INV and DEBT are insignificant. The
EARN coefficient estimate is positively large and has highly significant t-statistics,
whereas the FCASH coefficient estimate is negative and significant.
3. Vector Autorsgrotsion Results
The assumption of VAR is that the regression variables are stationary over time.
Since statistical inference based on non stationary variables results in incorrect
conclusions, it is imperative to scrutinize the stationarity of the variables. The tests of
stationarity applied here are the unit root tests and cointegration tests.
In Panel A of Table 3 .4 , unit root tests are conducted by using the augmented
Dickey Fuller (ADR statistic. The existence of a unit root implies non stationarity.
Thus, the null hypothesis of a single unit root or H*: p - 0 is tested against the
alternative or H^: p < 0 that the variable is stationary. Since the t statistic does not
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 45
m o m o w in » m mV o M #4 M
% o tA ri t m o in f# CD « pk N V m in #4 J"S O n (M
« % n S s o IS ; M M s ! ? « i! «o « y) a o K- iî :s ?
U # m Q m H** s 0 I ? V 1 1 î ?
t I - s Ê
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 46
U
I ?
?
>1 ÎS u i % s
n # # ? u * # — k ? 5 0 #4 W O 01 U 4» *» Ir t *4 V^ 4f: g y o u
9 ï 1; • : i - 4i 41 0 l U > n t
IL•' î-..** n # m I l 15 : 5 £“ . 41 o - >1 * * S o6 O # # 5 0 4» *4 # 0 O 0 4» # M * * U I 4» 4» !Hs I I 9 % %« *4 w IM #_ m»*» e «U c «4 e -4 «î i I /
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 47
have Student's t-distribution under the null hypothesis of non stationarity, critical values
for the t statistic on p, Z(t,), are simulated by Fuller (1976).
The single unit root test in Table 3.4 with the exception of DEBT does not reiect
the null hypothesis of a unit root at the 10% significance*. For the two-unit root test,
the procedure is repeated for first difference data. The null hypothesis of a unit root in
the differenced variables is rejected at the 10% level for all variables. Overall, the
results suggest that the variables become stationary only after first differencing. This
means that the variables follow an 1(1) process.
Panel B of Table 3.4 details the cointegration tests. Cointegration means that
the difference between two variables becomes stationary when these variables move
closely over time. If the variables are cointegrated, tests for stationarity may not be
reliable. Furthermore, Engle and Granger (1987) suggest that vector autoregressions
using differenced data will be misspecified because of omitted constraints when
cointegrated variables are used. Cointegration tests are conducted by repeating unit
root tests on the cointegrating residuals. The null hypothesis of no cointegration is that
the cointegrating residuals have a unit root. As it can be seen from Panel B of Table
3.4, no ADF statistic exhibits significance at the 10% level. Therefore, the null
hypothesis that the variables are not cointegrated cannot be rejected.
Based on these results, the first-order differenced data are used in the following
VAR analysis to satisfy the stationarity condition. The variance decompositions from
vector autoregressions are presented in Table 3.5. The ordering of variables used in this
analysis is INV, DIV, EARN, FCASH, and DEBT. The results reported here appear to be
quite robust to ordering since almost the same results are obtained with several different
orderings. Confidence intervals of 99% for the point estimates are computed by a
'It is customary to refer to the 10% critical values in stationarity tests.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 48
o • S d et m I s e *4 s"2 S" « et d d d d d d
O 9* e» o» o et in e* et O et e M lA W) a 10 O m et #4 S" et « «0 « «0 O d d d d o o o o O et O
et n r* • s m 0% OD O tf) et et 0» # M m e» m « O et et s: i = #4 <0 e» d d m d M o m O« o # O lA O
# « « « 0» o o o o d Ml r» #4 # e* Ot Ml » m O et m M N o O o M si « m r* m et m «0 m m n î Ê: I « m g e* «A r* o 10 Ml 10 S ï « o # O « «0 e* V e* m e* et O « I •o d A r* %0 e* r* Ml in # 44 «0 m ♦ w r* et o « O m g : in M V m m V m g et o r* e* ? a I et O m o e* o e* O
II iil l î
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. ::o a> ■o I 8 Û. Tabl# 3.5 (cont'd)
Panal Ci Relative percentage of earning, (BARM) forecaat variance h-quarter ahead "O explained by each ahock to CD
QUARTIR ihi DIV UU BARMDBBT FCASH % o =J 2 15.7961 2.7635 45.9658 18.4599 17.0145 o <0» 43.2) (0, 22.3) (18.1, 69.7)* (0, 35.1) (0, 35.7) S 4 20.5698 6.2978 38.9069 17.8462 16.3790 CD (G| 50.4) (0, 36.0) (9.9; 60.5)* (0, 33.4) (0, 34.5) 8 T3 • 22.9841 12.4979 34.9319 16.6153 12.9705 (0| 52.3) (0, 46.0) (6.8 , 52.9)* (0, 29.1) (0, 27.6) C5- zr o 12 28.1824 11.3649 34.1938 14.9142 11.3445 (1.2, 56.7)* (0, 46.7) (5.8, 51.1)* (0, 26.9) (0, 25.4) 3 nCD Panel Oi Relative percentage of debt (DEBT) forecaat "n c explained by each ahock to 3. 3" CD QS2MITEB. ihi BIV 1ÜÏ BARM DBBTFCASH
cB 36.3906 4.5996 8.6597 47.9804 2.3694 ■D 2 3 (5.0, 65.9)* (0, 25.3) (0, 25.9) (16.3, 68.5)" (0, 13.5) Q. C 4 55.9334 1.9179 20.2229 20.6968 1.2288 a (27.7, 76.4)* (0, 18.2) (0, 39.4) (4.0, 33.8)* (0, 9.3) O 3 "O 8 46.4060 7.1753 29.5955 13.0490 3.7739 O (15.7, 69.4)* (0, 36.6) (1.6, 50.2)* (0.3, 24.0)* (0, 13.9)
12 38.1480 12.5902 30.4431 9.8142 5.0042 CD (7.0, 65.3)* (0, 55.6) (0, 53.1) (0, 20.4) (0, 16.4) Q.
■D CD
C/) C/) (O 73 CD ■D 3 Q. C g Table 3.5 (cont'd) Q.
Panai li Ralativa parcantaga of fraa caah flow (PCASH) forecaat vari nca h-quartar ahead ■D aaplainad by aach ahock to CD OUARTIR PCASH (/) ihl BIX IBX BMWDBBT (O o" 2 16.3005 4.0063 1.2025 4.0588 74.4317 3 (0| 13.5) (38.4; 94.7)" 0 <0| 44.2) (0| 23.8) (0| 13.7) S 4 18.1732 4.6180 2.6570 6.6930 67.2585 CD (0| 48.7) (0| 27.8) (0} 16.9) (0; 17.0) (29.0; 82.3)" 8 "O a 11.1667 14.6336 5.2756 6.4788 62.4451 (0| 38.4) (0, 47.5) (0| 21.0)(0| 16.8) (21.0; 74.8)" C5- 3" 12 8.9709 20.9885 6.6926 6.8339 56.5139 1 (0| 38.9) (0> 58.9) (0| 25.0) (0| 17.5) (11.7; 71.5)" 3 nCD "n c *Tha 99% confidence Intervale are preaented 3. 3- ^Significant at tha 0.01 laval. CD
■DCD 3 a c a O 3 ■D 3
CD Q.
■D CD
C/) CJl (fi O 51
Monte Carlo simulation'^ with 1,000 replications. This confidence interval is bounded
by 0% and 100%. If the confidence interval includes 0. the null hypothesis of no
impact cannot be rejected.
Point estimates in Table 3.5 measure the degree of exogeneity for a given
variable in the sense that the h quarter ahead forecast error variance in a variable is
allocated to each source, and that a strictly exogenous variable explains all its forecast
error variance. If the forecast error variance in a variable is largely due to unexpected
shocks to itself in the last h quarters, then the implication is that there are no strong
interactions with other variables.
Panel A of Table 3.5 provides variance decompositions for DIV at the horizons of
2, 4, 8, and 12 quarters. At a horizon of two quarters, 96.15% of the variance in DIV
is explained by its own disturbances, whereas only 1.10% of the variance in DIV is
accounted for by INV. Other variables, including EARN, DEBT, and FCASH, also have
little impact on DIV. At the 12 quarter horizon, DIV explains 71.29% of its own
variance. However, INV explains only 4.87% of the variance in DIV, while EARN, DEBT,
and FCASH account for 12.10%, 5.61%, and 6.11% of the variance in DIV,
respectively. This suggests that nearly all variance in DIV is attributable to the shocks
to DIV itself, and that DIV has no dynamic interaction with other variables.
Variance decompositions for INV are shown in Panel B of Table 3.5. At the two-
quarter horizon, 87.20% of the variance in INV is attributed to its own shocks. The
percentage of the variance in INV explained by other variables ranges from 0.65% to
6.74%. At the 12-quarter horizon, the percentage of INV variance due to its own
shocks is a still dominating 75.31 %. DIV explains only 7.79% of the variance in INV,
which is insignificant at the 1 % level. The contributions of EARN, DEBT, and FCASH to
the variance in INV are 5.11%, 2.02%, and 9.76% respectively, insignificant at the
'This method constructs the posterior distributions of the estimate.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 52
0.01 level. These results imply that the variance in INV is fully explained by its own
shocks and thus INV appears to be exogenous.
Panel C of Table 3.5 presents variance decompositions for EARN. At the two-
quarter horizon, less than half the variance in EARN is accounted for by its own shocks.
At this short horizon, INV explains only 2.76% of the variance in EARN, while DIV,
DEBT, and FCASH account for 15.79%, 18.45%, and 17.01%, respectively. At a
longer horizon of 12 quarters, INV gradually becomes more important, explaining
11.36% of the variance in EARN. DIV accounts for 28.18% of the variance in EARN,
which is significant at the 1 % level. This would suggest that dividends explain future
earnings. DEBT and FCASH explains 14.91% and 11.34% of the variance in EARN,
respectively. EARN has 34.19% of its variance accounted for by its own shocks,
suggesting strong interactions among the variables.
Variance decompositions for DEBT are presented in Panel D of Table 3.5. An
interesting feature at the two-quarter horizon is that DIV explains 36.39% of the
variance in DEBT, which is significant. Of the variance in DEBT, 47.98% is attributable
to its own shocks. EARN explains only 8.65% of the variance in DEBT, while the
contributions by INV and FCASH are 4.59% and 2.36%, respectively. At the 12-quarter
horizon, the proportion of the variance in DEBT explained by DIV is 38.14%, which is
still significant. However, the proportion of the variance in DEBT attributable to its own
shocks decreases to 9.81 %, which is not significant. EARN contributes 30.44% to the
variance in DEBT, while INV and FCASH are responsible for 12.59% and 5.00%,
respectively. This result suggests that DEBT is set as a residual after all corporate
decisions are made. In particular, this implies that since DIV has a significant effect on
DEBT, firms borrow in order to finance dividend payments.
Variance decompositions for FCASH are also portrayed in Panel E of Table 3.5.
At the horizon of two quarters, DIV explains 16.30% of the variance in FCASH, implying
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 53
»
o o m o c!I! o
O (A m# «0 . 9* « #! I «Â o
O - «0 (A SS s : g «0 g : ; s SS e o m o o o S o S o -
• r» d o 0 m o M 0 • o • o 0 O • 0 . 0 V « m M r * d 0 0 0 0 m m 0 d • lA d 0 m 0 0 0 0 M m M m M M 0 0 m m d 0 0 0 8 . m o m e 0 O 0 o O o 0 O 0 O 0 o e
« « • 4 4 0 « 4 M 0 d 0 0 o o o 1 o o 0 0 O d lA M 0 O tn M m 0 0 0 m 0 0 0 m 0 W 0 0 o 0 «4 0 0 m V •» 0 «» O 0 m «% n — I 0 0 0 m d 0 0 0 m 0 d d O 0 (A « M 0 0 d 0 d d d 0 0 0 0 m %0 0 0 0 m r * d 0 d 0 0 d m d m 5
5- i n 4i I i I
m # S I S.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 54
that frea cash flows of the firm may be restricted by dividend payments. Of the
variance in FCASH, 74.43% is accounted for by its own shocks. The proportion of the
variance in FCASH explained by INV, EARN, and DEBT ranges between 1.20% and
4.05%. At a longer horizon of 12 quarters, the contribution of DIV to the variance in
FCASH decreases to 8.97%. INV explains 20.98% of the FCASH variance and becomes
more important, while EARN and DEBT account for 6.69% and 6.83%, respectively.
The proportion of the FCASH variance explained by its own shocks is 56.51%, more
than half the variance in FCASH.
In order to test for sensitivity of the VAR results to samples, two subsamples of
100 firms and 300 firms are selected randomly. Variance decompositions are calculated
for each subsamples. This procedure is also helpful in mitigating potential selection bias
due to sample screening. Table 3.6 reports the results of sample size sensitivity
analysis. Since tiie focus of this study is on DIV, only dividend variance decompositions
are presented.
Panel A of Table 3.6 shows dividend variance decompositions for a randomly
selected sample of 100 firms. At all horizons, the proportion of the variance in DIV
explained by its own shocks is more than 70%, while other variables explain less than
8% of the variance in DIV. These figures are consistent with the findings in Panel A of
Table 3.5. The results shown in Panel B of Table 3.6, in which dividend variance
decompositions are calculated for the sample of 300 firms, are also consistent with
those in Panel A of Table 3.5. At all horizons, DIV accounts for most of its forecast
variance. Overall, this suggests that the variance decomposition results reported in
Table 3.5 appear quite robust.
Figure 3.1 plots the impulse responses of DIV to each one standard deviation
shock to DIV, INV, EARN, DEBT, and FCASH. The 99% confidence intervals are also
plotted around the impulse response line. If this confidence interval band covers the
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. ■o
cI 8 Û.
■O CD DIVIDEND RESPONSE DIVIDEND RESPONSE DIVIDEND RESPONSE TO A ONE SID. SHOCK IN DIVIDENDS TO A ONE STD. SHOCK IN INVESTMO^TS TO A ONE SID. SHOCK IN EARNINGS C/) (II m (II (g O 3 B 0 U L S 1
U as
CD ao ao 8 5 c5' S i 3 -m -• -1 CD ■10 C -11 -IS p. -zs -25 ■OCD -9 0 10 u » 12 » o IC H-qUMCm AHIAO H-QUAKin AKIAD H-QMAKin A«AO a O 3 QBBHF fadrafe B d tr a ti ■O eeeooik 13^ Braid eeewmvsfrn eeesoM u|ip« Bond O
&
o c Figure 3.11 Impulee Response Functions of Dividends % ay ay Ç2C/) o' 3 CD "O O Q. C g Q.
■O CD DIVIDEND RESPONSE DIVIDEND RESPONSE TO A ONE STD. SHOCK IN DEBTS TO A ONE SID. SHOCK IN FREE CASH WC/) W o" 3 0 3 2$ as CD ao 8 TD '< 10 (O'3" 1 3 CD -• - s
3. 3" CD -a s as "OCD O Q. C 12 tt a H-qUAXmt AHEAD H - q u A n n a i b a d O 3 "O ErtSasli LLUUV ErtiM ti O ee etm vpfm naumà OOOwrnVppmBamA
CD Q.
Figure 3.1 (cont'd) "D CD cn o> (/) (/) 57
value of zero, the null hypothesis of no effects cannot be rejected. Diapram (1) shows
that a one standard deviation shock to DIV has significant short run effects on itself. At
a horizon of one quarter ahead, the effects are positive and significant. At a horizon of
two quarters ahead, negative and significant effects are found. At longer horizons, the
effects are insignificant since confidence interval bands fluctuate around the value of
zero. By contrast, diagrams (2)*(5) indicate that dividend responses to each one
standard deviation shock to INV, EARN, DEBT, snd FCASH are not significant at all
horizons. It appears that this evidence is consistent with the results using variance
decompositions.
4. Th» Laggtd-DMdênd Modê!
The results from Section D.2 suggest current relationships among dividends,
earnings, free cash flows, beta, and firm size. The analysis in Section D.3 provides
evidence of short memory in dividends. It is found that dividends are associated with
short memory of its own past, while any given shocks in all other variables under study
do not induce significant deviations in dividends. This would suggest that stronger lag
specifications might be useful in investigating determinants of dividend policy.
Accordingly, this section estimates several additional specifications of the time series
cross sectional regression model including lagged dividend variables, and tests for their
robustness. Since the early study of Lintner (1956), the empirical literature testing
lagged dividend models has been scant with the exception of Fama and Babiak 11968)
and Fama (1974). These studies also failed to explain the reason for the inclusion of
lagged dividends."
"Moreover, many econometric problems due to lagged dependent variable models or time series cross sectional data have not been effectively adjusted.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 58
Estimating lagged dependent variables in the regression causes many
econometric problems. If the disturbances are serially correlated in the lagged
dependent variable models, OLS estimates are inconsistent. Even if the disturbances are
"well-behaved", OLS estimation produces biased coefficients in finite samples. In order
to adjust for these problems in the context of time-series cross-sectional regressions, the
method of instrumental variables is suggested. The lagged-dividend variable is first
regressed on ali other explanatory variables. The resulting predicted values are then
used in an error components model to obtain time-series cross-sectional regression
estimates.
Table 3.7 reports time-series cross-sectional regression results using lagged
dividend variables. Model (1) is Lintner's (19561 model in which current dividends are
determined by the previous dividends and current earnings. The coefficient estimates on
DIV., and EARN are positive and significant. The adjusted R' is 51.89% and the F-test
reiects the nuli hypothesis at the level less than 0.01. Model (2) includes additional
lagged variables of dividends in the model of EARN and FCASH, which turned out
significant in Table 3.2. As predicted, the coefficient estimates are positive-significant
for EARN and negative-significant for FCASH. With the introduction of more lagged-
dividend variables, the first, second, and fourth lags show significant estimates and the
adjusted R' increases to 65.74%. This implies some persistence of dividend poiicy since
current dividends seem to be affected by the dividends paid one quarter, two quarters,
or four quarters before.
Similarly, models (3) and (4) introduce BETA and size dummy variables which
proved significant earlier. The coefficient estimate of BETA is negative-significant at the
0.01 level, whereas the coefficient estimates on the two largest size variables are
positive-significant at the 0.01 level. The positive impacts of the first, second, and
fourth lagged-dividend variables are the same as in model (2). In model (51, all variables
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. CD ■D O Q. C g Q. Table 3.7t Tt— Parle» Croaa-Sectlonal Regreaalone with Lagged Dependent Variable#
Model (1)1 DlVit - Tg * * ^5 ■D 2)1 CD Nodal ( DIV,'It »0 ♦ ♦ *2 0:V-2,it * »3 “IV-3.lt ♦ 4 “IV-4,it. ♦ Ts EANNit ♦ Tg rCASH^g ♦ “ It C/) W Nodal (3)1 o" “:Vit »0 ♦ ♦ »2 “IV-2.lt ♦ »3 “IV-a.lt ♦ % “*V-4. It: ♦ »S ««"it 3 ♦ BCTA^t * "It O Nodal (4)1 DIV,» ♦ Tg EANNit * “ It
8 Nodal (5)1 "O “ :V it - *0 + *1 “ : v _ i,it ♦ »2 “ :v _ 2 .it ♦ »3 “ :v _ 3 .it ♦ 4 “*v _ 4 .it♦ Tg BARNit ♦ Tft ♦ 1? — TAlt ♦ »J47 »J.lt i Indapandant Coafflclant (1) (2) (3) (4) (5) 3 Variable Batlmata CD INTBRCBPT *0 13.4902 8.9897 22.7459 1.7407 9.5618 (9.538)* (7.291)* (8.148)* (0.738) (2.847)* 3. 3" CD DIV_i *1 0.3829 0.2149 0.2116 0.2043 0.2029 (34.277)* (12.255)* (12.069)* (11.690)* (11.606)* CD ■D DlV.j 0.1676 O *2 0.1692 0.1602 0.1597 Q. (7.349)* (7.288)* (6.988)* (6.966)* C a -0.0361 -0.0375 O 01V_3 *3 -0.0392 -0.0398 3 (-1.521) (-1.580) (-1.656) (-1.684) ■D O DIV_4 *4 0.2642 0.2607 0.2510 0.2495 (14.982)* (14.790)* (14.273)* (14.191)* CD BABM *5 0.0293 0.0257 0.0257 0.0238 0.0240 Q. (9.684)* (8.252)* (8.279)* (7.693)* (7.730)* FCASH *6 -0.0054 -0.0055 -0.0050 -0.0051 (-7.921)* (-8.006)* (-7.360)* (-7.425)* ■D CD BETA *7 -16.0414 -9.8201 (-5.491)* (-3.272)* (/) Ol (/) Indapandant Coafflclant ( 1) ( 2 | (3) (S) "D Ml CD Varlabla datInata C/) -2.7386 -0.9233 W (-0.852) (-0.282) o" 3 0 -2.3470 -0.5164 (-0.732) (-0.158) 5 CD '10 8.5426 7.9051 8 (2.648)* (2.444)' ■D 36.6389 36.4119 CQ' '11 2 (11.2381* (11.159)' 1 3 AdJ. M 0.5109 0.6574 0.6580 0.6604 0.6610 CD Prob. of r < 0.0001 < 0.0001 < 0.0001 < 0.0001 < 0.0001 3. 3" CD 3Tha aima duany varlablaa, Sja, ara daflnad an followai Sj - 1 If SIZE < $161.21 million, Sj > O otharwlaa; ■DCD S; - 1 If $161.21 million s SIEE < $687.74 million. S; - O otharwlaa; S 3 • 1 If $687.74 million S SIEE < O $2,004.02 million, S 3 - 0 otharwlaa; S* - 1 If $2,004.02 million S SIEE < $5,265.82 million. S* - 0 Q. otharwlaa; Sj • 1 If $5,265.82 million S SIEE, Sj • 0 otharwlaa. O ^Por tha purpoaa of aatlmatlon, ona dummy varlabla la leat. Tha dummy varlabla not rapraaantad In tha tabla 3 la S 3 for SIEE varlabla. ■D O ^t-atatlatlca ara glvan In paranthaaaa. "significant at tha 0.01 laval. & ^significant at tha 0.05 laval. O c ■o CD % O O 3 61 used in models (1)-(4) are added. The estimate results are essentially the same as in other models, indicating the insensitivity of parameter estimates. The adjusted R' is 66.10%. The findings in Table 3.7 have several interesting implications. First, unlike Lintner's argument, more lags in dividends are important determinants of dividend policy. With the addition of more lags reflecting the past memory of dividends, the explanatory power of the model increases considerably. Second, as shown in Section 0.2, current earnings have positive impacts on dividends, while free cash flows and beta have negative effects. Also the firm size appears to play an important role in dividend policy. Since the sensitivity of the estimates to alternative model specifications has been established in previous sections, conducted in Table 3.8 are robustness analyses of the time-series cross-sectional model (7) in Table 3.7 for randomly selected subsamples and subperiods. In order to test for robustness of the model, the full sample is first split into three subsamples: randomly selected samples of 100 firms, 200 firms, and 300 firms. Then, the time-series cross-sectional model (7) in Table 3.7 is estimated on each subsample. As shown in Table 3.8, the sets of coefficient estimates on DIV.,, DIV.*, EARN, S*, and S, are the same as the findings in Table 3.7. However, the coefficient estimates on DIV., and DIV., show mixed results for each subsample. The coefficient estimate of BETA is insignificant in the subsample of 100 firms, whereas it is significant in the subsamples of 200 firms and 300 firms. To test for coefficient stability over time, the time-series and cross-sectional model is also estimated for two subperiods; 1979:11-1984:1V and 1985:1-1990:IV. The coefficient estimates on explanatory variables for the two subperiods shown in Table 3.8 are quite consistent with those in Table 3.7, except for the third lagged dividend variable during the subperiod 1979:11-1984:1V. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 62 m # ## lA O « P* f t m 4 2 2 « O m 0 O 0 « « 0 « W M N A T S 2 • • O « 0 o « 0 0 0 P# 0 0 « W f** M W 0» M o o • « • N O O 0 O M • o M M o • o • • 0 » fC o » O 0 O 1 • r* • 0 o • Ô O O 1 M P# 1 1 4 4 4 4 ♦ 4 4 f * O 0 « 0 0 O II PC m 0 M 0 0 0 0 M 0 0 PC o m 8 : 0 O r* p* O 0 M e 0 0 »• ♦ «0 M M # M 0 O P» « P» 0 • 0 . m PC M o • o • ♦ — ;; o e —’ O 1 • M • « t m m N o ^ o ^ ( % T • I 4 4 O « A 4 4 4 « T m 0 «CO PC 0 P» M 0 PC « #4 0 #c Pi £ 0 0 0 « 0 ^ 0 P» 0 « m fi 5 « 0 0 0 « • m « O 0 0 Pi p o Pi O O • O • 3 2 • M • « • 0 • 0 PC o ^ O 1 d O -» ? ? 1 T 4 4 4 4 4 —* A _ 0 PC « m Cr PC 2 P» 0 0 O m 0 0 « 0 M m ^ 0 0 0 M M PC O « 0 2PC • « 0 « • 0 • PC • * »• 0 Pi O • • P» 0 PC 3 * O PC • M • o O 1 • m II f* o %» O d Z O i l 7 > Q I 4 4 4 4 4 0 m 0 « p* m 0 PC « p* 0» m 0 « 0 « P* PC P» m m 0 o m ID m « « o n « m PC 0 m m r* m o • M , O • P • m • 0 • « m M o O o O <-4 O 1 ? 7 1 > Q g I ii > u O N >■ > 2 I O O & Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. CD ■D O Q. C g Table 3.8 |cont*d) Q. Randomly a.lMtad Sampl.a Total Sampl.a ■D Iadap.nd.nt Co.fflci.nt (79iII-90iIV) CD Varlobl. I.tiaat. 446 firms 446 firm. 100 f l n w 200 firm. 300 firm. (79iII-84iIV) (85:1-90:IV) WC/) o" 1 *8 -1.3309 -0.5976 -1.0523 -0.4959 -1.0510 3 (-0.1431 (-0.155) (-0.244) (-0.185) (-0.251) 0 3 ♦9 -2.0229 0.1095 -0.7056 -0.1524 -0.6896 CD (-0.272) (0.028) (-0.171) (-0.057) (-0.165) 8 ■D *4 *10 2.2909 6.6538 8.7361 6.9132 8.0647 (0.2841 (1.757) (2.001)** (2.606)* (1.941) CQ'3" »5 *11 36.9703 22.5234 29.3538 34.9107 36.4564 (5.447)* (6.961)* (13.106)* (8.576)* 1 (5.032)* 3 CD Adj. 0.7427 0.4990 0.5754 0.6664 0.5530 "n c 3. Prob. of F < 0.0001 < 0.0001 < 0.0001 < 0.0001 < 0.0001 3" CD CD ■D ^the aime diumy variable#, Sj#, are defined ae followes « 1 if S U E < $161.21 million. S* - 0 otherwiee; O S, " 1 if $161.21 million d SISE < $687.74 million, Sg " 0 otherwiee; S 3 • 1 if $687.74 million S SISE < Q. C $2,004.02 million. S 3 - 0 otherwiee; S, - 1 if $2,004.02 million & SISE < $5,265.82 million. S* - 0 a otherwiee; Sg - 1 if $5,265.82 million S SISE, Sg » O otherwiee. O 3 ^for the purpose of estimation, one dummy variable is lost. The dummy variable not represented in the tabl# ■D is 83 for SISE variable. O ^t-etatietice are given in parentheses. CD Q. "significant at the 0.01 level. "significant at the 0.05 level. "O CD (/) (/) S 64 Therefore, in spite of small differences, all coefficients of the time series cross- sectional regression in Table 3.8 display consistent signs and significances of estimates in all subsamples and subperiods. This confirms that the regression model and its coefficient estimates in this study are quite robust. 5 . Intarprêtaddn o f RtsultM The empirical results of this study are as follows: (1) contemporaneously, dividends are not affected by investments and long-term debts; (2) as in Linter (1956), dividends are positively predicted by current earnings; (3) dividends are negatively related to current free cash flows; (4) dynamically, dividends are not influenced by any other past variables except for lagged dividends;'' (5) dividends are negatively related to beta and positively related to firm size; and (6) there are no significant industry effects. The main results of this study concern the relationship between a firm's investments and financing decisions. According to the Miller and Modigliani (MM) proposition, investment decisions are independent of financing decisions such as new debts or dividends, since in perfect capital markets the value of a firm is not affected by financing but by investment. Dhrymes and Kurz (1964, 1967) claim that the complete separation of investments and financing decisions is implausible because in imperfect capital markets financing should be considered as part of the firm's investment decisions. The empirical literature associated with the MM proposition has shown mixed results. Dhrymes and Kurz (1967) and McCabe (1979) found that the firm's investment decision is linked to its financing decision. Higgins (1972), Fama (1974), and Smirlock "Moreover, investments and free cash flows are not dynamically affected by any other past variables. But, long-term debts are influenced by the past dividends and earnings. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 65 and Marshall 119831 documented no interdependence between investments and dividends. It is argued here that these tests seem to be misspecified or inappropriate because they have omitted important explanatory variables and/or lagged variables. The dynamic analysis in this study provides strong results. The MM proposition indicates that investments should not be influenced by dividends and debts. A stronger formulation of this proposition is that investment and financing variables also should not affect each other dynamically. For instance, current investment decisions should be determined independently of the previous debt and dividend decisions. Based on the dynamic investigation as wall as the contemporaneous tests, the results in this study support the MM proposition. The results of this study also show that dividends and new debts are contemporaneously independent, but that current dividend decisions affect future debt decisions. This is an indication that new debt decisions are subordinated to dividend decision. On the other hand, the results of this study confirm that, as Untner (1956) argues, current earnings and the first lagged dividends serve to predict current dividends. Furthermore, this study shows a stronger persistence in dividend policy. It is found that current dividend decision is affected by the dividends paid two quarters or four quarters before, in addition to the immediately previous dividends. The test results of free cash flows involve an intriguing interpretation. Although the contemporaneous relationship between dividends and free cash flows is negative, dividends are dynamically unaffected by free cash flows. The agency theory explanation of dividend payments is that the firm should pay dividends to reduce agency costs associated with excess cash flows [see Jensen (1986)]'*. In contrast, the pecking order theory states that dividend payments would be negatively influenced by free cash ''Furthermore, Easterbrook (1984) argues that paying dividends can reduce agency costs such as monitoring cost and risk aversion of managers, by keeping firms in the market for consistent monitoring. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 66 flows. Myers and Majluf (1984) claim that "firms can build up financial slack by restricting dividends when investment requirements are modest" [ibid., p. 220). Since in this study the current free cash flows are generated from the current accounting numbers including current dividends, an interpretation of this result should resort to the stronger implication whether current dividends are affected by the previous free cash flows (see Appendix A). Thus, the dynamic results provided here are not consistent with agency theory argument or pecking order theory explanation. A final interpretation concerns beta, size, and industry effects The negative correlation between dividends and beta indicates that firms with higher betas pay lower dividends to reduce the costs of external financing, since higher betas are associated with higher leverage [Rozeff (1982)1, or that firms distribute dividends to affect the systematic risk of their stocks (Dyl and Hoffmeister (1986)1. Since the variables related to firm size are adjusted explicitly, it should be noted that the differences in dividend policy between small firms and large firms are substantial. The evidence in this study seems to indicate that the larger firms would choose to increase dividend payments rather than to retain earnings for their faster growth. The tests in this study find no evidence for the systematic relationship between dividend policy and industrial classification, inconsistent with Michel (1979).'* This suggests that industrial variations in dividend policy might be attributable to firm size effects. E. Chapter Summary and Conclusions The purpose of this study was to analyze empirically a wide scope of dividend determinants including investments, earnings, debts, free cash flows, firm size, beta, and industry classification. '*Higgins (1972) also found no pervasive industry effects, whereas Dhrymes and Kurz (1967) documented significant inter industry differences. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 67 First, time-series cross-sectional regressions using an error components model were estimated to test for the contemporaneous relationships among these potential dividend determinants. It was found that a firm's dividend payments are significantly related to earnings, free cash flows, beta, and firm sizes. Perhaps the most important finding is that dividends are not influenced by investments and debts, consistent with the irrelevance proposition of Miller and Modigliani (1961). It was also shown that industry effects might not exist or might be only the manifestation of firm size effects. Second, in order to examine the dynamic relationships among these variables, the VAR analyses were conducted. The variance decompositions and impulse response analyses consistently demonstrated that dividends have significant short-run effects on itself, whereas other variables have no dynamic effects on dividends. It also appears that dividends have short memory of their own past and that other variables have no dynamic interactions with dividends. Finally, on the basis of both time-series cross-sectional regression results and vector autoregression analysis, a lagged-dividend model was developed and tested for robustness. The evidence suggests that dividend payments are heavily influenced by a series of past dividend payments, as well as contemporaneous factors such as earnings and free cash flows. Although this lagged-dividend model may be exploratory, it sheds some light on the dividend puzzle. One fruitful avenue for future study may be to apply a structural vector autoregression which simultaneously incorporates both contemporaneous variables and dynamic factors. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Chapter 4. DIVIDENDS, TAXES, AND PORTFOLIO CHOICES: The Effects of the Tax Reform Acts of 1984 and 1986 A. Introduction There have been two hypotheses relating to the ex dividend anomalies: (1) the tax-effect hypothesis; and (2) the short-term trading hypothesis. The tax-effect hypothesis predicts that positive excess returns are observable on the ex-dividend day because of the dividend tax compensation. This tax-effect hypothesis is first put forth by Elton and Gruber (1970). Later, Booth and Johnston (1984), Elton, Gruber and Rentzler (1984), Barclay (1987), and Michaely (1991) follow this reasoning. These authors find that, on ex-dividend day, stock prices fall by less than the dividend per share. These authors ascribe this phenomenon to the tax premium effect in which investors receive a compensating premium for the dividend tax penalization. This tax- effect hypothesis is questioned by Eades, Hess, and Kim (1984) and Grinblatt, Masulis, and Titman (1984), who find abnormal ex-dividend returns in non-taxable distributions. The short-term trading hypothesis predicts that positive excess volumes are observable around the ex-dividend day because there exist tax-induced arbitrage opportunities around the ex-dividend day for short-term traders, given transaction costs. Miller and Scholes (1982), Kalay (1982), and Karpoff and Walkling (1988, 1991) argue that short-term traders have incentives to eliminate arbitrage profits up to their marginal transaction costs around the ex-dividend day. Lakonishok and Vermaelen (1983) and Grammatikos (1989) investigate major tax reform options to support the short-term trading hypothesis. Lakonishok and Vermaelen (1986) and Fedenia and Grammatikos (1991) also examine excess trading volume around ex-dividend day and conclude that short-term traders are responsible for abnormal volume behavior, or that portfolio rebalancing takes place. 68 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 69 The motivation for this essay is that empirical tests of these two hypotheses in the context of the 1984 and 1986 tax reforms show mixed evidences (e.g., Michaely (1991) vs. Robin (1991)1', and that these two hypotheses seem to be mutually exclusive and neither is sufficient to fully explain the ex dividend anomalies. The purposes of this essay are; (1) to test the tax-effect hypothesis and the short-term trading hypothesis in the context of the 1984 and 1986 tax reforms; and 12) to propose an alternative approach to the ex-dividend anomalies that fully incorporates both the tax-effect hypothesis and the short-term trading hypothesis. In this essay, a portfolio approach is proposed that predicts some observable differences between cum and ex-dividend efficient portfolios based on a portfolio selection model to investigate the impact of the tax reform acts on portfolio choice. Portfolio theories allowing for personal taxes are divided into two camps. Brennan (1970) and Elton and Gruber (1978) derive a post-tax CAPM model, and Long (1977) posits a positive portfolio theory on an after-tax basis. Talmor (1985) and (amoureux (1990), on the other hand, take a normative portfolio approach, which allows for the differential taxation of dividends and capital gains. The portfolio model used in this essay follows that of Long (1977). Long (1977) focuses on the implausibility of the efficiency equivalence hypothesis between before- and after-tax portfolios, noting that (1) "the after-tax efficiency gains to shifts from portfolios that are before tax efficient to portfolios that are after-tax efficient" (Long (1977), p. 251; (2) "the greater the risk level of a before-tax efficient portfolio, the less efficient it is on any given after-tax basis" {ibid., p. 411; and (3) "any given before-tax efficient portfolio will be less efficient on an after-tax basis the greater the difference between an investor's marginal tax rates on dividends and capital gains"[ibid., p. 41 ]. ^Michaely (1991) teste the effect of the 1986 Tax Reform Act on ex- dividend day stock prices. He finds no support for the tax-effect hypothesis. In contrast, Robin (1991) argues the impact of the 1986 Tax Reform Act on ex-dividend day returns, which is consistent with the tax- effect hypothesis. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 70 The portfolio approach has several advantages over the other approaches. First, it fully incorporates the tax-effect since it differentiates between the tax treatments of dividends and capital gains. Second, it also encompasses the short-term trading hypothesis, according to which the firm's clientele is shifting and portfolio revisions are taking piace. Third, but not least, the portfolio approach is an equilibrium model with appropriate measures of risk and return, rather than a model of behavior. The remainder of this essay is organized as follows. Section B summarizes major changes in the Tax Reform Acts of 1984 and 1986. Section C reviews the extant models that have been advanced to explain the ex-dividend anomalies, and testable hypotheses are derived in light of the 1984 and 1986 tax reforms. Most important, this section presents the portfolio approach that is proposed to explain effectively the portfolio choices around the ex-dividend day. Section D describes the data and methodology used in the tests. Section E reports the results of the tests of the tax- effect hypothesis, the short-term trading hypothesis, and the proposed portfolio approach, and it explores the usefulness of the portfolio approach in explaining ex- dividend behavior. Section F summarizes and concludes the essay. 8. The Tax Reform Acts of 1984 and 1986 1. Th9 Têx Raform Act o f 1984 Under the Tax Reform Act of 1984, the required holding period for the dividend income tax deduction is extended from 16 to 46 days (see Panel A of Table 4.1). A taxable corporation owning stock of another corporation is allowed a deduction of 85%: gf fffg 3n,ount of any dividend received for that stock. At the same time, the corporation can claim the loss resulting from the ex-dividend decline in value of the stock as a short-term capital loss. This short-term loss can be used to offset an unrelated short-term gain otherwise taxable at the maximum corporate tax rate of 46%. ^The Tax Reform Act of 1986 decreased the dividend income tax deduction from 85% to 75%. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 71 TmW# 4.1: Sununarv of Chong## in th# T#x Roform Acts of 1984 #nd 1986 Pantl A : 1984 Tax Reform Act Topic Before 1984 TRA After 1984 TRA ( Corporation» i Holding period for 16 days 46 days 85% dividends received tax deduction PaneiB: 1986 Tax Reform Act Topic Before 1986 TRA After 1986 TRA 1987 After 1987 ( Individuais ) Tex Rate Schedule Min 11% Min 11% Min 15% Max 50% Max 38.5% Max 28% (IS bracketsi (5 bracketsi (2 brackets) Dividends Taxable at income tax rate Taxable at Taxable at (11% 50%) income tax income tax rate (11 %, rate (15%, 38.5%) 28%) Dividend Exclusion of ) 100 Repealed Repealed (Of $ 200) Capital Gains Taxable at income tax rate Taxable at Taxable at the lower rate Income tax of income tax rate (15%, rate and 28% 28%) (11%. 28%) 60% capital gains deduction Repealed Repealed * The highest effective capitai gains tax rate ■ 20% (50% x 40%) * The lowest effective capitai gains tax rate - 4.4% (11 % x40%) Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 72 Tabla 4.1 (cont'd) I Corporationa I Income Tax Rate Schedule Min 15% Min 15% Min 15% Max 46% Max 34% Max 34% IS brackets) (3 brackets) (3 brackeu) Dividende Taxable at income tax rate Taxable at Taxable at income tax income tax rate rate 85% dividends received 80% dividend 70% dividend deduction received received deduction deduction * The highest effective 6.9% (46% X 15%) 6 .8 % (34% X 10.2% (34% dividend tax rate 20%) x30%) * The lowest effective 2.25% (15% X 15%) 3.0% 115% X 4.5% (15% X dividend tax rate 20%) 30%) Capital Gains Taxable at special corporate Taxable at Taxable at capital gains tax rate (Max income tax income tax 28%) rate (15%, rate (15%, 34%) 34%) Note : I , ) denotes (Minimum rate. Maximum rate). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 73 If the corporation is financing an Investment in stock with borrowed funds, interest on the indebtedness is also tax deductible. These corporations are generally taxed at a lower rate on dividends than on capital gains. 2. Th» T»x R»fom Act o f 1986 The Tax Reform Act of 1986 equalizes the nominal taxation of dividends and capital gains for individuals (see Panel 8 of Table 4.1). 8efore passage of the act, capital gains were preferantially taxed compared to dividends for individuals. Individuals we;e taxable in as many as 15 brackets at rates thst varied from 11% to 50%. Dividends were taxable at these ordinary income tax rates for individuals and were subject to a maximum tax rate of 50% after deducting dividend exclusion'. Individuals cc'uld deduct 60% of net capital gain from gross income. As a result, the highest tax rate applicable to an individual's net capital gain was only 20% (the 50% maximum individual tax rate times the 40% of net capital gain). Under the 1986 Tax Reform Act, the tax schedule was reduced to two brackets of 15% and 28%, starting with the 1988 tax year. There was a blended rate schedule for 1987 with five brackets, varying from 11% to 38.5%. The dividend exclusion for individuals also was repealed. Thus, dividends were taxed at a maximum rate of 38.5% in the transitional year of 1987 and were subject to a maximum rate of 28%' after 1987. In addition, the 1986 Tax Reform Act repealed the 60% net capital gain deduction for individuals. Beginning in 1987, capital gains are taxed at a maximum rate of 28%. Thus, for 1988, there was no longer a distinction between dividends and capital gains for tax purposes for individuals. The 1986 Tax Reform Act generally reduced corporate tax rates, but did not change much the effective tax differentials between dividends and capital gains. Prior to the 1986 tax reform, corporations were taxed from a five bracket tax schedule ranging ^The exclueion amount# were $100 of dividend# received by an individual atockholder and $200 by a married couple. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 74 from 15% to 46%. Since corporations could deduct 85% of the dividends received, corporate income from dividends was taxed at a maximum effective rate of 6.9% (the 15% of dividends received times the maximum corporate tax rate of 46%). Capital gains were also subject to a preferential tax rate of 28% under prior law. Under the 1986 Tax Reform Act, the old-five bracket corporate tax schedule was reduced to a three-bracket schedule varying from 15% to 34%, effective July 1987. Under the new act, the 85% dividends received deduction is lowered to 70%. Thus, the highest effective tax rate of corporate dividends received became 10.2% (the 30% of dividends received times the maximum corporate tax rate of 34%). Also, prior law's preferential 28% tax rate for corporate capital gains is repealed. Beginning in July 1987, corporate capital gains were taxed as ordinary income and were subject to a maximum tax rate of 34%. C. Models and Testable Hypotheses 7. Th* Tax-£ff$et Modxt The tax-effect model was originally proposed by Elton and Gruber (1970). Later, Kalay (1982), Booth and Johnston (1984), Barclay (1987), and Michaely (1991) pursue this logic. Elton and Gruber (1970) suggest that the ex-dividend price of a stock should be related to the tax rates of marginal stockholders. Because dividends and capital gains are taxed at different rates, they argue that the equilibrium market price on the ex- dividend day should reflect the value of dividends vis-a-vis capital gains to the marginal stockholders. Elton and Gruber (1970) assume long-term traders who have already decided to sell a stock around the ex-dividend day. These investors' only concern is whether to sell before or after the ex-dividend day. In case of no transaction costs, the long-term trader would be indifferent between selling the stock cum- or ex-dividend day as long as the returns are the same. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 75 - E{P„] -r,(E[P„] -Phy) *D{1-Ta) . (« 1) where Pan is the cum dividend stock price, P* is the ex-dividend stock price, Ptuy is the stock purchase price, T, is the marginal tax rate on capital gains, and Ty is the marginal tax rate on dividends. Equation (4.1) means that the cum-dividend return would be equal to the ex- dividend return. This can be rearranged as (4*2) D 1-T, Equation (4.2) is the market pricing model for long-term traders on the ex- dividend day. In equilibrium, market ex-dividend prices will be determined such that a stockholder with different tax rates on dividends and capital gains will be indifferent as to selling the stock between before or after the ex-dividend day. Rewriting equation (4.2) as the ex-dividend rate of return yields; ^ where P., is the ex-dividend rate of return and d = is the dividend yield. Equation (4.3) implies that observable positive excess returns on ex-dividend day reflect the higher taxes imposed on dividends compared to taxes on capital gains. According to the tax-effect hypothesis, investors receive a compensating premium for the tax penalization of dividends relative to capital gains on the ex-dividend day. This means that the effective taxation of dividends and capital gains alone determines the price behaviors on the ex-dividend day. Thus, it is predicted that the higher the effective dividend tax vis-a-vis the capital gains tax, the greater the dividend tax penalization and the larger the positive excess returns on ex-dividend day. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 76 Hypothesis 1 : Prior to the 1986 tax reform act, significant positive excess returns wifi be observed on the ex-dividend day. Since the 1986 Tax Reform Act eliminated the nominal tax differentials between dividends and capital gains, the tax-effect hypothesis predicts that positive excess returns observed on the ex-dividend day will be drastically reduced to an insignificant level during the post-1986 tax reform period compared to the pre-1986 tax reform period. However, as Booth and Johnston (1984) argue, capital gains may be still treated more favorably than dividends because individual investors can maintain a tax timing option on the realization of capital gains. The effects of tax deferral of capital gains may prevent excess ex-dividend returns from being completely eliminated. Hypothesis 2; During the post-1986 tax reform period, average ex-dividend day excess returns will be significantly lower than those during the pre-1986 tax reform period, or will not be different from zero. Based on the tax-effect hypothesis, many studies document the positive correlation of the price change-to-dividend ratio with the dividend yield of the stock, suggesting dividend clientele effects. For example, Elton and Gruber (1970) report that with the exception of the first and eighth decile, the price change-to-dividend ratio increases up to one as the dividend yield rises. Since price change-to-dividend ratios less than unit would indicate low dividend yields and high implied tax rates, thay argue that low dividend yields should attract stockholders in higher tax brackets. This can be interpreted in terms of excess returns on the ex-dividend day. If dividend clientele effects exist, negative correlations between the dividend yields and ex-dividend day excess returns will be observed. Marginal stockholder's tax rates can be inferred from those ex-dividend day excess returns. Hypothesis 3: There will be negative correlation between dividend yields and ex- dividend day excess returns. This correlation will be weakened between the pre-1986 tax reform period and the post-1986 tax reform period. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 77 2. Short'Ttm Tnding Modal The short-term trading model was first studied by Miller and Scholes (1982) and Kalay (1982). Later, Lakonishok and Vermaelen (1983, 1986) and Karpoff and Walkling (1988, 1991) follow this reasoning. Miller and Scholes (1982) argue that every investor can exploit profit opportunities from short-term trading, buying cum-dividend shares and selling them ex- dividend. Such trading activities are expected to eliminate the excess returns on the ex- dividend day, but transaction costs may well prevent the excess returns from being completely eliminated. The marginal short-term traders include taxable corporations and dealers in securities. Since individual short-term traders may be constrained by the capital loss limitations and the wash-sale rules. Miller and Scholes (1982) point out that corporations eligible for the dividend income deduction or security dealers taxable at the same rates on dividends and capital gains could have incentives to make short-term trading profits. Taxable corporations are considered to have stronger incentives to make short-term trading than dealers (see Grammatikos (1989)1. Generally, the preferential tax treatment of dividends relative to capital gains may give a taxable corporation strong incentives for short-term trading around the ex- dividend day. As a result, a corporation can buy shares of stock in another corporation cum-dividend in anticipation of receiving a dividend that will be eligible for the dividend income tax deduction. The corporation may finance the cost of an investment in stock with borrowed money having tax-deductible interest. After receiving the dividend, the corporation can sell the stock ex-dividend and claim any capital loss from selling the stock at a lower ex-dividend price in order to offset other capital gains taxable at a higher tax rate. However, this dividend tax deduction is not allowed unless the corporation holds the dividend-paying stock for a minimum period. This implies that firms must expose themselves to capital loss risks in order to exploit the dividend tax deduction. The 1984 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 78 Tax Reform Act requires that. In order to qualify for the dividend tax deduction, the corporation must hold the stock for a minimum 46 days instead of a minimum 16 days, as allowed under prior law. This change increases the capital loss risk of firms that engage in short-term trading around the ex-dividend day. Since the risk exposure of short-term traders has been increased after the 1984 tax reform,* it is predicted that excess returns on the ex-dividend day for those short-term traders will increase during the post-1984 tax reform period, compared to the pre-1984 tax reform period. Hypothesis 4: Average ex-dividend day excess raturns will increase significantly during the post-1984 tax reform period compared to the pre-1984 tax reform period. Since the 1984 tax reform increased the exposure risks of short-term traders, these increased risks would be reflected in the stock prices for several short-term trading days before and after the ex-dividend day, in addition to on the ex-dividend day during the post-1984 tax reform period. Hypothesis 5; During the post-1984 tax reform period, more significant excess returns will be observed for several trading days before and after the ex-dividend day than in the pre-1984 tax reform period. In order to investigate the effects of the 1986 Tax Reform Act on short-term trading, the marginal rates of substitution of dividends and capital gains between the pre- and post-1986 tax reform period are compared. Under the 1986 Tax Reform Act, the effective corporate tax differentials between dividends and capital gains increased slightly to 23.8% (the capital gains tax rate of 34% minus the effective dividend tax rate of 10.2%| from 21.1% (the capital gains tax rate of 28% minus the effective dividend tax rate of 6.9%) during the pre-1986 tax reform period. Michaely (1991) argues that, because the marginal rate of substitution between dividends and capital gains for highly taxed investors decreased from 1.72 in 1986 to 1.36 in 1987, the corporation's incentives to engage in short-term trading is reduced. The calculation. *The rieke of ahort-term traders due to the 1984 Tax Reform Act seemed to be largely systematic (see Brown and Lunmer (1986) and Grammatikos (1989)]. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 79 however, shows that the marginal rate of substitution between dividends and capital gains during the post-1986 tax reform period is 1.29 (> (1-0.0691/(1-0.28)1, whereas the marginal rate of substitution between dividends and capital gains during the post- 1986 tax reform period is 1.36 ( - (1-0.1021/(1-0.34)). It is argued that, if one allows for transaction costs, the 1986 Tax Reform Act seems to give little incentive to firms to engage in tax-induced short-term trading (see Appendix B). 3. Th* Portfolio Mod*! The basic idea of our portfolio approach is that the ex-dividend value of stock and short-term trading behavior around ex-dividend day should reflect the trade off between risk and expected return. Long (1977) argues that 'the portfolio dividend yield choice cannot be made independently of the risk-expected return trade-off, since the dividend yield of all mean variance efficient portfolios is a linear function of their non- diversifiable risk' [Kalay (1982), p. 1059). Thus, it can be inferred that any change in the risk-expected return trade-off induced by the tax reforms should be manifested in the portfolio choices around the ex-dividend day. To show how a change in the risk-expected return trade-off is reflected in portfolio choices around the ex-dividend day, we construct efficient portfolios using the portfolio selection model posed by Markowitz (1952, 1959, 1981) as follows: Maximize subject to where i f is the transpose of the expected return vector, W is the optimal portfolio weights, V is the covariance matrix of the returns, A is the investor's risk aversion parameter, and A and b form a set of linear constraints. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 80 Efficient portfolio frontiers are computed. First, the efficient portfolio frontiers between the pre-1984 tax reform period and the post-1984 tax reform period are compared. Within each period, cum-dividend efficient frontiers and ex-dividend efficient frontiers are compared. Short-selling is not considered because it seems not to be a common practice around the ex-dividend day [see Karpoff and Walkling (1991)1. While the portfolio optimization model is usually applied in a normative context, our approach is positive. Before the 1984 tax reform, the required holding period for dividend income tax deduction was 16 days. It is predicted that cum-dividend portfolios will be more efficient than ex-dividend portfolios, because ex-dividend portfolios are more likely to be restricted from trading. This inference is supported by the arguments of Green (1980), Grundy (1985)*, and Pogue (1970)*. Hypothesis 6: Cum-dividend portfoiio efficient frontiers wiil dominete ex-dividend portfolio efficient frontiers. The 1984 Tax Reform Act extended the minimum holding period for the dividend income tax deduction to 46 days. Since trading may be restricted for longer ex-dividend days, the efficiency gap between cum-dividend efficient frontiers and ex-dividend efficient frontiers will be greater after the 1984 tax reform than before the 1984 tax reform. Hypothesis 7: The efficiency gap between cum-dividend efficient frontiers and ex- dividend efficient frontiers will be greater during the post-1984 tax reform period than during the pre-1984 tax reform period. ^Green (1980) and Grundy (1985) suggest the delay and acceleration hypothesis. Their arguments state that trading activities around the ex-dividend day will concentrate on the last cum-dividend day and on the first ex-dividend day, because investors attempt to avoid costs resulting from delaying or accelerating transactions relative to their optimum trading date for tax purposes. ^Pogue (1970) reports that portfolios restricted from trading are dominated by the unrestricted portfolios. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 81 Long (1977) shows that, with the Introduction of Income taxation, Investors will demand after tax efficient portfolios and that those portfolios may not be before tax efficient. For example, under the tax regime that raised the dividend tax rate relative to the capital gains tax. Investors are Induced to revise their portfolio holdings to reflect these differential taxes. That sort of portfolio revision will result in an efficiency gain at the new tax rates because It will Increase the expected after tax return at the new tax rates and reduce the after tax variance of the portfolio [see Long (1977)1. Accordingly, the after tax efficient portfolio will dominate any given before-tax efficient portfolio on an after tax basis. Furthermore, Long predicts that the potential after tax efficiency gains due to moving from the before-tax efficient portfolio to the dominating after-tax frontier will be smaller If the correlations between two portfolios are large. The 1986 Tax Reform Act decreased the dividend tax rate from a maximum rate of 50% to 2 8 % \ According to the Long's hypothesis. Investors are motivated to restructure their portfolio holdings to reflect the new tax rates. However, since the dividend tax rate under the 1986 Tax Reform Act has decreased. It Is expected that the expected after-tax return will decrease and that the after-tax variance will Increase at the new tax rates^ (see Long (1977), p. 391. Thus, It Is predicted that the after-1986 tax reform efficient portfolios will be dominated by the before-1986 tax reform efficient portfolios. Hypothesis 8: The efficient portfolio frontier under the pre-1986 tax reform period will dominate the efficient portfolio frontier under the post-1986 tax reform period. ^Unless the individual must pay a 5% surcharge, he is taxed at a maximum rate o f 28%. decrease in the expected return is to be anticipated in the tax- effect hypothesis since the tax premiums decrease with a drop in dividend tax rate, but its variance is not considered in the tax-effect hypothesis. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 82 D. Data and Mathodology 1. Data The sample for this study consists of taxable dividend payment events by all firms listed on the New York Stock Exchange (NYSE) and American Stock Exchange (AMEX) from January 1, 1980, to December 31, 1989. In order to examine the effects of the Tax Reform Acts of 1984 and 1986, the entire sample period is divided into three periods: Period I (1/1/80-7/17/84), Period II (7/18/84-12/31/86), and Period III (1/1/87- 12/31/89). To obtain uncontaminated samples, the following selection criteria are used for each sub-sample period. First, dividend payments whose ex-dividend date are less than 57 trading days apart from the ex-dividend date of the previous dividend payment are deleted. Second, a dividend payment is not considered unless its ex-dividend dates are at least 11 trading days away from the declaration (announcement) date of the subsequent dividend payment. Third, dividend payment events that have missing returns during the period - including the estimation period and the test period (-55 through + 10 days relative to the ex-dividend day) - are eliminated. After this screening, the sample consists of the following dividend payments and number of firms by subperiods: 21,334 dividend payments by 2,000 firms in Period I; 10,335 dividend payments by 1,643 firms in Period II; and 12,525 dividend payments by 1,776 firms in Period III. These are reported in Table 4.2. All dividend payments are taxable, and cash dividends are paid in U.S. dollars. Stock dividends and splits are excluded. Quarterly dividends, semi annual dividends, and annual dividends are identified by the CRSP distribution type codes 1232, 1242, and 1252, respectively. Reproduced witli permission of the copyright owner. Further reproduction prohibited without permission. 83 Tabla 4.2: Numbar of Sainpia Rmia and Oividand Paymant Evants from January 1980 throuQh Oaeambar 1989 Data wara obtainad from tha CRSP daily maatar tapa. AH dividand paymanu ara taxabla and cash dividands paid in U.S. dollars. CRSP distribution typa codas ara 1232 for quarterly dividends. 1242 for semi annual dividands, and 1252 for annual dividands. Period 1 Period II Period III 11/1/80 - 7/17/84) 17/18/84 - 12/31/86) (1/1/87 • 12/31/89) Numbar of Firms 2,000 1,643 1,776 Numbar of Dividand Payment Evants 21,334 10,335 12,525 Dividand Types Quarterly dividand 20,622 9,912 11,851 Sami annual 292 121 148 dividand Annual dividand 82 37 33 Others 338 265 493 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 84 2. Mtthodahgy The mean mdjueted return model* Is used to calculate ex dividend day excess returns. Trading days 55 through 11 relative to the ex dividend day are used as the estimation period. The 11 day window (-5 through +5) surrounding the ex dividend day is observed to teat the null hypothesis of zero excess returns for the ex dividend period. The excess returns'* around the ex dividend day are estimated as follows. ER,t • R,t - AR,. (4.6) where ERj, is the excess return for stock i on day t, Rh is the realized return for stock i on day t. and AR| is the arithmetic average of stock i’a returns in the estimation period. The average excess return is defined as the cross-sectional mean of excess returns. Average standardized excess returns are calculated as the cross-sectional mean of excess returns standardized by the standard deviation, which were estimated for the period from -55 through -11 days before the ex-dividend day. Average market-adjusted excess returns are computed as the average excess returns adjusted by the CRSP equally-weighted index for day t. Following Brown and Warner (1985), the t-test statistic is calculated as the ratio of the mean excess return to its standard deviation that is estimated from the time-series of mean excess returns." ^Brown and Warner (1985) report that the mean-adjusted return methodology ie ae powerful as the market model. ^^Since the analyeia of ex-dividend excess returns carries much leas risk of heteroscedaaticity than that of the price change-to- dividend ratios, exceae returns are examined in this study. The heteroacedaeticity problem in the price change-to-dividend ratios can be adjusted by the GLS estimation as fades, Hess and Kim (1984) euggeated (see Michaely (1991)]. ^^Thia procedure takes care of croaa-eectional dependence in excess returns. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 85 E. Empirical Rasuitt 1. D»$e/iptivm R»suttM o f Conslstoney Tabla 4.3 shows the characteristics of the taxable dividend-paying firms for the period January 1980 to December 1989. In order to investigate dividend clientele effects, four quartile dividend yield groups are examined: Group 1 for stocks with a dividend yield less than or equal to 0.7%; Group 2, between 0.7% and 1.1%; Group 3, between 1.1% and 1.5%: and Group 4. higher than 1.5%. Consistent with the tax-effect hypothesis, the average ex-dividend day excess return for the overall period is significantly different from zero at the 1 % level. In Panel A of Table 4.3, the average ex-dividend day excess return for the overall period is 0.1126%, with a standard deviation of 2.219%. The overall sample also has an average dividend yield of 1.09%, with a standard deviation of 0.9%. For the overall sample in Panel A of Table 4.3, the negative correlation between ex-dividend day excess returns and dividend yields is observed only in the high-yield groups. The dividend yield quartiles 3 and 4 have mean yields of 1.29% and 2.21 %, respectively, whereas they have respective ex-dividend day excess returns of 0.2371 % and 0.0357%. This indicates that there is a clientele effect present in the high dividend yields group. Similar trends are observed within each sub-period. Panel B of Table 4.3 shows the descriptive results for Period I (1/1/80-7/17/84), which serves as the control period before the Tax Reform Acts of 1984 and 1986. The average ex-dividend day excess return is 0.1376%, which is significantly different from zero at the 1% level. The sample for Period I has a dividend yield mean of 1.15%, with a standard deviation of 0.7%. During Period II (7/18/84-12/31/86), the 1984 Tax Reform Act was in effect, but the 1986 Tax Reform Act had not yet materialized. Since the 1984 Tax Reform Act extended the required holding period for dividend income tax deduction from 16 to 46 days, the risk of corporate short-term traders has been increased. It is therefore predicted that, following the increased risk, excess returns for short-term traders will Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 86 Tabto 4.3: Sampl* Dncriptiva Statistics on Ex-dividsnd Day Excasa Ratums and Dividand YWda Tha mean and tha standard deviation for ax-dividand day axcass returns and dividend yields are calculated for tha overall sampla. tha sub-period samples, and tha quartila dividend yield groups. Quartiia dividand yield groups are formed as follows: Group 1 for stocks with a dividend yield less than or equal to 0.7%, Group 2 batwoan 0.7% and 1.1%, Group 3 batwaan 1.1% and 1.5%, and Group 4 higher than 1.5%. All estimâtes of tha mean axcass return are significantly different from zero at tha 1 % lavai. The dividand yields are calculated by dividing the dividend par share by tha cum-dividend price. Tha market values of tha firm are computed by multiplying the cum-dividand price par share by tha numbar of shares outstanding. Panel A : Overall Panod (1/1/80-12/31/891 Total Dividand Yield Quartile Evants 1 2 3 4 ( lAW 1 ( High ) Numbar of 44194 15709 11312 7055 10118 Dividand Payment (100%) (35.55%) (25.60%) (15.96%) (22.89%) Evants (parcantagal Ex-dividand Day Excess Returns (%) Moan 0.1126** 0.0410 0.2032 0.2371 0.0357 Standard 2.219 2.359 2.184 2.158 2.061 deviation Dividend Yields Mean 0.0109 0.0043 0.0089 0.0129 0.0221 Standard 0.009 0.002 0.001 0.001 0.012 deviation Market Value of the Firm ( $ mil. ) Mean 1889 1640 1478 1637 2910 Standard 39885 35254 8272 11582 69644 deviation Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 87 Table 4.3 (cont'd) Panel B: Period 1 (1/1/80-7/17/84) Total Dividend Yield Quartile Events 1 2 3 4 (Low) ( High ) Number of 21334 6139 5341 4407 5447 Dividend Payment (100%) (28.78%) (25.04%) (20.88%) (25.53%) Eventa (percentage) Ex-dividand Day Exceaa Returns (%) Mean 0.1378" •0.0092 0.1389 0.2781 0.1898 Standard 2.329 2.551 2.335 2.302 2.057 deviation Dividend Yields Mean 0.0115 0.0044 0.0090 0.0129 0.0208 Standard 0.007 0.002 0.001 0.001 0.008 deviation Market Value of the Firm ( $ mil. ) Mean 921 875 855 1025 1178 Standard 11809 8793 3782 4333 20298 deviation Panel C : Period II (7/18/84-12/31/86) Total Dividend Yield Quartile Events 1 2 3 4 ( Low ) ( High ) Number of 10335 4424 3024 1428 1461 Dividend Payment (100%) (42.81%) (29.26%) (13.80%) (14.14%) Events (percentage) Ex-dividend Day Excess Returns (%) Mean 0.1518" 0.1287 0.2881 0.2130 •0.1204 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 88 Tabla 4.3 (corn'd) Standard 1.962 2.085 1.836 1.715 2.030 deviation Oividand Yialda Mean 0.0092 0.0044 0.0088 0.0127 0.0214 Standard 0.007 0.002 0.001 0.001 0 009 deviation Market Value of the Firm ( $ mil. ) Mean 2837 1937 1505 2359 8782 Standard 71192 44447 5309 23765 170961 deviation Panel D: Period III (1/1/87* 12/31/89 ) Total Dividend Yield Quartile Events 1 2 3 4 ( Low ) ( High ) Number of 12525 5146 2947 1222 3210 Dividend Payment (100%) (41.09%) (23.53%) (9.76%) (25.63%) Events (percentage) Ex-dividend Day Excess Returns (%) Mean 0.0379 0.0256 0.2329 0.1246 *0.1545 Standard 2.225 2.341 2.229 2.082 2.061 deviation Dividend Yields Mean 0.0114 0.0043 0.0088 0.0128 0.0247 Sundard 0.013 0.002 0.001 0.001 0.019 deviation Markat Value of the Firm ( $ mil. 1 Mean 2755 2536 2579 3003 3171 Standard 34623 44749 14361 6707 35515 deviation • 9 CixeeeSdaMAM H M em 1 9 L Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 89 increase. Consistent with this expectation, the average ex dividend day excess return increased from 0.1376% during control Period I (1/1/80-7/17/84) to 0.1518% during Period II (7/18/84-12/31/86). This is shown in Panel C of Table 4.3. Dividend clientele behavior is also observed in the high dividend yield quartiles of Period II, which is the same pattern as that observed during Period I. Panel D of Table 4.3 presents the results observed during Period III (1/1/87- 12/31/89) when the 1986 Tax Reform Act had been implemented. Since the 1986 Tax Reform Act has equalized the tax rates on dividends and capital gains, the tax-effect hypothesis predicts that the price change-to-dividend ratio would be statistically indistinguishable from one, or that excess return on the ex-dividend day would not be statistically different from zero. During the post-1986 tax reform period, ex dividend day excess return decreased to 0.0379%, which is not different from zero at the 1 % level of significance. This finding is consistent with the tax-effect hypothesis. 2. Ttst of Short-Torn Troding Hypothosis Table 4.4 reports non-parametric test results of equality of ex-dividend day excess returns between the two periods. In Panel A of Table 4.4, ex-dividend day excess returns during the pre-1984 tax reform act are compared to those during the post-1984 tax reform act. Average ex-dividend day excess returns after the 1984 tax reform increased to 0.1518% from 0.1376% before the 1984 tax reform, and this increased percentage seems to be statistically different from zero at the 1 % level. This finding is similar to what Grammatikos (1989)" obtained by using the raw price ratios on the ex-dividend day. Grammatikos reports that the overall raw price ratio declines my beet Itnowledge, cramnetikos (1989) is the only study that deals with the effects of the 1984 Tax Reform Act. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 90 Taw* 4.4: Nflfi-Paramatrie Tnta of EquaUty of Ex-Oividand Day Exeau Ratuma Batwaan tho Two Pariods Savaral non-paramatric ttatittica ara ahown to taat tha null hypothatit that tha distribution of ax-dividand day axcats ratuma haa tha sama location paramatar across tha two diffarant pariods. Statistics basad on tha rank scoras of axcuss ratuma across the two pariods ara tha Wilcoxon test. Median test. Van dor Waardan test, and Savage test. Statistics basad on tha empirical distribution function ara tho Kolmogorov Smimov test and Kuipar test. Average ax- dividand day axcasa returns ara cakulatad as tha maan diffaranca batwaan tha obsarvad ratum on tha ax-dividand day and tha simpla avaraga of aach stock's daily ratuma in tha estimation period (-55 through -11). Tha Wilcoxon, Madian, Van dar Waardan, and Savage Tests ara simpla linaar rank statistics that ara powerful for location shifts. Tha Kolmogorov test and the Kuiptar test ara statistics basad on tha empirical distribution functions. Panel A : Tha Pra-1984 Tax Reform Period Versus tha Post-1984 Tax Reform Period Dividend Avaraga Ex-divldand P-Valua of Non-paramatric Test Yield Quartila Day Exeau Ratum Safora tha After tha Wilco Medi Van Sava Kolm Kuip 1984 Tax 1984 xon an dar ge ogor ar Reform Tax Test Test Waar Test ov Test Reform dan Test Test 1 -.0092 .1287 .0001 .0003 .0001 .3452 .0001 .0001 ( Low I 2 .1389 .2881 .0005 .0106 .0003 .3432 .0001 .0001 3 .2761 .2130 .6294 .1981 .9665 .0028 .0048 .0001 4 .1896 -.1204 .0001 .0001 .0001 .0001 .0001 .0001 ( High ) Sample Maan .1378 .1518 .1739 1821 .1479 .0001 .0001 .0001 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 91 Tabit 4.4 (corn'd) Panal B : Tha Pra-1986 Tax Raform Pariod Vartus tha Post-1986 Tax Raform Pariod Dividand Avaraga Ex-dividand P-Valua of Non-paramatric Taat Yiald Quartila Day Excaaa Ratum Bafora tha Attar tha Wilco Madi Van Sava- Kolm Kuip 1986 Tax 1986 xon an dar 0« ogor ar Raform Tax Taat Taat Waar Taat ov Taat Raform dan Teat Taat 1 .1287 .0256 .0242 .0330 .0217 .0608 .0549 .2254 I Low 1 2 .2881 .2329 .2471 .2493 .2178 .8050 .2771 .2812 3 .2130 .1246 .2216 .1190 .2600 .4563 .0595 .0467 4 -.1204 -.1545 .8969 .6889 .8735 .8886 .9661 .8384 ( High ) Sampla Maan .1518 .0379 .0001 .0005 .0001 .0028 .0001 .0006 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 92 from 0.877 before the 1984 tax reform (1/1/75-7/17/84) to 0.796 after the 1984 tax reform (7/18/84-12/31/85).'* To further examine the changes in behavior of short-term traders between the before- and after-1984 tax reform period, uniiice Grammatilcos (1989), we compare excess returns for 11 days (-5 through + 5) surrounding the ex-dividend day. Because the 1984 Tax Reform Act was intended to expose short-term traders to more risic, increased exposure would be reflected in the return behavior for several short-term trading days before and after the ex-dividend day in addition to on the ex-dividend day during the post-1984 tax reform period. According to the short-term trading hypothesis, it is furthermore predicted that there more significant excess returns will occur for 11 days surrounding the ex-dividend day during the post-1984 tax reform than prior to the 1984 tax reform. Table 4.5 shows average excess returns around the ex-dividend day for the period from January 1, 1980, to July 17,1984, before implementation of the 1984 and 1986 Tax Reform Acts. The 11-day window for average excess return has two significant excess returns at the 5% level, when the ex-dividend day excess returns are excluded. Tables 4.6 and 4.7 present average excess returns around the ex-dividend day for Period II (7/18/84-12/31/86) and Period III (1/1/87-12/31/89) when the 1984 Tax Reform Act was in effect. During the latter period, the 1986 Tax Reform Act was also in effect, but it does not appear to have affected the behavior of short-term traders (see Section C.2 for discussion). Consistent with expectation, the 11-day window for average excess returns during Period II (7/18/84-12/31/86) has six significant excess returns on each side, excluding the ex-dividend day, two of which are significant at the 1% level. During Period III (1/1/87-12/31/89), the 11-day window for average excess ^^GranmatDcce (1989) compares a sample of 19,407 observations before the 1984 tax reform (1/1/75-7/17/85) and a sample of 2,032 observations after the 1984 tax reform (7/18/84-12/31/85). His samples appear to be seriously unbalanced. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 93 TaWt 4.5: Avtrag* Excms RMwnt Around Iho Ex>Oividmd Day for tha Pariod January 1, 1980. to July 17.1984 Tha «ample pariod contains 21.334 dividend payment event*. Average excess returns are calculated as tha mean difference between the observed return on day t and the simple average of each stock's daily ratuma in the estimation period 1-55 through -11). Average standardized excess returns are the average excess returns standardized by the standard deviation which was estimated during the same estimation period. Average market-adjusted excess returns are the average excess ratuma adjustad by the CRSP equally waighted index for day t. Tha t-tast statistic is the ratio of the mean excess retum to its standard deviation that was astimatsd from the time-serle* of mean excess retums. **’ and ' denote the average excess retums significantly different from zero at the 0.05 and 0.01 level, respectively. Trading Day Average t-value Average t-value Average t-value Excess Stand Markat- Return ardized Adjust-ed (%) Excess Excess Retum Retum (%) {%) -5 -.0544 -1.53 -.0160 -1.13 .0032 0.19 -4 -.0124 -35 -.0032 -.20 .0065 0.39 3 .0820 2.31* .0362 2.27* .0332 2.00 -2 .0349 0.96 .0254 1.59 .0207 1.24 -1 .0645 1.62 .0429 2.69* .0995 5.99** Ex-Dividend Day .1375 3.67" .0741 4.66** ;1651 11.14** + 1 .0360 1.01 .0150 0.94 .0556 3.36* + 2 .0525 1.46 .0319 2.00 .0326 1.97 + 3 .0344 .97 .0224 1.41 -.0233 -1.40 + 4 .0547 1.54 0327 2.05* .0041 .25 + 5 -.0763 -2.15* -.0279 -1.75 -.0121 -.73 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 94 Tibto 4.6: Avtrigc Excm * R#lwm# Around 6w Ex-OivMond Day for Iho Poriod July 18.1984. to Docombor 31,1986 Tho sompio poriod comoino 10.335 dividond poymont ovonts. Avorago oxcou ratuma ara calculatad aa tha maan diffaranea batwaan tha obaarvad ratum on day t and tha almpla average of each atock'a daily retuma in tha aatimatkm period i-S5 through -11). Avaraga atandardizad axcaaa returns are tha avaraga axcaaa ratuma atandardizad by the atandard daviation which waa eadmatad during the aama eadmation period. Average market edjuated exceaa ratuma ara the avaraga axcaaa ratuma adjuatad by the CRSP equally weighted index for day t. Tho t taat atadadc la the redo of tha maan axcaaa ratum to ita atandard daviadon that waa aadmatad from tha dma-tarlas of maan axcaaa ratuma. and " " denote the average exceaa retuma aignificandy diffarant from zero at the 0.05 and 0.01 level, respectively. Trading Day Avaraga t-value Average t-value Average t-value Exceaa Stand Markat- Return ardized Adjuat-od Exceaa Excaaa Retum Ratum 1%) (%) -5 .0219 -.74 .0014 .09 .0055 0.28 -4 .0755 2.56* .0530 3.48* .0684 3.51" -3 .0754 2.55* .0448 2.95*' .0435 2.23* -2 .0896 3.04** .0678 4.46** .0436 2.23* -1 .1004 3.40** .0751 4.94* .0722 3.70" Ex Dividend Day .1517 5.14** .0905 5.95" .1840 9.43" + 1 .0775 2.63* .0516 3.39" .0695 3.56" + 2 .0624 2.11* .0378 2.48* .0655 3.36" + 3 .0469 1.59 .0356 2.34" -.0092 -.47 + 4 .0551 1.87 .0449 2.95" -.0159 -.82 + 5 -.0125 -.42 .0007 .05 .0010 .05 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 95 Tabla 4.7: Avaraga Excaaa Ratuma Around tha Ex«Dlvldand Day for tha Pariod January 1.1987. to Daeambar 31.1989 Tha aampla pariod containa 12.525 dividand payment avants. Avaraga axcasa retums ara caiculatad as tha maan diffaranca batwaan tha obsarvad ratum on day t and tha simple avaraga of aach atock's daiiy ratums in tfia aatimation pariod ( 55 through -11). Avaraga atandardizad axcasa ratums ara tha avaraga axcasa ratums standardized by tha standard daviation which waa aatimatad during tha aama estimation pariod. Avaraga markot-adjustad axcasa ratums ara tha avaraga axcasa ratums adiustad by tha CRSP aquaily waightad indax for day t. Tha t-tast statistic ia tha ratio of tha maan excess ratum to its standard daviation that was estimated from tha tims-sarias of maan excess ratums. *** and ' denote tha average excess ratums significantly different from zero at tha 0.05 and 0.01 level, raspactivaly. Trading Day Average t-value Average t-valua Avaraga t-valua Excess Stand Marfcat- Retum ardized Adjust-ed (%) Excess Excess Ratum Ratum (%) (%) -5 .0163 .39 -.0137 -.86 .0545 2.14* -4 .1021 2.43" .0431 2.72" .0418 1.64 -3 .1096 2.6 V .0495 3.13" .0383 1.50 -2 .1377 3.27" .0660 4.17" .0529 2.08’ -1 .0862 2.05* .0403 2 54" .0420 1.65 Ex-Dividand Day .0378 .90 .0077 .49 .0586 2.30* + 1 .0021 .05 -.0082 .52 .0046 .18 + 2 .0534 1.27 .0214 1.35 -.0168 -.66 + 3 .0274 .65 .0206 1.30 -.0290 -1.14 + 4 .0169 .40 .0021 .13 -.0807 -3.17" + 5 .0011 .03 -.0346 -2.18' -.0194 -.76 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 96 retums has four cum dividand significant excess returns, one of which is significant at the 1 % level. During periods II and III, excess returns appear to concentrate cum- dividend because ex-dividend days are likely to include a minimum period of 46 days in which short-term traders are restricted to trade by the law. Thus, these evidences strengthen the short-term trading hypothesis, predicting that the 1984 Tax Reform Act would have an effect on the retum behavior of short-term traders. When average excess retums are compared for 11 days, including the ex- dividend day between the before- and after-1984 Tax Reform Act by using average standardized excess return and average market-adjusted excess return, the evidence is consistent with what is shown in Tables 4.5, 4.6, and 4.7. 3. Têst of TêX’Effoct HypothoMis Panel B of Table 4.4 compares average excess returns on the ex-dividend day between the before-1986 tax reform period and the after-1986 tax reform period. Consistent with the predictions of tax-effect hypothesis, the overall average excess return on the ex-dividend day has decreased from 0.1518% before the 1986 tax reform to 0.0379% after the 1986 tax reform. The null hypothesis of equality of excess retums between the two periods is rejected at the 1 % level of significance in five non- parametric tests. Further comparisons of ex-dividend day excess retums between the before- and after-1986 tax reform are reported in Tables 4.5, 4.6, and 4.7. Tables 4.5 and 4.6 show that, prior to the 1986 tax reform, average standardized excess returns and market-adjusted excess retums, as well as average raw excess returns, are positively significant at the 1 % level. However, as shown in Table 4.7, after the 1986 tax reform, both average raw excess retum and average standardized excess retum are reduced to levels which are not different from zero at the 1 % and 5% levels of significance. Only market-adjusted ex-dividend excess returns are significant at the 5% level. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 97 These results lend strong support to the tax-effect hypothesis. It can be also inferred that the differential taxation of dividends and capital gains has affected the market valuation of the stock on the ex-dividend day. Previous studies'* on the ex-dividend day effects of the 1986 Tax Reform Act show mixed evidences. For example, Michaely (1991) reports that the 1986 average price change-to-dividend ratio is not significantly lower than those of 1987, rejecting the tax interpretation of the ex-dividend day behavior. In contrast, Robin (1991)'* argues that the ex-day abnormal retums declined between the pre 1986 tax reform period and the post-1986 tax reform period. The finding reported is consistent with Robin's. Dividend clientele effects imply the negative correlation of ex-dividend day excess returns with dividend yields. In order to examine possible changes in the dividend clientele behavior, we regress ex-dividend day excess retums on dividend yields for the pre- and post-1986 tax reform periods. The cross-sectional regression results are reported in Table 4.8. The coefficient estimates of dividend yields are negative and statistically significant at the 1 % level for each of the two periods. A comparison of the coefficient estimates of dividend yields between the pre- and post-1986 tax reform periods shows an interesting evidence on the change of dividend clientele behavior. Since the coefficient estimate of dividend yields changed from -8.6683 during the pre- 1986 tax reform period to -4.9625 during the post-1986 tax reform period, it can be argued that the 1986 Tax Reform Act somewhat discouraged dividend clientele behavior of investors through the equalization of differential tax rates on dividends and capital gains. ^^Other ceeearchera who examine the effect of the 1986 Tax Reform Act include Bolster. Lindsey and Mitrusi (1989). and Fedenia and Grammatikos (1991). Bolster. Lindsey and Mitrusi (1989) compare trading behavior in December 1986 and January 1987 and find that relative trading volume was considerably high in December 1986 for long-term winners. Fedenia and Grammatikos (1991) argue that the abnormal volume around the ex-dividend day increased in 1986 and 1987. ^^Robin (1991) was published at the end of June 1991, at which time this study had been completed. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 98 T#W# 4.8: Crou-SaetioiMl ftosrtsaion of Ex*Oivid«nd Day Exeoss Ratuma on Dividand Yialda EXRET, - dto + YIELD, + ( Panal A : Bafora tho 1986 Tax Raform ( 7/18/84 . 12/31/861 EXRET, - 0.2320 • 8.6683 YIELD, (7.085"! (-3.015") Adi-R: - 0.0008 Obaarvationa » 10331 Panal 8 : Aftar tha 1886 Tax Raform ( 1/1/87- 12/31 /88 I EXRET, - 0.0848 • 4.8625 YIELD, (3.562! (-3.203"! Adj-R* - 0.0007 Obaarvationa - 12521 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 99 4. Portfolio Tost RosuNs This section reports the main results of this study. The aim of this section is to investigate the relationships between the ex-dividend retum behavior and the ex-post portfolio choice around the ex-dividend day in the context of the 1984 and 1986 tax reforms. Previous studies [see, e.g., Lakonishok and Vermaelen (1986) and Fedenia and Grammatikos (1990)1 which advance or support the short-term trading hypothesis seem to have followed a model of trading volume behavior, rather than an equilibrium model with appropriate measure of risk and return. Thus, the portfolio approach is taken in this section. Following Long's (1977) argument, it is expected that the Tax Reform Acts of 1984 and 1986 have effects on the risk-expected retum trade-off of investors, which wiil be reflected in the portfolio choices around the ex-dividend day. To show possible changes in the risk-expected retum trade-off of investors, efficient portfolios around the ex-dividend day are constructed, using Markowitz's (1959) portfolio selection model. Assuming that investors take into account short-term trading activities after the announcement of dividend payments, the windows between the dividend announcement day and the ex-dividend day should be chosen for their portfolio selection." In order to examine the effects of the 1984 Tax Reform Act, which increased minimum holding period from 16 days to 46 days, it would be reasonable to consider the windows from -25 to 2 days and from +2 to +25 days relative to the ex-dividend day." To this aim, securities are omitted whose ex-dividend dates are 25 or fewer trading days apart l*Thia can alao iaolata potential bias due to the proximity of the announcement day and the ex-dividend day [see Bades, Hess and Kim (1984)). ^^To test for the sensitivity, these procedures can be repeated on different windows. However, these efforts may be restricted by the inefficiencies of the extant portfolio selection algorithms in dealing with sparse matrices. Nevertheless, we repeated the window of 10 days, which yielded consistent results with our conclusion reported in this study. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 100 Tabic 4.9: Comparison of Cum Dividand and Ex-Oividand Efficient Portfoiioa batwaan lha Pra 1984 Tax Raform Pariod and tha Pott-1984 Tax Raform Pariod For aach pariod, 14 comar portfoiioa baginning tha portfolio with maximum ratum ara shown. Portfolio revisions around tha ax-dividand day ara axpactad to concantrate on about 14 stocks bacausa pravious rasaarchaa indicate that an avaraga of 14 firms want ax-dividand on aach day Isaa Karpoff and Walkling i1990)1, or an avaraga number of 18.6 stocks arc in aach ax-day portfolio isaa Eadas, Hess and Kim (1984)1. 4 is tha investor's trada-off batwaan risk and ratum. E(r,| and o ', denote tha axpactad ratum and standard daviation of tha portfolio, raspactivaly. Panal A ;: Bafora tha 1984 Tax Raform Cum-dividand Efficient Portfolio Comar À E(r^ Average Portfolio Portfolio Dividand Yiald 1 509.34 1.8578 14.7883 0.0058 2 108.87 1.8480 7.5858 0.0043 3 40.92 1.8359 8.0959 0.0082 4 29.78 1.8204 5.0044 0.0082 5 25.18 1.8128 4.5735 0.0058 6 24.17 1.8097 4.4324 0.0085 7 12.05 1.5518 2.3280 0.0088 8 7.42 1.5088 1.4898 0.0071 9 8.78 1.4998 1.3814 0.0078 10 2.87 1.4197 0.5895 0.0071 11 2.84 1.4188 0.5835 0.0073 12 2.29 1.3985 0.4805 0.0070 13 1.90 1.3784 0.3881 0.0078 14 1.87 1.3745 0.3808 0.0078 Avaraga Sampla Dividand Yield - 0.0112 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 101 Tabla 4.9 (corn'd) Ex*dividand Efliciant Portfolio Comar À EIrp Avaraga Portfolio Portfolio Dividand Yield 1 51.76 1.6605 16.1636 0.0042 2 47.84 1.6469 15.03 0.0043 3 45.26 1.6356 13.79 0.0045 4 12.69 1.4372 2.2959 0.0049 5 4.32 1.3432 0.6966 0.0066 6 3.22 1.3229 0.5436 0.0076 7 2.96 1.3151 0.4955 0.0067 8 2.42 1.2960 0.4032 0.0062 9 1.76 1.2771 0.3155 0.0067 10 1.46 1.2667 0.2612 0.0076 11 1.05 1.2496 0.2366 0.0060 12 0.91 1.2411 0.2216 0.0066 13 0.66 1.2362 0.2164 0.0065 14 0.60 1.2260 0.1994 0.0116 Avaraga Sampla Dividand Yiald- 0.0112 Panal B : Aftar tfw 1964 Tax Raform Cum-dividand Efficiant Portfolio Comar À Elr.) Avaraga Portfolio Portfolio Dividand Yield 1 10.06 2.0560 6.9653 0.0017 2 9.11 2.0257 6.3653 0.0033 3 6.79 2.0066 6.0236 0.0029 4 6.34 1.6460 3.6240 0.0024 5 5.44 1.7666 2.9234 0.0029 6 3.51 1.6577 1.7510 0.0031 7 2.94 1.6147 1.4739 0.0062 a 2.92 1.6124 1.4604 0.0057 9 2.14 1.5054 0.9197 0.0057 10 1.74 1.4392 0.6633 0.0056 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 102 Tabla 4.9 (cant'd) 11 1.18 1J348 0.3589 0.0056 12 1.12 1.3218 0.3291 0.0055 13 0.94 1.2857 0.2547 0.0060 14 0.73 1.2340 0.1684 0.0057 Avaraga Sampla Dividand Yiald- 0.0097 Ex-dividand Efficiant Portfdie Comar À Elr.) a*-. Avaraga Portfolio Portfolio Dividand Yiald 1 118.63 1.8080 22.3477 0.0246 2 26.31 1.7007 6.7936 0.0135 3 8.85 1.6105 3.6198 0.0119 4 6.40 1.5583 2.8245 0.0101 5 4.95 1.4910 2.0612 0.0097 6 4.73 1.4808 1.9627 0.0084 7 3.59 1.4089 1.3645 0.0081 8 3.47 1.4015 1.3129 0.0085 9 1.47 1.2328 0.4792 0.0077 10 1.26 1.1988 0.3866 0.0076 11 1.10 1.1714 0.3219 0.0077 12 1.07 1.1656 0.3093 0.0078 13 0.71 1.1001 0.1931 0.0081 14 0.61 1.0700 0.1534 0.0077 Avaraga Sampla Dividand Yiald- 0.0097 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 103 from the declaration dates. As a result, 445 securities are obtained from the Period I (1/1/80-7/17/84) sample, 224 securities from the Period II (7/18/84-12/31/86) sample, and 224 securities from the Period III (1/1/87-12/31/89) sample.'* These sub-samples satisfy the previous selection criteria used for the uncontaminated sample. Rrst, to consider the effects of the 1984 Tax Reform Act that increased the risk exposure of short-term traders, cum-dividend and ex-dividend efficient frontiers were compared for the pre-1984 tax reform period and the post-1984 tax reform period in Table 4.9. For each period, 14 corner portfolios beginning the portfolio with maximum return are shown. Portfolio revisions around the ex-dividend day are expected to concentrate on about 14 stocks because previous studies indicate that an average of 14 firms went ex-dividend on each day [see Karpoff and Walkling (1990)1, or an average number of 18.6 stocks are in each ex-day portfolio (see Eades, Hess and Kim (1984)1. Consistent with expectations, cum-dividend efficient portfolios dominate ex- dividend efficient portfolios during the before- and after-1984 tax reform periods.'* Furthermore, the efficiency gap between these cum- and ex-dividend efficient portfolios expanded during the after-1984 tax reform period, compared to those during the before- 1984 tax reform period. Graphic comparisons of cum- and ex-dividend efficient frontiers for the two periods are shown in Rgures 4.1 and 4.2. These results indicate that the portfolio choices around the ex-dividend day cannot be independent of the risk-expected retum trade-off of traders as suggested by Long (1977). Table 4.9 reports further evidence on the clientele behavior of investors. A comparison of average portfolio dividend yields for the cum- and ex-dividend efficient portfolios indicates that the average portfolio dividends yields for the before-1984 tax this study, portfolios ars formed by selecting securities (not observations) that meet our selection criteria. l*The statistical tests to be conducted are the joint tests of equality of mean and variance between the cum- and ex-dividend efficient portfolios. Non-parametric tests rejected the null hypothesis at the 1% level. For the pre-1984 TRA period, P-values of Wilcoxon tests were: 0.0033 for mean, 0.7227 for variance. For the post-1984 TRA period, P- values of Wilcoxon tests were: 0.0009 for mean, 0.8030 for variance. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 104 Ejqwcted Return 2.7 2.4 2.1 18 15 H 1 2 0.9 0.6 0.3 0.0 I I I I T 1 r n I r -2 0 2 4 6 6 10 Standard Deviation •••E x -d iv id e n d BBB cum—dividend Figura 4.1 COMPARISON OF EFFICIENT FRONTIERS (befme-1964 TRA) Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 105 E)qiected Ratum 2.7 2.4 2.1 18 15 12 0.9 0.6 0.3 0.0 -2 0 2 4 6 8 10 Standard Deviation # # # Ex—dividend B S S cum—dividand Rguw4.2 COMPARISON OF EFFICIENT FRONTIERS (after-1984 TRA) Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 106 reform period and those for the cum-dividend efficient portfolio after the 1984 tax reform show an increasing trend as comer portfolios increase. The average portfolio dividend yields for ex-dividend efficient portfolios after the 1984 tax reform show a decreasing trend vis-a-vis corner portfolios. An interpretation of this result is as follows. If tax-induced short-term trading occurs, investors are likely to have more interest in the capital loss deduction and financial cost deduction than in the dividend income tax deduction because dividends are progressively taxed. Thus, tax-induced short-term traders are motivated to choose first the securities with lower dividend yields in their portfolios. However, if tax-induced trading is restricted such as in the ex-dividend days after the 1984 tax reform, there will exist only dividend capture traders who are interested in buying and selling securities to strip dividends for purposes other than taxes." Next, to examine the effects of the 1986 Tax Reform Act, cum-dividend efficient portfolios between the before-1986 tax reform period and the after-1986 tax reform period were compared. Table 4.10 presents evidence that the after-1986 tax reform portfolios are dominated by the before-1986 tax reform portfolios.'' This result is also depicted in Figure 4 .3 . These findings are consistent with Long's prediction that, under the new tax regime, investors are motivated to restructure their portfolio holdings to reflect the new tax rates. Another evidence reported in Table 4 .1 0 is that the average portfolio dividend yields, vis-a-vis corner portfolios, show an increasing trend during the before-1986 tax reform period, whereas those average portfolio dividend yields display a stabilized trend during the after-1986 tax reform period. Prior to the 1986 Tax Reform Act, long-term traders are taxable at the unfavorable tax rates on dividends relative to capital gains. 20on# exemple is Jepsnese life insurance companies which have a preference for dividend income because of regulatory laws. These traders have incentives to buy the securities with higher dividend yields. 21p-.values of Wilcoxon tests ttere: 0 .0 0 0 9 for mean, 0.9S 33 for variance. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 107 T#W# 4.10: Comparison of Cum4Nvidand Efficlsnt Portfolios botwssn tha Pro 1986 Tax Raform Poriod snd tha Post 1988 Tax Raform Poriod For aach poriod, 14 comar portfolios boginninp tho portfolio with maximum rotum ara shown. Portfolio ravisions around tho ox-dividond day sro oxpoctod to concantrate on about 14 stocks bacausa pravious rasaarchaa indicsto that an avorago of 14 firms wont ox*dividond on aach day (sao Karpoff and Walkling (1990)1, or an avaraga numbar of 18.6 stocks ara in aach ax-day portfolio Isaa Eadas, Haas and Kim (1984)1. 4 ia tha invaator'a trada-off batwaan risk and ratum. Eir,) and o ', danote tha axpactad ratum and atandard daviation of tha portfolio, raspactivaly. Panel A : 8afore tha 1986 Tax Raform Comer À Eir^ o *. Avaraga Portfolio Portfolio Dividand Yiald 1 10.06 2.0580 6.9853 0.0017 2 7.45 1.96S1 5.4289 0.0033 3 5.81 1.8833 4.2903 0.0039 4 3.94 1.7234 2.7321 0.0034 5 3.93 1.7222 2.7225 0.0071 6 3.66 1.6809 2.4094 0.0100 7 3.49 1.6539 2.2156 0.0095 8 1.79 1.3788 0.7628 0.0084 9 1.57 1.3411 0.6360 0.0082 10 1.55 1.3357 0.6193 0.0081 11 1.39 1.3013 0.5181 0.0076 12 1.32 1.2844 0.4724 0.0071 13 1.12 1.2367 0.3561 0.0066 14 1.04 1.2173 0.3141 0.0070 Avaraga Sampla Dividand Yiald - 0.0069 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 108 Tabto 4.10 (corn'd) Pant) B ; Aftar tha 1986 Tax Raform Comar À E(r^ Avaraga Portfolio Portfolio Dividand Yiald 1 98.07 1.3623 8.2020 0.0010 2 79.99 1.3552 6.9429 0.0031 3 41.78 1.3312 4.0240 0.0027 4 29.39 1.3204 3.2535 0.0030 5 9.67 1.2996 2.4411 0.0032 6 7.51 1.2621 1.7967 0.0033 7 2.71 1.1482 0.6336 0.0031 8 1.11 1.0414 0.2255 0.0032 9 0.98 1.0246 0.1905 0.0031 10 0.88 1.0089 0.1614 0.0031 11 0.71 0.9813 0.1175 0.0032 12 0.40 0.9202 0.04S7 0.0038 13 0.36 0.9127 0.0441 0.0038 14 0.31 0.9034 0.0379 0.0038 Avaraga Sampla Dividand Yiald - 0.0069 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 109 Expected Retajzn 2.7 2.4 2.1 18 15 12 i 0.9 0.6 0.3 0.0 I I I I I I I I I “ I 1 I I - 2 0 2 4 6 8 10 Standard Deviation BBB cum-dividend after-1986 TRA 1 cum-dividend befiote-1988 THA Raure4.3 œMPARISON OF EFFICIENT FRONTIERS (befare-1986 TRA \%nua after-1986 TRA) Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 110 Thus, during the before-1986 tax reform period, dividend clientele effects would be in effect, and it is expected that the negative correlations between dividend yields and excess returns will be observed during the same period. After the 1986 tax reform, since investors are taxed at the same rates on dividends and capital gains, the clientele incentives of long-term traders are predicted to be reduced. Average portfolio dividend yields are expected to maintain a stable trend regardless of the expected return during the post-1986 tax reform period. These evidences are consistent with the expectations of the dividend clientele hypothesis. As shown in Table 4 .1 0 , the average portfolio dividend yields show a negative correlation with the expected portfolio return during the before-1986 tax reform period, but the dividend yields seem to have remained at a stable level during the after-1986 tax reform period. F. Chapter Summary and Conclusions This essay has investigated the portfolio choice around the ex-dividend day in the context of the 1984 and 1986 tax tax reforms. The fundamental idea has been to show that the ex-dividend value of stock and short-term trading behavior around ex- dividend day reflect the trade-off of risk and expected return. The 1984 Tax Reform Act increased the risk exposure of tax-induced short-term traders by extending the minimum holding period for the dividend income tax deduction. The 1986 Tax Reform Act decreased the dividend tax rate from a maximum rate of 50% to 28%, leading to the equalization of the tax rates on dividends and capital gains. By comparing the efficient portfolio frontiers between the pre-tax reform period and the post-tax reform period, an attempt was made to report changes in the risk-expected retum trade-off around the ex- dividend day. These portfolio results was related to the excess retum behaviors on which the 1984 and 1986 Tax Reform Acts are also documented to have had effects. First, the effects of the 1984 and 1986 Tax Reform Acts on the retum behaviors around the ex-dividend day were documented. Numerous studies have evidenced the Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Ill tax-effect hypothesis and the short-term trading hypothesis. But, although these tax reforms provide a valuable laboratory to test the extant hypotheses, researches on the ex-dividend day effects of the 1984 and 1986 Tax Reform Acts are very few (see Grammatikos (1989), Michaely (1991), and Robin (1991)], and ihei, evidences are mixed. The findings in this essay are consistent with both the tax-effect hypothesis and the short-term trading hypothesis, and they are summarized as follows: (1) Prior to the 1986 tax reform, significant positive excess returns were observed on the ex-dividend day. (2) During the post-1986 tax reform period, average ex-dividend excess returns were not different from zero at the 1 % significance level. (3) There existed the negative correlationships between the dividends yields and ex-dividend day excess retums, which correlationships decreased from the pre-1986 tax reform period to the post-1986 tax reform period. (4) Average ex-dividend day excess returns increased significantly during the post-1984 tax reform period compared to the pre-1984 tax reform period. (5) During the post-1984 tax reform period, more significant excess returns were observed for 11 trading days before and after the ex-dividend day compared to the pre-1984 tax reform period. The second approach is quite different from the extant researches, and more insightful and consistent conclusions were obtained. The cum- and ex-dividend efficient portfolios between the pre-1984 tax reform period and the post-1984 tax reform period were compared. The following evidences to support the short-term trading hypothesis were found. (6) Cum-dividend efficient portfolios dominate ex-dividend efficient portfolios. (7) The efficiency gap between cum-dividend efficient frontiers and ex-dividend efficient frontiers has expanded more during the post-1984 tax reform period than during the pre-1984 tax reform period. Following Long's (1977) argument, it was predicted that the 1986 Tax Reform Act would decrease the expected after-tax return ar the new tax rates and increase the after-tax variance of the portfolio. Consistent with this expectation, it was found that (8) The efficient portfolio frontier during the post-1986 period was dominated by the efficient portfolio frontier during the pre-1986 tax reform period. This result may be interpreted as a tax effect since the tax premiums decrease with an decrease in dividend tax rate. But, the variance as risk factor was not considered in the tax-effect hypothesis. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 112 To conclude, the evidence of excess return behaviors supports the existence of both tax effects and short-term trading around the ex-dividend day. These tax effects and short-term trading are not mutually exclusive, but co-existent as long as short-term trading is tax-induced. They interact to have an impact on the ex-dividend return and volume behaviors. This finding was effectively supported by portfolio test results. The portfolio approach is quite advantageous over the extant approaches, because this approach incorporates the tax effect as well as the short-term trading. The portfolio approach also seems to be useful in explaining dividend capture tradings by Japanese life insurance companies that are not tax-induced. It can be emphasized that the ex-dividend behavior cannot be understood independently of the risk-expected return trade-off. Also, it is proposed that this portfolio approach can be best used in many branches of corporate finance ( for example, turn of-the-year effect ) where portfolio revisions are considered to occur. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Chapter 5. EMPIRICAL TESTS OF DIVIDEND SIGNALLING A. Introduction With the publication of Bhattacharya’s (1979) paper, the notion of imparting information to shareholders with dividends (commonly referred to as dividend signalling) has attracted considerable attention among financial economists. Miller and Rock (1985), John and Williams (1985), Ambarlsh, John and Williams (1987), Ofer and Thakor (1987), and Williams (1988) develop formal models of dividend signalling. These models, in general, derive their logic from the seminal papers of Spence (1973, 1974) and Riley (1979). Miller's (1987) suggestion of the stringency conditions necessary for a sustainable dividend signalling equilibrium notwithstanding, signalling still appears to be one of the most catching theories explaining the enigma of dividend policy. The need and opportunity for dividend signalling arise when corporate agents know more about the quality of the enterprise than outsiders do. Outside investors cannot differentiate the quality of projects undertaken by the firm, because they lack perfect information. They value and buy the firm's stock at the price that corresponds to only the average quality of projects. Agents (insiders) are the only ones who know the true prospects facing their firms and attempt to optimize the objective function of shareholders. Thus, they have an incentive to avoid the adverse selection problem by informing the market of their firms' excellence. They can signal high cash flows of the firm by paying higher dividends. If this signal is correctly received in the market, the shares of the firm will command a higher price, benefiting the firm and its shareholders. The intention of this essay is to test empirically the John and Williams' (1985) (JW) signalling equilibrium model. This test is conducted in order to provide statistical evidence of the empirical validity of this model. The reason for selecting this model is twofold. First, the authors argue that their model is empirically testable and provide guidelines for such testing. Second, this model is a reasonable representation, given its 113 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 114 assumption, of the class of models dealing with the so-called signalling equilibrium. There are no reports of large-scale empirical validation of this model. The JW model implies that, in equilibrium, corporate insiders with superior information about future cash inflows have incentives to signal by distributing larger dividends "... if and only if the demand for cash by both their firm for investment and their stockholders for liquidity exceeds the internal supply of corporate cash" (John and Williams (1985). p. 1055]. The signal of larger dividends results in raising the stock price and consequently benefiting the firm or current shareholders when selling the stock. The JW model attempts to explain why many firms prefer paying dissipative' dividends to repurchasing shares, while others simultaneously pay dividends and float new shares. It is argued that the payment of dividends is induced by clienteles of stockholders who demand current cash. Their model also shows the explicit relationship among the announcement effect, dividends and eu/n-dividend market values. The intrinsic weakness of signalling models is the difficulty in testing empirically their theoretical implications. JW assert that their signalling model is testable, because the cross-sectional impact of dividends is a simple function of observable variables. In the remainder of this chapter, the task of empirical testing is undertaken. Section B contains a more detailed discussion and review of the JW model. In Section C. the methodology is explained. Section D describes the data and the proxy variables. Section E presents tests, results, and their interpretation. Section F is a summary, including some guidelines for future research. ^The term "dissipative" originates from electrical engineering and it means wasteful. In this context, the meaning is that shareholders are required to bear the cost of this signal via their tax liability for the dividend payments. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 115 B. Th# John and WBliams Modal 1. Nênativ» JW develop a signalling equilibrium with dividends and single dissipative costs (i.e., a proportional tax on dividends). In this model, capital gains are not taxable, and issuing, retiring, and trading shares are costless. According to their arguments, if the liquidity demand by the firm and its current shareholders exceeds internally generated funds, corporate insiders are motivated to distribute a taxable cash dividend and to reveal to outside investors the present value of their firm's future cash inflows. This signal would result in raising the stock price to the benefit of current stockholders. JW attempt to explain why some firms do not pay dividends, whereas others do and simultaneously sell new shares to investors. In order to finance investments, a firm must either issue new shares or retire fewer outstanding shares. Similarly, to collect cash for personal use, current stockholders must sell their shares. In either case, stockholders have to suffer some dilution in their fractional ownership of the firm. Current shareholders desire to reduce this dilution on corporate or personal accounts. Hence, insiders, in the best interest of current stockholders, are inclined to convey favorable information by paying a taxable dividend. Recognizing the relationship between favorable information and the value of the shares, the market bids up the stock price, mitigating stockholders' dilution. The JW signalling equilibrium is achieved because the marginal gain from paying dividends is balanced against the marginal cost incurred by the tax on dividends. In the market, stocks with marginally larger dividends are rewarded with a premium, compensating current shareholders for the dissipative effect of their marginal income tax. For firms paying marginally smaller dividends, the dead weight cost of dividends outstrips the marginal benefit of reducing ownership dilution. In the JW model, an optimal signalling equilibrium is reached when firms with favorable inside information distribute higher dividends than firms without favorable information. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 116 2. Tntêblu Analytkêl Modêl^ The JW formulation of the problem is anchored in the idea that "... other things equal, firms which pay dividends have clienteles of stockholders who demand current cash-such as widows, senior citizens, and financial institutions" {ibid., p. 1065). This concept is expressed mathematically, as a signalling equilibrium. In their Equation 1 3 .' Accordingly, D{X) ~ 'lMax{I-C*L. 0 )'La(X)] fOT X i L. (5.1) t where D(X) is the optimal dividend at equilibrium; t is the constant, marginal personal tax rate on the dividend; I is investment; C is corporate cash available; L is the liquidity need of the shareholders; and X is the present value of the future cash inflow. Equation (5.1) implies the equilibrium condition. A t equilibrium, the dividend increases proportionally to the logarithm of the stock's present value, a property common to many other dividend models. Equation (5.1) also shows that dividends decrease in t and C and increase in L. The key element of this model is that the firm has exogenous demand for liquidity by its shareholders. The demand for cash is the catalyst to signal via dividends. This, in turn, implies that the boundary condition for insiders to pay dividends is C g I + L; that is, the demand for cash from the firm is non negative. HYPOTHESIS #1 : Demand for liquidity causes dividend payments. ^The models in this section build on John and Williams (1985) to posit JW'e model correctly for this testing. ^The mathsmatical proof of JW's models is presented in Appendix C of this dissertation. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 117 When insiders optimally declare a dividend (when shareholders' demand for cash, I + L, exceeds the firm's internal supply of cash. Cl, the model for the marginal effect of announced increments in dividends becomes* P'[D{X)] . g. - L ^ for X > 1 and C < I*L, (5.2) where P'(D(X)] is the change in the stock price with respect to the announced increments in dividends; and P[D(X|] is the firm's market price. According to equation (5.2), increments in dividends around the optimal dividend, D(X), increase the market price of the firm's stock, P(D(X)1, measured cum- dividend. This means that larger dividends are associated with cum-dividend higher stock prices. This is not surprising, since prior to JW many scholars document dividend announcement effects both theoretically and empirically. However, unlike many event- study models, a testable implication of the JW signalling model is that the stock price is an increasing function of the signal.* HYPOTHESIS #2: The equilibrium price is monotonie in the dividend signel (or private information!. Another aspect of JW's equation is that its cross-sectional effects are explicit and empirically testable. JW contend that in equilibrium, shareholders are compensated by a proportional increment in their stock price for incurring the proportional cost of personal taxes levied on dividends. Equation (5.2) makes the cross-sectional connection for dividend announcement effects on stock prices, market value of the firm, tax rate, dividends, and shareholders' demand for cash. HYPOTHESIS #3: Announcement effects (excess returns during the event period) are an endogenous function of dividend levels, the demand for liquidity, and market value o f the firm. ^Equation (15) of JW. ^Sm Acharya (1988). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 118 C. Mathodoiogy Stock price reactions to quarterly dividend announcements are estimated by calculating prediction errors for stock returns of dividend paying firms from a market model. Then, the return prediction error for stock i on day t is expressed as PEic - Rte ■ «i - (5*3) where PE,, is the prediction error for stock i's return on day t; Rh is the realized return to firm i on day t: a„ fii are the market model's parameter estimates; and Rm, is the return on the equally weighted Center for Research in Security Prices (CRSP) market portfolio on day t. The market model parameter estimates are obtained using the stock returns of each firm over 120 trading days, ending 30 days before the dividend announcement date. For each firm i and each trading day t within the event period It = 30 to t = + 15, 46 days), the prediction error, PE*, is calculated. The average prediction error on event day t for a portfolio of N securities is then expressed as APB^ • ( 1 ) (5 4 ) The test statistic, Z,, for APE, is based on the standardized prediction errors (SPE*), and is obtained from It is customary to accept that the distribution of Z, is unit normal. Each standardized prediction error of firm i on day t is defined as Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 119 . ID - S ’ 1* SPE,,» PEu/ lo ^ a * ) 1, (5.6) i j where a, is the standard deviation of the residuals in the market modei estimation period: T is the number of days in the estimation period; Rm, is the return on the market portfolio for day t; Rm is the mean return on the market portfolio over the estimation period; and Rmk is the return on the market portfolio for the k th day of the estimation period. The cumulative prediction error for stock i, CPE,. is obtained by summing prediction errors over the event time, (5-7) while the cumulative average prediction error over the event time is calculated from CAPE • (1) CPEi . (5.8) The prediction error for trading volume is also estimated, using a "trade-volume” model, similar to equation (5.31: “ Ti ♦ «1 V^e ♦ 5ie (5.9) where V„ is the rate of change in the trading volume from day t * 1 to day t for firm i. V„ > (trading volume for firm i on day t - trading volume for firm i on day t * 1 ) / trading volume for firm i on day t ■ 1 ); Vm, is the rate of change in the trading volume from day t 1 to day t for the market. - (trading volume for the market portfolio on day t minus trading volume for the market portfolio on day t * 1 ) / trading volume for the market portfolio on day t -1 ); K„d, are trade volume model parameter estimates; and C, is the prediction error for the trading volume of stock i on day t. In order to obtain the average prediction error, the Z-statistic and the cumulative average prediction error, the procedure outlined for equations (5.4), (5.5), and (5.8), respectively, is followed. This "trade-volume" model [equation (5.9)1 is different from prior models-i.e., those that typically follow Beaver (1968)"in two ways. First, prior models regress each Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 120 firm's percentage of shares traded to its shares outstanding on the percentage of total shares traded on exchange to the market shares outstanding. These models carry an inherent tendency of heteroscedasticity in spite of attempts by several authors (e.g., Pincus (1983), Richardson, Sefcik and Thompson (1986)1 to correct for this problem. The model here is more analogous to the market model and less heteroscedastic since each firm's trading change rate is regressed on the market trading change rate. Second, the focus of prior models has been restricted to the abnormal trading volume to model information effects. More important, this paper highlights the role of normal trading volume, which has been mentioned only sparingly in prior research. Traders in the market are generally divided into two groups; information traders and liquidity traders. While abnormal trading volume may reflect the trading activities of information traders, normal trading volume itself may exhibit a perpetual level of demand of liquidity traders. The rationale for using this normal trading volume is that the level of liquidity demand may vary with market conditions, but it is independent of any information motivated trading. D. Data and Proxy Variables Firms traded on either the New York Stock Exchange (NYSE) or the American Stock Exchange (AMEX) are selected to form a primary data base. In order to be included in the primary selection, firms must appear in both the CRSP Daily Returns, 1991 Files and the Quarterly Industrial COMPUSTAT, 1991 File. For selection, stocks must also have consecutive data for 1986 through 1990. The study period is limited to 20 quarters, commencing from the first quarter of 1986. The constraint on the starting quarter is due to the availability of daily trading volume data on the CRSP tape.* This database is called the "initial sample." ^Trading volume data on stocks appear in the CRSP file only after January 1986. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 121 For a firm to be included in the final aample, quarterly dividend announcement dates must be both available and verifiable. Quarterly dividend announcement dates are identified from the declaration dates of quarterly dividend payments on the CRSP tape by assuming that these events are known to investors on the next business day through public media. Other sources of announcement dates are the Wall Street Journal Index and Moody's Annual Dividend Record. All stocks for which these dates cannot be ascertained are excluded from the final sample. Stocks that report stock splits and stock dividends are also eliminated from consideration. For each quarter, the sample data are grouped according to the direction of changes in dividend per share from the previous quarter to the current quarter. The vast majority of the data are for firms whose dividends are unchanged from quarter to quarter. Consequently, the final sample consists of firms that increase dividend per share, decrease dividend per share, and do not change dividend per share on a quarter- to-quarter basis. Table 5.1 presents the quarterly frequencies for each category, for both the initial and the final samples, by year. The cardinal problem of testing models such as (5.1) and (5.2) is the derivation of the theoretical variables for the purpose of empirical measurement. Even variables that are unambiguously defined are hard to measure because of lack or inaccuracy of data. Other variables are more troublesome in that they are defined only in a theoretical sense and cannot be observed directly. Thus, empirical research has to face the impasse of (1) resorting to the use of proxy variables, or (2) grinding to a halt. Option (1) is followed in this study, and the uneasy task of the empirical derivation of the theoretical variables is undertaken. Data used in this study are obtained from the CRSP tape and the COMPUSTAT 1991 Quarterly Industrial File. Because firms use different fiscal years, those listed on the CRSP tape are matched with those in the COMPUSTAT file, on a calendar basis. Table 5 .2 shows the names and definitions of COMPUSTAT variables and their data item Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 122 » I l s «o e a i « M « I « s s «CD I If u O i s tn î M m i S # i II II S I ? wo ! II «0 0 u il h» S i i I » Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 123 numbers. In the following subsections, these proxy variables are discussed in more detail. 1. U q u U itv In the JW model, the demand for liquidity is an exogenously specified function, unrelated to the quality of the firm. Also, liquidity is not influenced by investors' assessment of the signal used by the firm. A reasonable proxy for liquidity demand seems to be trading volume, although in such situations one must exercise exceptional care. Trading volume is affected by investors' perception of firm quality and their reassessments caused by changes in the firm's dividend policy. Release of information concerning quality through dividends could cause investors to alter their portfolio compositions, thereby creating an endogenous link between the signal (firm quality) and trading volume (demand for liquidity). This link is at odds with the JW notion that demand for liquidity gives rise to the desire to signal. Liquidity demand of current shareholders is characterized by the trading volume and the number of shares outstanding. Hence, liquidity demand is measured by price per share times the number of shares traded as estimated from the "trade volume" model. LIQUlDTYit - . (5.10) where ^ is the stock price of firm i on day t; and VOL* is the expected trading volume for firm i on day t: v s ü T t ■ (i*?n) • voi-f.t-i. (5.11) where V* is the expected trading volume change rate: and VOLi,,., is the actual trading volume on day t 1. Then, liquidity demand is standardized by the market value of the firm for sake of comparability across firms. For each firm, liquidity demand on day t is computed as a liquidity ratio in relation to firm size: Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 124 L IQ U ID ^ (5.12) - - I S C T ^ Finally, the unique liquidity proxy for each firm is obtained by taking the average of the liquidity ratios for trading days t - 30 through t - 8, in order to nullify possible informational effects. Consequently, RATIOBAR, is a proxy for a normal level of liquidity demand the firm attempts to maintain. RATIOBAR, . LIgMTIO,, (»•«! In order to double check the validity of this proxy, an alternative measure of liquidity demand, the natural logarithm of the number of shares outstanding before the dividend announcement date, is also used. LNSHARt > iTi (SHAR^m,) , (5.14) where SHARESNO) is the number of shares outstanding as oft - 30. 2. F im SU» Arm size is measured by the natural logarithm of the firm's market value, which is obtained by multiplying the number of common shares outstanding at t « 30 by the average stock price for t= 30 through t= 8. MKTVALi » SHARESNOi • 7 ] 1 i (5.15) . SHARESNOi * ^ ^ P it ■* e— 30 where P* is the stock price for firm i on day t. 3. Dhridtnd PoSey In the JW model, the firm attempts to satisfy a certain perpetual level of liquidity by maintaining its level of dividends. The firm's dividend policy is measured by its Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 125 dividend yield before the dividend announcement. The dividend yield for firm i is estimated as the dividend per share divided by its average stock price. DIVYIELDi - (5.16) where DjVIPSHA, is the dividend per share for firm i; and PRC, is the average stock price for t « 30 through t= 8. Larger changes in dividends are construed as stronger signals and are expected to draw larger price reactions, ceteris peribus. To control for the actual magnitude of the dividend change, the difference between the current and previous dividend yields is considered. DIVYDIFF, • DIVYIELDi^t - DIVYIELD,,^.^ , (5.17) where Dl WIELD,,, is the dividend yield for the current quarter; and Dl WIELD,,,., is the dividend yield for the previous quarter. D IW D IFF represents the dividend signal corporate agents use to signal the firm's quality to the market. 4. Aeeountihg Veriebles Accounting data are collected from the quarterly COMPUSTAT tape. They include increases in investment (INVEST), long-term debt issued (DEBTISSU), long-term debt retired (DEBTRETI), sale of common and preferred stock (EQUSELL), purchase of common and preferred stock (EOUBUY), cash dividend (DIVIDEND), cash and cash equivalents (CASH), and increase in short-term working capital (WORKCAP).' The demand for cash by the firm, if measured by net investment, may ignore demands for cash for other purposes. Other reasons for cash needs may include repayment of debt, repurchase of shares, and increases in short-term working capital needs. As a result, the level of investment may vary, cross-sectionally. To control for ^Definitions of COMPOSTAT data are presented in Table 5.2. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 126 Î i S S i S S a e I S 'i ! g ? 1 I 2 E E ? I ? ? 5 3 a if I 5 ; 5 I' I I 1 1 I Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 127 the cross-sectional effect of external financing, an approach of grouping firms by external financial needs is taken. The cross-sectionally observable model of the theoretical equilibrium condition of Equation (5.1) is then DEHANDi • mVEST, * LIOUIDTY, - CASH,, i » l N. (5.18) where DEMAND, is the demand for cash from firm i; INVEST, is the increase in investments in firm i; CASH, is the cash and cash equivalents of firm i; and LIQUIDTY, is the liquidity demand of current shareholders. E. Testing 1. Sêmplê StsH$Hes Table 5.3 presents quarterly summary statistics of the variables included in the tests. The contrast of the quarterly statistics for dividend-increasing and dividend- decreasing firms suggests several interesting differences. First, a simple one-way analysis of variance for the variables in the table indicates that the dividend-increasing and dividend-decreasing groups are different at the 0.01 level for stock repurchase.' This is a fascinating observation in and of itself. It suggests that the amounts dissipated by firms that show a propensity to increase dividends is in fact larger than the amount of dividends paid. This is so since the repurchase, de facto, amounts to a selective dividend disbursement scheme. Second, at the 0.01 level of significance, the null hypothesis of the ANOVA (no difference between the two groups) is also rejected for the market value of the firm. This difference may suggest that larger firms are more willing (or perhaps better expected) to pay dividends. Finally, at the 0.01 level, there are significant differences in dividend yields for dividend-increasing, dividend-decreasing, and dividend-unchanging firms. However, no significant differences in total dividend payments are found, although net stock ie speculated that this difference is due to the influence of the repurchase component. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. CD ■D O Q. C g Q. Tmbl# S.3i Quarterly SuBary Statistica of tba Pimal Saapla, 1986 1/4 - 1990 4/4 ■D Incraasa in Dividend# Oacraaaa in Oividanda Mo Change in Oividanda CD (3,650 Obaarvationa) (287 Obaarvationa) (10,435 Obaarvationa) WC/) Naan Standard Oaviation Maan Standard Oaviation Maan Standard Oaviation o" 3 OIWIILO (») 1.23 2.55 1.70 1.23 1.05 0.83 0 3 DIVIOBNO 39. SB 142.43 31.34 61.77 38.96 143.10 CD 8 MTIOUK («> 0.36 0.37 0.38 0.35 0.39 0.41 ■D • H A M S NO |MM) 56.50 95.35 32.67 62.23 50.71 93.36 (O'3" NKTVAL (HHS) 2404.33 502B.34 574.45 1612.26 2010.06 5151.81 1 3 CASH (HMS) 15.07 213.02 0.46 68.46 2.59 185.05 CD INVEST (HHS) 29B.45 25B2.2B 121.52 784.77 255.99 2108.00 "n c 3. DEBTISSU (HHS) 149.B3 1011.07 54.38 265.73 123.24 833.66 3" CD DEBTBETI (HHS) 100.99 771.50 53.68 315.81 92.11 727.81 CD ■D EQUSELL (HHS) 11.45 62.65 13.26 60.98 12.33 65.31 O Q. C EQUBUy (HHS) 27.48 121.54 1.00 5.66 23.26 128.38 a O MORKCAP (HHS) -3.B2 79.07 -1.85 38.45 -4.79 101.18 3 ■D O CD Q. ■D CD (/) ro (/) 0 0 129 repurchases are significantly greater for dividend paying firms. Since the bulk of the data is post-1986, this cannot be attributable to a tendency to reduce shareholders' tax liability. Table 5.4 presents the Pearson correlation matrix of the proxy variables. Each cell in the table contains the correlation coefficients of dividend-increasing, dividend- decreasing, and dividend-unchanging firms. The correlation coefficient of DIVYIELD with all the other variables and their level of significance are shown in italics in the table. For dividend-increasing firms, DIVYIELD is negatively correlated with RATIOBAR and MKTVAL at the 0.05 level. However, the correlation with EQUSELL is positive and significant at the 0.01 level. Correlations are not significant for CASH, INVEST, DEBTISSU, and EQUBUY. For dividend-decreasing firms, the picture is quite different. Except for the negative correlation with RATIOBAR (-0.251, the correlations with all other variables are insignificant. This is consistent with the JW argument to the extent that the model should hold for increases in dividends only. 2. HYPOTHESIS #1: Demand for UguiditY eêus»s divkfend paynMnts. a. Information-Motivated Trading Several studies, notably in the accounting literature, are concerned with abnormal trading volume to explore the information content of a corporate event. Some authors (e.g., Bamber (19B6)] acknowledge that using abnormal trading volume does not provide a better theoretical tool to determine the information content of an event. Still, it is believed that abnormal trading volumes are indicative of thedepth of the market's response to information, while abnormal returns reflect thebreadth of such response. Accordingly, by regressing each firm's trading volume change rate on the market trading volume change rate, the market's reaction to the dividend announcement can be measured. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 130 M #4 ^ I MOM 5 V 0 MOMM> « Ml MOMo m M GOO « M o m m Ml o G m m O o o M G M m O M d d d d d d d d d S 8 M ^ im m r* 8 sss U GOO OGG G G G : Ml # O MOM G M t MOO MOO GOG G M O o e o o d o O O o G G G If I I I I I I i s # « « GG o M 4»G o m o o Ml o o o G M G MM G MM m Ml O I d e d d d d d d d d d d d d d d d d 0 I! s M M o M M m r § O O O o o o o o o O #4 O G o G e o o I I d d d d d d o o o G G G G G G II I I I I I I n a « « « m Ml m m M m o G M O M M 2 GG M G o o M o G G OG G GG G G GGGM o GG S G iiAl o d d d d d d d d d d d d d d d d d d I | i I I I ??? 9 ? « * lA il i 5 5 H i II I i l S I s I I Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 131 Table 5.5 exhibits the statistical properties of the trading model [equation (5.9)1. For a 7-day window around the announcement day. the dividend-increasing firms have two of the APE'S significant and positive at the 0.01 level. In contrast, dividend-decreasing firms have three significant APE's, one prior to the announcement day and two after the announcement. When one looks at the plot of the ACPE's for the entire 46-day time horizon (Figure 5.1), it seems that just prior to the announcement, for both types of firms, trading volume changes are mostly random. The trading volume changes become systematic at and after the announcement day. Dividend-increasing firms have decreasing trading volumes. Dividend-decreasing firms also have declining trading volumes. For this subsample, all APE's are negative with the exception of days 6, 11, and 15. The data and Figure 5.1 indicate that, while the market correctly anticipates the positive price effect of dividends, trading decreases for both declining and increasing firms. The anticipation of dividend changes should have been coupled with a change in trading around the same days. Instead, it is shown that the increase in dividends decreases trading considerably. For stable dividend firms, trading increases. Thus, the paradox here is that trading decreases with a change in dividend policy and increases in the absence of change. The ad-hoc conclusion is that a change in policy, whether it is construed as good or bad, thwarts trading. This evidence is in complete juxtaposition to the JW model and are perplexing as well. The findings of stock volume responses to dividend announcements reported here are different from the empirical studies of Harris (1986) and Richardson, Sefcik and Thompson (1986). These two papers document a positive correlation between daily stock volume and stock price changes. The evidence reported here, however, is consistent with a recent study by John and Lang (1991). John and lang demonstrate the impact of insider trading around dividend announcements, congruous with the Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 132 aeaawxaria SSRSSRSSSSSRRSgSSRâsâggâgSSSSg S: OOOOOOOOOM 1 I I I I I I OOOONOONMMM 'omMvinmiow m I# p* m »0% @1 W M S om«#vvr*vim # o m » mM # p. O I o I ? OinOMO#NMf4MOOMMOOWWw6 ip*mm#a%miAp»m i - mm#imominv## imo%ewMMinmin s s s s s s g s MioMONimvNmwlONOP^ioemwo# mMiomwmooM I e e ooeoeooooo »cSdododoed m II II II II I n iAiAO#oeoo»inMmp#OMm#nor**NA»No#»»mMoo SSSRSSSSRaSSSSRSaSRSSRSRS^ o P» m • m £ « • p» 9t 40 #4 40 « 40 Pi m m 0 0 0 5 n « m n 40 n f* « O «0 04 9t 0 0 m S #4 r*» # Pi M m OD « 0 m 0 44 0 M p* n o M 04 P» m m 40 m 0 44 m P* 0 0 44 0 M M M M d d Pi fi #4 d 44 •H Pi m m N ? ? I 1 1 1 1 1 ? I ? 9 7 1 1 1 7 ? 1 t 9 1 t 7 7 1 1 t s « « « « : lA « p» m g 2 e S • m 04 m i m P» m o 0 0 0 O 9 fi 04 « o m m Pi 0 0 Pi 0 O 0 S 0 5 0 0 #! MM O m d 6 d Pi d d #4 0 Pi 44 d fi Pi r» 1 ?? 1 ? 1 9 1 7 9 9 1 1 7 7 O 5 « Ml m* 2 m 2 £ in M « 40 m m # 0 0 0 0 2 £ o e • 40 m # m m Pi 0 44 0 0 0 M 3 i g 2 V #4 m S • 40 f* O m 40 V 40 0 0 0 0 p r» m 0 0 o M tfl Ml V fi p* 40 Pi p» O Pi 0 Pi 0 p» SO 0 S 0 0 fill ????o ?d o ? 9 9d 9 9 9 d d d 9 d d d 4^ d d d d O I 0 p%0 0 P* O 0 0 0 0 0 0 O 0 0 0 44 P» 44 0 0 0 0 0 0 O 0 O d d 0 0 d 0 0 0 2 0 0 0 0 0 0 0 0 0 0 0 44 0 O 0 O 0 4«î I l | î 0 0 0 Pi 0 0 Ci 0 0 f i 0 Pi 0 Ci 0 0 0 0 0 0 0 0 0 0 0 0 "I 2 O 0 0 0 0 O 0 0 0 0 0 P* 0 « 0 0 r* 0 0 01 I g 0 O 0 0 0 0 n 0 0 0 0 0 0 P*» 0 P» 2 0 0 0 o 2 0 0 0 0 0 0 0 0 O 0 0 0 0 0 O 0 0 0 0 0 0 0 o O 0 0 P* P* P" 0 m 0 O m 0 0 0 d d d d d d d d d d 44 44 44 M 1 ?9 9 9 9 i 9 1 9 9 1 9 9 9 1 1 1 g o lii «« # in « lO 0 44 e 0 o o 0 0 o 0 0 0 44 0 0 • m0 0 o 0 0 0 0 0 0 0 0 O 0 0 0 0 o 0 0 0 0 p» 0 O 0 0 ? d 0 0 0 0 0 d 0 0 0 44 0 d d 44 d 0 P* 1 99 9 d d d lA m! # mMMoaa^aaamwMOa r> a a « n n •< h I I I I I I I I ...... I I I I I I I I I I Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 133 M tf) O r* « o m Ci p A» P p Ci A» o 6 r • • # M d d PA d PA Ci P m PA Ci d Ci Ci ■ m m m m M m m Ci m m mm m m Ci m 1 g « 1 • 1 g b r* w m PA P A* PA p p p PA p m p r t f S 0 p in O # A»WPP m o A» P O O p p m > s w V m m P O Pi m m PA Pp S O p m m • P m P m p A*P m m m m o § t & 0 £ M Pi e g 2 d d d d M Pi PA Ci Ci N Ci Ci 1 1 g ï < 3 ►MAl • 0 < flU « r* lA m lA p PP p P PA Ci O p p M pAm P#pP P O A» p m O p p S w S o M w M m Ci PA m Ci m : ? ? d d d w • 2* u # o M p Ci p p PpP Ci PA O g g b »lA # PA p Ci p p A* p PA P O o p g g P g » 2 m p m PA pP p A» m o • g 0 b O M o e p m Ci Ci o PA po o » w b < flb Al m e O o d d d d o ? ? ? ? ? ? ?? 5 m m PA fp pp p m o p Ci A* p o p 2 j i d m PA p A» A» m p Pi Ci Pi P p p d 5 ? m A# M Ci Ci Ci Ci Ci n Ci mCiCiCi m e# # % 1e •> e0 > I b » m lA O p P p p p p mp 9 ■2#2 0 « » «0« A» O P p p p O p u g u b A» m lAp P PA O p O P p pAn r»r* • m P m p p m p p P p £ 5 1 Pp p p p P & I 1 ? 7 7 7 7 7 K at 7 7 7 7 S £ 2 4# # * I • m « m» 2 PA p p p P p m Ci o O #APA O PA pCi O m o PA p Ci M 8 M o tf> pA PA d d PA d d P & £ 7 ? ? T 7 7 7 ? & 1 « 2 m p p P m p m p m p Ci g w g m *a M pm P p pp Ci p p po O I g b g e V * p P P p p p p pp p O O V O o p P p pCi p p m p p p 0 2 I 8 % «4 d d d d d d d ? i £ Al a ? ? ?? ? ? 5 p%#A m m O O A* p P 6 li-L A» o o p ■ g g # • 2 d d PA d p p d p Ci d d p d p « A# m mm Ci m Ci Ci Ci Ci Ci ? AI 2 m mmm «# te I # e » 9 > 0 w m4 M O g p* p PA o O A» A* A»lA p 2 £ 2 r* » o m p mP P P Ci 2 g 0 b tfi m VlA m P PP O m p p m «M « p#m m m m p P Ci Pmp P P p p I S i 2 pA Ci Ci CiCi Ci Ci # g 1 s s M ?? 7 7 7 7 7 7 7 # < fr cod 2 e #« • Î 2 ® 5 2 M 2 P P P p Ci PA P P mP O p P P “O f AI « mo « P p p Ci 'î P A» Ci P PP g f# 3 AI PA PA Ci PA ? ? d d d d P d d d d Pd M ' • 1 O # o m PA p p mp O p p P p g 0 Q 2 2 m P p p p p o P p p p p p g b « • o M PA p m p p Q P p o Ci pA #A O O o Ci PA p Ci PA Ci mm S% 0 2 —4 d d d d d d d d d 2 £ Al 6 ? ? ? ? ? i ? I 1 S p I"«1 e M Û II Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. CD ■D O Q. C g Q. "O Dividend-Increasing Firm s Dividend-Decreasing Firms Dividend-Unchanging Firms CD wC/) o' 3 0 3 CD 8 ■D (O' 3" W 1 3 CD "n c 3. 3" CD ■DCD O Q. C a O 3 ■D # # # m « • ■ « # 'M « O CD Q. Note: 1) Connected curves indicate CAPEs. 2) Needle type bars indicate AFEs. ■D CD Figure 5.1 Information-Motivated Trading Volume (/) Volume APE# and CAPEs (/) w 135 results reported in Table 5.5 and Figure 5.1. All data considered, the verification of the empirical validity of the JW model is still missing. b. Liquidity Trading The motivation of liquidity traders is different from that of information traders Information traders enter the market with information that must have changed the fundamental price of the stock. Liquidity traders sell their stocks because they desire liquidity to meet consumption or savings objectives. Liquidity demand by shareholders reflects the unique ownership structure of the firm. Consequently, it can be safely assumed that (1) liquidity demand shows a stable or perpetual trend over time ("normal" trading volume): (21 it is firm-unique: and (3) it does not reflect unexpected information in the market. "Normal" trading volume is obtained from the predicted values of model (5.9), theoretically clear of any noise in volume created by information flows. The approach that is taken here is to extract the "normal" level of trading as a proxy for liquidity demand. Predicted trading volume is standardized by firm size and averaged before the dividend announcement for each firm. In Figure 5.2, the ratio of normal liquidity trading to the market value of the firm around the dividend announcement day is charted. For dividend-increasing firms, liquidity ratios maintain a stable level before the dividend announcement date, increase slightly on the announcement day, and come back to the "normal" level thereafter. Dividend-decreasing firms also display a rather stable liquidity ratio before the dividend announcement date. The ratio Jumps dramatically at the announcement date, only to regress to the normal level three days after the announcement. Dividend-unchanging firms always maintain a normal level of liquidity, even around announcement days. The evidence that liquidity ratios rise sharply at the dividend announcement date for dividend-decreasing firms appears to be radically at odds with JW's argument that a higher level of liquidity trading should occur only for dividend-increasing firms. Unlike the JW signalling model, it is argued here that dividend policy is useo to satisfy liquidity Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. CD ■D O Q. C g Q. Dividend Unchanging Firms ■D Dividend-Increasing Firms Dividend Decreasing Firms CD C/) UQunnv W uQUonv UQunnv M o' HMioa 3 •an •on 0 3 CD 8 •OM ■D (O'3" 1 3 CD "n UM c 3. 3" CD ■DCD O Q. C a ,, 11 ...... o < I I i I I I I I I ' I I I ' -T 3 T3 O CD Q. Note: Needle-type bars Indicate the average ratios of normal liquidity trading to the market value of the firm. ■D CD Figure 5.2 Liquidity Trading around the Dividend Announcement Day w (/) o> (/) 137 demands of current shareholders. When the level of liquidity trading is low. the firm increases dividends to satisfy the liquidity needs of shareholders. When the normal trading level is too high, the firm decreases dividends, lowering the liquidity level of shareholders. Consequently, shareholders faced with declining dividends increase liquidity trading to boost their liquidity level. In the long term, a separating equilibrium will be achieved, and a certain perpetual level of liquidity trading will be maintained as portrayed by the case of dividend unchanging firms, c. Regressions o f Dividends on Liquidity Demends Feldstein and Green (1983) argue that "... there is the desire on the part of small investors, fiduciaries, and nonprofit organizations for a steady stream of dividends with which to finance consumption" [ibid., p. 17). JW encapsulates this argument in the firm's value maximization problem for its current shareholders such that liquidity demands may cause dividend payments because of information asymmetry. In the JW model, the demand for cash from the firm (DEMAND) is the total of the demand by current stockholders on personal accounts (LIQUIDTY) and the liquidity demand on corporate account (INVEST), less the firm's internal supply of cash (CASH). JW argue that the dividend is paid if and only if DEMAND is positive. Whenever DEMAND is positive, current stockholders and the firm collectively sell shares to new investors, thereby diluting the fractional claim of current owners. As a consequence, stock prices go up to reduce the dilution and to meet the cash demand of current stockholders. In this scenario, the line of causality goes from the demand for cash from the firm (DEMAND) that should motivate the payment of dividends. The optimal dividend condition of the JW signalling equilibrium of equation (5.1) implies that insiders pay dividends whenever C < I + L. In addition, when insiders pay dividends, the magnitude of dividends increases on the demand for personal liquidity (LIQUIDTY) and investment (INVEST) and decreases on the supply of corporate cash (CASH). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 138 I I I I m w* to o >4 o> at i ; s COCOM O #4 M I g § g V0 g% 82 I r 8: I I I 8 : O 8 R 88 I II » (0 I "c c M « n c n « n « 0 ^ 0 ^ 0 o•o n O #4 in o in o m o « 8 M o m o m o m V o V 88 o n o « 3 I m I m # m P# I M I I I I I ■7 I i ♦ ♦ « o « m # n « 8 PI O O «"* 5 ÎT S o PI 91 O r* #4 p" M RS m V m in mom n O n o ? I M I N # f! fiff I I I I 7 ' f ' i 'i I I e p* « 91 « in 1*8 1* o « m • ^ m m o m o m #4 w#4lA A91 m#4 V 11 3333 O m° S ss mom m o n o OD o o n o #4 o # o n s 's O u • 8 o o « s n n III! n il & t 1 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 139 In order to examine the causal relationship between liquidity demands and dividend payments, two puxy variables of shareholders' liquidity demand, RATIOBAR and SHARESNO, are regressed on dividend yield, DIVYIELD. The estimates obtained from these two regressions are shown in Table 5.6. Models (1), (21, and (3) are designed to prevent potential multicollinearity between two liquidity variables and assess each variable's marginal explanation for variations in dividend policy. Model (4) is constructed to control for cash demand by the firm itself. Toward this objective, INVEST and CASH are standardized by the market value of the firm. In the table, all the regressions have significant t statistics for RATIOBAR and In(SHARESNO). CASH is insignificant for all samples but dividend unchanging firms. As discussed earlier (see correlation matrix), a strong, negative correlation between RATIOBAR/SHARESNO and DIVYIELD persists in these regressions. This significantly negative relationship between liquidity demand and dividend policy contradicts the JW's optimal dividend condition [Equation (5.1)1 that higher demand for liquidity motivates the firm to pay larger dividends. Instead, this outcome lends strong support to the contention posited in the previous section: that higher (lower) levels of liquidity demand are associated with lower (higher) dividend payments. Since the estimates for INVEST and CASH are not significant, it appears that cash demand by the firm and the firm's cash holdings do not play a significant role in dividend policy. In short, this finding shows that liquidity demand by shareholders is one of the negative determinants for the firm's dividend policy. 3. HYPOTHESIS #2: The aquXbrium prie» is monotonie in the dividend signei lor privets kiformetionl. e. Single Lineer Pricing Reietion in Signelling Table 5.7 presents the average return prediction error. As the table shows, the announcements of quarterly dividend increases generate, on average, significant positive return prediction errors around the announcement day. In a 7-day window around the Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 140 6 r - m m ^ ^ im P» !!,Ô. g m M « I i “ >00000000000 00000000000 §§ 0000000000000000000000000000 I 1 II I I I I t I I I I I I I I I I I ?? » r# w o # o8;33%:888%3%:s;33g;ss3; :3 o «4 O ^ «4 ^ O 0000«4MM,#OOOMOOOMWONOOM I I I I I I I I I I I I I « M N M omomoNOm i'i o o o 9 o o o S 00000000 itÛ • • • • • # # • • • • • • g# ooeeeeoeeoeeoeeoooeeeooeeeoe o d I I ’ * I I I I I I I I I I I I m HI ti III m 0 0 f* 0 M M m 0 <0 0 m 0 •M 0 p* <0 0 52 0 r*»m d« <0 <0 10 m « m 0 d in A in i p m m « in in V m m m m inovetr^MOr^oo o M $ 0 # o r* I. o M m # P» m M P# M N M MMmmmwmmmmmm I i liiiiiiisis 000000 00000000000 23 >0000000000000000 00000000000 1 I I I I I I I I I I 1 I I I I I ?? i ÎI' % !3oSnSo?S3nS3S?oS3Sr;3o3Sl??SK ? ggggg^ • •••••■•••g g g g g g g ggggggggggg ii 0000000000 000000 f i l l 1 I I I I 00000000000I I I I I I II ?? I nmor>c I « n tfi H 0 M O....ôôiiiiiiiiiiiiiiiii O f ill ??? 000000000000000000000000000 i Ï M m wMOOo^oim#mwMM M M t 77777777777T777T- I I I I I I I I I I I I I I Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 141 • m « »4 « lA o i t I ^ P» HI «0 lA r> ' g0 mO m «w « m« w9 8 g g g o o o ll II iillissiil o o o o o o 0 0 0^00 0 o o d Ô d I ?? ?? 1 I I I I I 01 : : ; 8 3 g #0000imNVM O ? M M O M Ô M 0-4, 0 0 --0 0 !S CÎ • ^ ^ i i | So io §o ooogeooo ?? ooooo I i a in >4 ^!C?;oom^*iOM»in 8 SSg8SSSS8SSS85S ! I o o o o o o ?????????TÏTÎ i I 1 I 2 - Î I 8888 ? i0in»iaM»ioi Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 142 announcement day (15 days total), three average prediction errors are significantly positive at the 0.01 level. Of the three, one each is at, prior to, and after the announcement day. This is a strong indication that the market correctly anticipates a rise in dividends. The largest prediction error occurs one day prior to the announcement day (0.53 percent), more than any other prediction error in the same window. The announcement day and the following day have significant prediction errors of 0.35 and 0.16 percent, respectively. This is interesting because it is evidence of a somewhat slow reaction speed to "favorable" information. Still, even these errors may be deemed immaterial in magnitude. The cumulative average return prediction error, CAPE, is plotted in Figure 5.3 for visual dramatization. The CAPE rises to 0.9 8 percent by the announcement day and continues to drift upward through event day t - + 13, when it reaches its peak at 1.30 percent. In contrast with dividend increases, firms that announce quarterly dividend decreases, on average, experience significant negative return prediction errors around the announcement day. For an identical 7 day window for this sub-group, four average prediction errors are significantly negative at the 0.01 level. Three of the four are prior to the announcement day, and one is at the announcement day. This may indicate that while "bad news" is not very well anticipated, the reaction time to bad news is short. In the interest of visual contrast, the CAPE for dividend reducing firms is plotted in Figure 5 3. For these firms, the CAPE decreases to -4.37 percent by event day t= -2 and drops significantly after that until it reaches a minimum of -8.37 percent on event day t> +10. The only conclusion one may draw from these data and the figure (in addition to what was already surmised) is rather trivial: the market considers a dividend increase good news and a reduction in dividend unfavorable news. These findings are not at all surprising. Several empirical studies document the announcement e ffects wf dividends, supporting the information content of dividends (Aharony and Swary (1980), Asquith and Mullins (1983), Penman (1983), Patell and Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. (D "O O Q. C g Q. "O Dividend-Increasing Firms Dividend-Decreesing Firms Dividend Unchanging Firms CD pwmcnoii C/) o" 3 O 8 ■D (O'3" i 3 CD 3. 3" CD ■DCD O Q. C -a « &« ^ Il. H ■ If H 11111111111 n I I ■ . » Il ■ I ■ \ a O 3 'M n -M - « - « - t e B « « M -w -« -i e t « « 9t -tt -10 -tS i • 1 0 1 0 ■D wm ttvm O CD Note; 1) Connected curves indicate CAPEs. Q. 2) Needle-type bars indicate AFEs. ■D CD F ig u re S 3 The Reiationship between Signei end Price (/) Return APEe and CAPEe w o' 3 144 Wolfson (1984), Kalay and Loawenstein (1985) and Richardson, Sefcik and Thompson (1986)]. Hence, the empirical evidence that is shown here in the context of the JW model is not forthright. JW's contention is the formulation of a theoretical model for the announcement effects of dividends, documented in numerous empirical studies. Their model is a pricing function, dependent on dividends culminating in a signalling equilibrium. The positive relation between the dividend signal and price is just one of the outcomes of the signalling equilibrium. b. Firm Size and the Signai-Price Relationship in order to examine a potential size effect in signalling, every subsample is grouped into quintiles, based on the market value (MKTVAL) of the firm. Since it was shown in the previous section that the stock price at the dividend announcement date is an increasing function of dividend changes, three return interval prediction errors (CPE(k), k - 7, 5, 1) are regressed on dividend changes (DIVYDIFF) for each size- quintile. CPE(k), is the cumulative return prediction error of firm i for a k-day window, respectively. Table 5 .8 reports the regression results by size quintiles. Regressions for the three smallest size quintiles (quintiles 1, 2 and 3) show significant t-statistics for DIVYDIFF. In contrast, the two largest size quintiles (quintiles 4 and 5) have insignificant t-statistics for DIVYDIFF. When the R* and the P-value of the F-statistics are compared across each size quintile, it is found that dividend changes (DIVYDIFF) explain much better the variation in price changes for the three smallest size quintiles than for the two largest size quintiles. These differences areconsistent with the "small- firm effect" reported in many other studies (e.g., Banz (1981),Zeghal (1983), Barry and Brown (1984) and others). To interpret the differences between the small and large quintiles in the regression results here, it is argued that larger firms present easier and more timely access to information than do smaller firms. Further, more individuals disseminate this information. Because small firms present to outsiders more limited information. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 145 o e M tt < i s ^ i i il I I m « «A o M # 1 0 N ^ ^ M ^ i f i f M 8Slsl 8S338: # I I £ M $ 10 # # « «A A 10 I S ^ 0 ^ 10 O O O# m ## O 0 0 0^00 o M m #4 o O M M 0 O O CO m M m i 0 « «4 « *4 a is î.,3 —‘ M A tfi e 1 SSSS8! sssa s s 83^ S 8 S I î î M e M # M f# # m « M m « o # #4 s 10 -0 ^ po Q 0 0 0 O 9 A 10 A 0 o o N 0 0 w o I f S i SSiSiSSi 833388 e 0 0 0 O Q II SL in e H S=S I oS - ? î i II ë "S " ■ M'-J 1 T,£ 4i s s £ n i ! lA H I S I s Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 146 invastors require a compensating premium in the form of stock price increases when evaluating their investment risk. What the small-firm effect implies in the context of signalling is that the signal- price relationship may occur only in small firms and that signalling is not evident in large firms. JW argue that dividend and market value of the firm combine to increase stock prices positively at the dividend announcement date. The evidence presented here is not consistent with this argument. c. Polynomial Transformations o f tha JW Signalling Equilibrium To test the empirical validity of the JW signalling equilibrium, the consistency of the signalling-pricing function with the data must be explored. Toward this end, the independent variables of the regressions of dividend changes (DIVYDIFF) on the return interval prediction errors (CPEIk], k « 7, 5, 1) are introduced as polynomials. This non linear transformation is permissible because it does not affect the parameters of the model, but provides a better fit for estimation. Three models are fitted and tested, and their results shown in Table 5.9. In order to control for possible correlations among the independent variables, a step-wise regression is applied, where the independent variables enter one at a time. Since the JW dividend-signalling model implies a separating equilibrium, no regressions for dividend-decreasing firms show significance. When (DIVYDIFF)* is entered in the dividend-unchanging and dividend-increasing samples, its coefficients are significant and negative. Coincidentally, the coefficients of DIVYDIFF remain significant and positive. Likewise, adding (DIVYDIFF)* shows a much greater explanatory power of the model as evidenced by the significant increase of the R*’s (adjusted for degrees of freedom) in each sample. When (DIVYDIFF)* enters the regression, it always has a significant and positive coefficient in each sample, while the significance and signs of the other coefficient estimates remain unchanged. Adding (DIVYDIFF)* also raises significantly the explanatory power of the model. From the test results shown in Table 5 .9 , it can be concluded that the polynomial form of an empirically estimable signalling-pricing model Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 147 k 0 5 § § § II II i I I : I I in i l s s s 3 s : t t o o o o o I 3 I I s n î I Sî. in « 2? " « • » 10 1o * N ♦ ♦ h ew t t t #«««J I M M M # e #4 p* Pi # « in . o # m # 1 0 o«o« 2 10 o 01 o m Oi ^ O MlO ^ ^ ^ o e o n# 10 n P4 r * lO P» 01 o o m miooowp^NOioiooo III o1- p#M m o • n n V 0» 10 Pi 0* o*»m#4Mioo^vP*mo o o o m M P4 m 01 I o in ^ V N V ^ m #4 in % t I 8 m « S f s L s .31 m #» n# #» M o « in « 3 ? S f g S3S 83 83 83 S O 58 o o o r« 83 83 82 r» r* r* o d O iii ! 1 # 0 ÎÎÎ # S 2 ! t ’IZ Z g II u u u i III < 2 Î II 1 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 73 CD ■D O Q. C g Q. Tabla 5.9 (cont'd) ■D CD Dapandant Modal INTBRCBPT DivyoïFP (DIVYDIPP)^ (DlVrOIPP)-* R' Pcob. of P Slsa Vaclabla TO Tl T2 »3 WC/) o" CPlfl» m .0070 .0812 .003 .0052 3 (9.91)** (2.79)** O (21 .0063 .4155 -.3576 .011 .0001 |8.66)** (5.31)** (-4.60)** (3) .0045 1.3824 -6.4790 4.7079 .034 .0001 8 (6.061** (9.52)** (-8.29)** (7-87)** ■D Panai C. Olvldand-Oacraaalng Plraa 287 CPB(7) (1> -.0413 .4278 .003 .3481 (-4.471** (.94) (2) -.0363 1.1074 10.2969 .005 .5195 (-3.03)** (.97) (.657) (3| -.0262 3.2800 90.0410 655.1571 .009 4515 (-1.76) (1.49) (1-26) (1-15) CPE(5| -.0403 -.0147 .000 .9707 a (1) 3* (-4.96)** (-.03) CD (2| -.0362 .5446 6.4755 .001 .8271 (-3.44)** (.54) (.61) ■DCD (3| -.0261 2.7064 87.8215 651.8804 .007 .5555 O (-2.00)* (1.40) (1-41) (1-30) Q. CPB(l) (1) -.0205 .2990 .004 .3158 (-3.40)** (1.00) O (2) -.0136 1.2491 14.3967 .010 .2252 3 (-1.74) (1.69) (1-40) ■D (3) -.0089 2.2547 51.3062 303.2357 .013 .3025 O (-.92) (1.57) (1-10) (-81) Mot#* t valu## ar# in paranth####. CD Q. «Significant at th# 0.05 laval. •«Significant at th# 0.01 laval. ■D CD C/) C/) A 00 149 fits the data well. This polynomial pattern is inconsistent with the monotonie relationship between price and signal that other authors assume (e.g., Acharya (1988)1. d. Likelihood Retio Tests o f Non-UnearitY in Peremeters In this section, the cross sectional model of dividend announcements of equation (5.2) is examined. Equation (5.2) is the testable cross-sectional relation among the dividend announcement effect, the market value of equity, dividends, liquidity demand by current shareholders, and the cash demand from the firm. The non linear equation (5.2) is proposed to be tested with the following regression model: C- (MKTVALj * fPIVIDENDi - LIQUIDTYj) (5.19) CPE(k) I DEMAND where CPE(k), is the cumulative return prediction error of firm i for a k-day window; k is 7, 5, and 1, respectively; t is the constant, marginal personal tax rate on dividends; MKTVAL, is the market value of firm i at the announcement date; DIVIDEND, is the gross dollar amount of dividends; LIQUIDTY, is total liquidity demand by current stockholders; and DEMAND, is the combined net demand for cash on personal and corporate accounts. Multiplying the numerator by t and carrying through the division, equation (5.19) can be rewritten as CPE{k)t ■ t-MKTADJ, * C^-DIVADJ, - C-LIQADJ,. (5.20) where MKTADJ, is MKTVAL, / DEMAND,; DIVAOJ, is DIVIDEND, / DEMAND,; LIQADJ, is LIQUIDTY, / DEMAND,; and all other variables are as defined in equation (5.19). Using proxy variables and imposing non-linear parameter constraints on equation (5.20), the testable equation (5.21) is obtained: CPEik)i - Po * P,ln(AarvaL), ♦ PjOJVYDIFF, ♦ fijRATIOBAJi, , (5.21) subject to the constraints: A) ” 0; A —A; A - (fi) '. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 150 JW argue that Not surprisingly. Increments In dividends around the optimal dividend, D(X), Increase the market price of the firm's stock, P(D(X)1, measured cum-dividend. More surprisingiy, this Impact Is proportional to the personal tax rate, t, on dividends as shown by substituting (14) into (151 (i.e., equation (5.19)1. In other words, in equilibrium stockholders must be compensated at the margin by a proportional increment in their stock price for Incurring the proportional taxes from dividends [ibid., p. 1063- 10641. Consequently, the estimation of equation (5.21) should yield a significantly different set of parameter estimates than an unconstrained model. The null hypothesis that the coefficient estimates of regression (equation (5.21)1 and the unconstrained estimates of the same model are not significantly different Is tested. Two sets of parameter estimates are obtained from regression equation (5.21); one set by minimizing the error sum of squares subject to the equality constraints of (5.21), and the other set by simple OLS. Since the constraints of regression (5.21) are non linear In parameter space, the multivariate secant method* Is used to minimize the sum of squared residuals. A likelihood ratio test Is applied to compare the error sums of square of the unconstrained model with that of the constrained model. The likelihood ratio Is then defined as: ■21ogX •= Wlog(.^^l - Wlog[.^^] , (5-22) where N Is the total number of observations; N 8 (6 ) = Z(y,-f(x„0))* Is the sum of squares estimated from the constrained I « 1 model; and N 8 (6 ) - Z(y,-f(x„6)l* Is the sum of squares estimated from the unconstrained I m 1 model. ^This is an iterative process, whereas a starting value for the coefficient estimatee is chosen. This set of values is changed continuously until the error sum of squares cannot be further reduced. This multivariate secant method is selected over other methods (for example, the Gause-Newton method) because it estimates the parameters from the history of iterations, rather than from an analytical derivation. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 151 0 • 0 • m 0 m 0 0 0 #4O 0 0 o 0 s S V m o 0 0 0 m 0 0 PI m p4 0 0 0 r* 0 P* 0 0 0 #4PI m 0 0 0 0 0 0 lA m 0 m #4O N 0 #4 0 m 0 0 0 £J0 P» 0 P» 0 o 0 0 0 0 0 #40 0 2 s «A 0 0 0 0 0 0 0 o 0 0 n 0 0 Q f#0 #4 #40 0 #4 0 0 0 0 O 0 II lA lA #4 P# #4 0 0 m 0 m m lA lA lA O o o m m m 0 0 0 m m m 0 0 0 0 0 0 0 0 0 0 0 0 0 p * p * PI 0 0 PI M f * 0 0 P I 0 m m 0 0 o o PI P» PI PI O PI 0 0 0 m r * N P* 2 0 0 m Q o #4 o 0 #4 0 m o O o e e #4 o 5 0 0 o 0 0 0 o V tA 2 M 0 0 m 0 0 0 0 0 P I m o in 0 m 0 m 0 PI 0 #40 #40 0 o o N p * 0 0 0 0 o 0 0 0 0 M 0 0 0 P# P* 0 0 0 PI 0 0 P I m 0 0 n m m 0 0 0 0 o 0 O 0 PIPI 0 m o 0 #4 #4PI m JN M M P» P» P» 0 0 0 0 0 0 #4 m m m 0 0 0 0 0 0 0 0 0 n f # n 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 PIPI M e « 0 0 r * M p * 0 0 0 0 0 *4 m 0 o Ô o 9 s 0 0 m 0 m 0 0 0 0 O o o o 0 0 lA 5 n 0 0 n o O o #4O o o o 0 0 0 0 0 0 0 • 2 m 2 e PI pk 0 0 0 0 0 0 3 s • p4 0 #4 (A 0 0 0 0 0 2 0 0 0 0 0 m 0 0 0 m PI 0 0 m 0 f* 0 0 #4 m PI 0 0 o 0 m #40 «40 t i M 0 0 p* 0 0 0 PIPI PI #4 PI M 0 0 0 O O o #4 0 0 0 P I 2 2 2 O O O 0 0 0 0 0 0 0 0 0 P» P» (3 0 P* P» r * 0 0 0 o o OP* P» P» m m m M M M M p# M P I P I m m m C4 PI PI m m m M m u! I IS 0 ** i ? lA r * 0 #4 0 #4 f* 0 #4 0 P» 0 SU • # M M 0 M 0 M 0 M 0 M 0 M 0 0 0 MM 0 li OS w & 6 W o 8 8 8 U 8 8 8 8 8 8 tfi ? 111 <0 2 0 0 O 5 0 0 0 0 «1 0 0 0 0 0 t s •*4 #4 #4 l?f Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 152 Asymptotically, 2lo(M - /If), where r It the number of restrictions. Table 5.10 reports the results of these likelihood ratio tests for dividend- increasing, dividend-decreasing, and dividend-unchanging firms, for each year and for the combined sample. For each subsample, the number of observations, the likelihood ratio statistic, and its probability are show n.N one of the likelihood ratio test statistics for the total sample is significant at the 0.01 or less level. For dividend- increasing firms, only the statistics of 1989 and 1990 using CPEID are significant at the 0.01 level. But the same results are obtained for the full sample as well. Otherwise, probabilities (of no difference) are high for both types of firms and for the pooled sample. Hence, the null hypothesis of "no difference" between the constrained and unconstrained models cannot be rejected. This is interpreted as strong empirical evidence that the proportionality condition of equation (5.21) and the implied signalling equilibrium of equation (5.2) posited by JW is either inconsistent with the data, or the empirical model is misspecified. 4 . HYPOTHESIS #3: Announemnmtt «fftaets (9xeess ntum s during thn nvnnt pnriod) « r* en nndogtnous function of dMdond hvois, the domond for Squidity. morkot vaiuo ond fkm sin. a. Tests of the Cross-Sectional Relation of the JW Mode! Equation (5.2) implies that the potential causes for stock price reactions to dividend announcements are the market value of the firm, gross dividends, liquidity demand by current stockholders, and the demand for cash from the firm. To test the intensity and duration of the prediction errors, the same windows around the announcement day as before are examined. The linear relationship between the quarterly obeorvatione can be pooled because the sum of independent etatietice is also Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 153 cumulative return prediction error (CPE(k), k « 7, 5, 1) and the proxy variables MKTVAL, DIVIDEND, LIQUIDTY, and DEMAND is studied. One should note, however, that DEMAND is defined as the sum of INVEST and LIQUIDTY, less CASH. In a linear relation, this presents a potential multicollinearity problem. In order to circumvent this problem arising from the correlation between LIQUIDTY and DEMAND, the latter variable is excluded from the cross-sectional regression. CPEIkl's are regressed on In(MKTVAL), DIVYDIFF and RATIOBAR for each sample. The linear relation to be estimated is CPE(k)i - «0 ♦ t^ln(HKTVAL), ♦ b^DIVYDIFF, * ijMTIOBARj, (5.23) where CPE(k), is the cumulative return prediction error of firm i; and k is a 7-, 5-, and 1-day window, respectively. The parameter estimates obtained from regressing each quarter's observations are checked for the OLS assumptions of normality and homoscedasticity. The analysis of residuals indicate that the assumptions of normality and homoscedasticity are not violated. Interestingly, contrary to recent arguments" by Eckbc, Maksimovic, and Williams (1990), no evidence of distribution truncation of the standardized residuals at the dividend announcement date for dividend-increasing and dividend-decreasing firms is found. The analysis of "condition indices" is applied as the multicollinearity diagnostic. The condition index number is the square root of the ratio of the largest eigen value to each individual eigen value. When this number is very large, estimates may be biased. Belsley, Kuh and Welsch (1980) suggest that condition indices greater than 10 indicate weak dependence and that condition indices of 30 to 100 indicate moderate to strong ^^Bckbo, MakalnoviCr and William# (1990) argue that bocauae of corporate inaider# ' private information, the diatributiona of reaiduala at the voluntary event announcement date will be truncated, and that thia truncation biaa will lead to the inconalatent eatimatora of OLS and GLS. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 154 h 0 8lA 3 f4 8 9 i i i 8 1 I m m # i a #4 #4 to Tc 8 8 8 I s s 3 8 8 f# ni « « « S13Î.R1 #m« # #4 « # «M « lA et m 3888:3 m #4 #4 # W» m # A O A m ^ • «0 M O « 0k * ? S 5 S 5 V i m t m ! fi L L L 1 w #4 V I ri et d r d d L ' L L II 1 ^ t «0 « «4 « o « V 0k^ ^ 9 o p^4 # m 0 5 #4 # M * M O mS » #4 p# S ^ vk Ok^ #A 0 0 0 001 0 i o m e !# O ^ O m O k0 et 0 o #4 pt Ok O d * et did d f ♦ « « « 1 • « M m etm # et9» -i*# et 10 « ê 5 -• S — m -• 0 s.- is is is 888:88 838388 c f! M # m 1 et 1 et 1 d d d ’ d L ' ♦ w ; ^ * * • Ï f 3 — 2V *•« S •— »# «a.# e|m Ok# ^ #4 «ib « 1- 0 « o • et m m m et f» # m0 N m m 1 1 88888: O m O k0 O V o 0 o m o m e d d d *4 s s * a a i d1 r et 1 r et 1 M • -4b C «P k k ♦ O • 1 1 a: : : : o o o s S 2 2 « « 10 8 8 : I m m m I S i!i a g 8 i Ï i i i iiî p4 s « 1 2 s a M s < m c! lii 1 _ _ _ 1 1 S S 9 a |T M M 2 0 0 0 i i 2 & u 8 2 o u u 2 U V o ill Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 155 collinearity. All condition index numbers are found to be less than 10. Hence, it is concluded that the parameter estimates are unbiased. The estimated coefficients, F statistic probabilities, and the R"s of regression equation (5.23) for each sample for the three types of firms for the three different windows are displayed in Table 5 .1 1 . The table is organized in three panels: Panel A portrays the estimates and statistics of the full sample, while panels B and C show the same for dividend-increasing and dividend decreasing subsamples, respectively. In Panel A of Table 5.11, the full sample has a positive and significant coefficient for DIVYDIFF and negative and significant coefficient for RATIOBAR, as reported already In the previous sections. The MKTVAL estimate is significant for the 7-day window only. For dividend-decreasing firms, none of the variables have significant coefficients. This seems to be consistent with the JW model since dividend-decreasing firms may not signal in a separating signalling equilibrium. Panel B of the table demonstrates negative and significant coefficient estimates for RATIOBAR and MKTVAL. Nevertheless, DIVYDIFF is significant only for the 1 -day window, as a result of controlling for the magnitude of the dividend changes. The striking evidence reflected in the table is that the signs of the MKTVAL estimates are negative for dividend-increasing firms. This is completely unexplainable by the JW model. JW suggests that when the firm announces a dividend payment to signal, the magnitude of the dividend and the market value of the firm impact positively on the stock price.'' The inescapable conclusion is that the JW model cannot be supported by empirical evidence. Stock price reactions to dividend announcements are not too well explained by the market value of the firm’s equity, dividends, and cash demand the JW model, the market value of the firm is endogenously determined when dividend payments affect the stock price. In order to verify the results of Table 3.11, the market value of the firm at the announcement day is estimated using the stock price. With these estimated values, in place of the actual values, the tests in this section are repeated. Results identical to those of Table S.11 are obtained with those estimated values. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. CD ■D O Q. C g Table 5.12$ Teate of Liquidity belated Variables by Pire Sise, Squat loo (5.23) Q. CPB(1)£ - 5o + InlMRTVALj^) + 63 DIVDIPPj^ + 63 SATIOBAR^ Ind.pand.nt Quintll. 1 Qulntil. 2 Qulntil. 3 Qulntil. 4 Oiintlli "O V.riabl.a (Sa.ll) (Largi CD Pan.1 A. Qulntil.a Groupad by INVI8T C/) C/) mmciPT .0026 .0101 .0014 .0099 .0055 «0 (0.90) ( 3 . 3 9 ) » (0.54) (3.20)" (1.68) ln(MXTVAL| .0002 -.0009 .00001 -.0009 -.0003 «1 (0.59) (-2.15)' (0.04) (-1.92) (-0.86) olvioirr .4596 .0687 .3665 .6893 .1039 8 «2 (1.94) (0.65) (4.44)" (5.81)" (3.02)' ■D RAtlOWUl -.4778 -.3579 .0260 -.4145 -.0308 (-2.51)* (-2.29)' (-0.19) _ «3 (-1.60) (0.18) .0039 .0028 .0074 .0152 .0036 Prob. of r .0140 .0600 .0002 .0001 .0200 Panel B. Quintiles Grouped by BQUBUY CD INTIRC8PT .0050 .0089 .0028 .0076 .0021 «0 (1.64) ( 3 . 1 2 ) " (0.99) (2.33)' (0.57) 3. In(NKTVAL) -.0002 -.0008 -.00005 —.0006 .00001 3" «1 (-0.58) (-1.84) (-0.13) (-1.19) (-0.02) CD DIWDIPP .6327 .7967 .4812 .9981 .0343 CD «2 ( 2 . 5 8 ) " ( 5 . 2 1 ) " (5.66)" (7.35)" (1.23) ■D RATIOBAR -.3151 -.1543 -.1186 -.3320 -.0241 O Q. , *3 (-2.15)' (-0.86) (-0.73) (-1.48) (-0.13) C R* .0042 .0116 .0120 .0211 .0006 a Pcob. of P .0100 .0001 .0001 .0001 .6740 O 3 "O Panel C. Qulntil.. Groupad by DBBTRBTI o IRTRRCBPT .0073 .0034 .0086 .0058 .0151 «0 (2.53)' (1.34) ( 2 . 6 1 ) " (1.83) ( 3 .72)" CD In(HXTVAL) -.0003 -.00005 -.0011 -.0005 -.0014 Q. «1 (-0.89 ) (-0.14) (-2.09)' (-1.21) ( - 2.85)" DIVTDIPP .3276 .9376 .5494 .5731 .0592 *2 (1.94) (7.67)" ( 3 . 6 1 ) " ( 5 . 9 2 ) " (1.81) RATIOBAR -.3874 -.3836 -.1120 -.2070 -.1754 (-1 02) (-2.27)' (-0.57) (-1.55) (-0.81) ■D , S3 CD R* .0030 .0239 .0063 .0141 .0042 Prob. of r .0440 .0001 .0007 .0001 .0100 C/) C/) (XI O) 157 «A # g M* # N M O Mp ^O #4 *4 ^m e m *»O# o P P gSg^^OP^pp Ü s s i s s s sS83 — O m o m o M M I o M # O O I m / o # # 2 N # M 10*4 « ^ O *»10 # o O«O*4in«0f«l*«*4Q OpoPr»r»mpor# ^ o ^ p ^ N ^ r» o*40*4io««4eioo PPPMNPPPP N I M « I fH I 25 M m # m «o M N«P«O«0i 0*4 P * P ^ V *»*4*fc^O ^ ^ m ^ ^ ^ o i #4 o PPP«P«m^*4p MPOPPPPPNO PMO*OI»#»WOPP PPPPPPOOPO P # I N m I M P O t in « 10 I ^ P P m M p « *4 <4 0 ^ M o I PPPPPPP^O#425 <^o ^P ^ 9 —*M p 838:: 838: 0PP#4(4M*40PM *••••••••• * fî I* H o o M I M . ' ? ■ ■ p # p m # M # ^ o ^ ^p ^p *»p P ^« OP ^ w P e ^ _ 18 o p Q ^ p o il 838S3Ü!SS8 8 O <4 O P P P 38 & n I *4 Üî ti« «I w %# 0 îîiî ni t fg •-S •'* 2 ^ li ! g î H 1 i!i Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 158 from the firm (or liquidity demand by current shareholders). Stated differently, the theoretical equilibrium relationship described by equation (5.2) empirically does not hold. As it was indicated earlier, the results of likelihood ratio tests are evidence that the JW 's dividend signalling model is inconsistent with the data and is possibly misspecified. The results of the cross sectional analyses give additional weight to this argument. b. Controlling for Liquidity Factors In addition to the liquidity demand by current shareholders, cash demand from the firm may be influenced by investment, repayment of debt, repurchase of shares, and funding increases in short-term working capital. These other liquidity factors may impact the analysis in two ways. It may have an effect on the cross-sectional relation between dividend levels, liquidity demands, and firm size. It may also provide biased estimates of coefficients. In addition, in a Modigliani and Miller (1958) world, the firm's external financing is construed as the residual of its investment decision. Since levels of investment vary across firms, external financing also varies cross-sectionally. It follows that differences in external financing across firms must be isolated. To control for liquidity-related factors and to verify the robustness of the estimator function, the regression of the previous section is repeated for each quintile grouped by INVEST, EQUBUY. DEBTRETI, WORKCAP, and DEBTISSU, respectively. The parameter estimates for each of these regressions are reported in Table 5.12. Comparisons between five quintiles and with the previous results indicate that the estimates between quintiles are at variance with each other. In the interest of brevity, the results are given only for the regression in which CPE(1)'s are used as the dependent variable." The figures shown in the table imply that estimates obtained from the linear model might be biased. Accordingly, the robustness of the tests of the previous section can be scrutinized. Clearly, more ^^The cegreseione using CPE(7) and CPE(5) result in similar estimates. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 159 3 : <0 r» S I 8 3 « P i o d d d d d d S i I i i o i I 8 d d g Q o o * I A o 8 8 OOP i s i f i isl 38 S 8 M8 383 8:8388 sssfss o -M 0o -M 0o -M e S s 8 § 8 8 U 8 u 8 S s 8 S I O 9 9 I 4 4 •4 i 85 & I I i i ie PI lA 4 !l 4 4 4 O O e S» I o m Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 160 rigorous testing of the effects of the liquidity-related variables on the cross-sectional model is required. An additional set of tests of the liquidity-related variables (INVEST, EQUBUY, DEBTRETI, WORKCAP, and DEBTISSU) is performed, applying a two-step procedure. At the first step, the error terms are estimated from the cross-sectional regression equation (5.23). At the second step, the estimated error terms are regressed on each of the liquidity-related variables. The null hypothesis is that the errors obtained at the first step are unrelated to each one of the explanatory variables of the second step. The results of this last set of tests are reported in Table 5.13. The estimated coefficients of the liquidity-related variables are insignificant in each secondary regression except for two cases: regressions using CPE(I) show significant estimates for INVEST and DEBTRETI at the 0.0 5 level. Overall, it is safe to conclude that the cross- sectional model of equation (5.23) appears to be robust enough to portray the firm's liquidity-related variables. F. Chapter Summary and Conclusions In this essay, the dividend signalling model developed by John and Williams (1985) is empirically tested. First, it is examined whether the demand for cash from the firm motivates the payment of dividends as JW, or Feldstein and Green (1983) suggest. JW assert that, in equilibrium, larger dividends are paid only if the demand for cash from the firm is high. It is found that there is a negative relation between liquidity demands and dividend payments, inconsistent with the JW contention. Instead, it seems that the empirical evidence supports Feldstein and Green's reasoning that firms have a strong desire to satisfy liquidity demands of their current shareholders by paying dividends. It is documented here that liquidity demand by shareholders is one of the determinants of the firm's dividend policy. Second, stock price and volume reactions to quarterly dividend announcements are studied. It is observed that announcements of dividend increases bid up the stock Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 161 price while trading volume decreases. In contrast, announcements of dividend decreases reduce both stock price and trading volume. The JW model posits that higher (lower) dividends are associated with higher (lower) stock price and trading volume and dividends can be used as a signal (see also Miller's (1987) signalling conditions). Hence, the findings of stock price and volume reaction to dividend announcements are not consistent with some of the basic relations of the JW model. Third, it is also discovered that the positive relationship between dividend changes and price changes is significant for smaller firms and insignificant for larger ones. This may imply that a signal price relationship exists for small firms, but not for large firms. At best, the indication is that the signal price relationship is cross- sectionally negatively related to firm size. This finding is not consistent with the JW model's notion that dividends and the market value of the firm combine to increase stock prices positively at the dividend announcement date. The likelihood ratio tests indicate that the cross-sectional model of JW [equation (5.2)1 is inconsistent with the data or empirically may be misspecified. Fourth, it is found that the polynomial form for the signalling-pricing function fits the data very well. This polynomial pattern is neither consistent with the monotonie relationship between signal and price that many authors assume, nor compatible with the non-linear relationship that JW posits. Fifth, the key argument that JW assert to be empirically testable is examined. It is this argument around which are developed the tests of the cross-sectional model that relate the dividend announcement effects, the market value of the firm, dividends, liquidity demand by current shareholders, and the cash demand from the firm. A cross- sectional regression analysis shows that stock price reactions to dividend announcements are not explained by the JW model (and that the latter is also inconsistent with the data). Despite all the empirical evidence reported here, a caveat is in order. It is quite possible that some of the conclusions reached here may hinge on measurement Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 162 problems. Unfortunately, these problems will exist in all empirical work, making the verifiability (or perhaps, more important, the refutabilityl of a theory impossible. Notwithstanding, while scores of studies focus on formulating theoretical models of signalling, empirical research is a must for the testing of the validity of these models. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Chapter 6. SUMMARY, CONCLUSIONS, AND FUTURE RESEARCH The objective of this dissertation has been to extend empirical work on dividend policy in three ways. The dissertation encompasses three essays.' The first essay empirically investigated a wide range of dividend determinants including investments, earnings, debt, free cash flows, firm size, beta, and industry classification. This essay employed time series cross sectional tests and vector autoregression (VAR) methodology to analyze the explanatory power of the various theories of dividend policy. The second essay proposed an alternative approach to the ex dividend anomalies that fully incorporates both the tax effect hypothesis and the short-term trading hypothesis. This essay investigated some observable differences between cum and ex-dividend efficient portfolios based on a portfolio selection model in the context of the 1984 and 1986 tax reforms. The third essay analyzed the empirical validity of a dividend signalling equilibrium model and tested the explicit relationship among the announcement effect, the dividends, and the cum-dividend market values, suggested by a signalling equilibrium model. In Chapter 3, time-series cross-sectional regressions using an error components model were estimated to test for the contemporaneous relationships among theoretical dividend determinants. It was found that a firm's dividend payments are significantly related to earnings, free cash flows, beta, and firm sizes. Perhaps the most important finding is that dividends are not influenced by investments and debts, consistent with the irrelevance proposition of Miller and Modigliani (1961). It was also shown that industry effects might not exist or might be only the manifestation of firm size effects. 'These essays did not converge to certain unified conclusions because they empirically examined separate dividend theories based on different assumptions and models. 163 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 164 On the other hand, in order to examine the dynamic relationships among these variables, the vector autoregression analyses were conducted. The variance decompositions and impulse response analyses consistently demonstrated that dividends have significant short-term effects on itself, whereas other variables have no dynamic effects on dividends. It appears that only dividends have short memory of its own past and that other variables have no dynamic interactions with dividends. Based on these results, a lagged dividend model was developed and tested for robustness. The evidence suggests that dividend payments are heavily influenced by a series of the past dividend payments as well as contemporaneous factors such as earnings and free cash flows. Although this lagged dividend model may be rather exploratory, it will shed some light on the model of puzzling dividend policy. One fruitful avenue for future study may be to apply a structural vector autoregression which simultaneously incorporates both contemporaneous variables and dynamic factors. Chapter 4 attempted to report changes in the risk-expected return trade-off around the ex-dividend day by comparing the efficient portfolio frontiers between the pre-tax reform period and the post-tax reform period. These portfolio results were related to the excess return behaviors on which the 1984 and 1986 Tax Reform Acts are also documented to have had effects. First, the effects of the 1984 and 1986 Tax Reform Acts on the return behaviors around the ex-dividend day were documented. The results are consistent with both the tax-effect hypothesis and the short-term trading hypothesis, and they are summarized as follows: (1) prior to the 1986 tax reform, significant positive excess returns were observed on the ex-dividend day; (2) during the post-1986 tax reform period, average ex-dividend excess returns were not different from zero at the 1 % significance level; (3) there existed the negative correlationships between the dividends yields and ex-dividend day excess returns, which correlationships decreased from the pre-1986 tax reform Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 165 period to the post-1986 tax reform period; (4) average ex-dividend day excess returns increased significantly during the post-1984 tax reform period compared to the pre-1984 Tax Reform period; (51 during the post-1984 tax reform period, more significant excess returns were observed for 11 trading days before and after the ex-dividend day compared to the pre-1984 tax reform period. Second, the cum- and ex-dividend efficient portfolios between the pre-1984 tax reform period and the post-1984 tax reform period were compared. The following evidences supporting the short-term trading hypothesis were found: ID cum-dividend efficient portfolios dominated ex-dividend efficient portfolios; and (2) the efficiency gap between cum-dividend efficient frontiers and ex-dividend efficient frontiers expanded more during the post-1984 tax reform period than during the pre-1984 tax reform period. Following Long's (1977) argument, it was predicted that the 1986 Tax Reform Act would decrease the expected after-tax return at the new tax rates and increase the after-tax variance of the portfolio. Consistent with this expectation, it was also found that the efficient portfolio frontier during the post-1986 tax reform period was dominated by the efficient portfolio frontier during the pre-1986 tax reform period. This result may be interpreted as the tax-effects since the tax premiums decrease with an decrease in dividend tax rate. But, the variance as risk factor was not considered in the tax-effect hypothesis. To conclude, the evidence of excess return behaviors supported the existence of both tax effects and short-term trading around the ex-dividend day. It appeared that these tax effects and short-term trading are not mutually exclusive, but co-existent as long as short-term trading is tax-induced. And this finding was effectively supported by portfolio test results. It was argued that portfolio approach is quite advantageous over the extant approaches, because this approach incorporates the tax effect as well as Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 166 short-term trading. It was noted that ex-dividend behavior cannot be understood independently of the risk-expected return trade-off. It was also proposed that this portfolio approach can best be used in many branches of corporate finance (e.g., turn-of- the-year effect I where portfolio revisions are considered to occur. In Chapter 5, the dividend signalling model developed by John and Williams (1985) was empirically tested. It was examined whether the demand for cash from the firm motivates the payment of dividends (see John and Williams (1985) or Feldstein and Green (1983)]. It was found that there is a negative relation between liquidity demands and dividend payments, inconsistent with the JW contention. The empirical evidence supported Feldstein and Green's reasoning that firms have a strong desire to satisfy liquidity demands of their current shareholders by paying dividends. Stock price and volume reactions to quarterly dividend announcements were studied. It was observed that announcements of dividend increases bid up the stock price while trading volume decreases. In contrast, announcements of dividend decreases reduce both stock price and trading volume. The JW model posits that higher (lower) dividends are associated with higher (lower) stock price and that trading volume and dividends can be used as a signal. Hence, the findings of stock price and volume reaction to dividend announcements are not consistent with some of the basic relations of the JW model. It was also found that the polynomial form for the signalling-pricing function fits the data very well. This polynomial pattern was neither consistent with the monotonie relationship between signal and price that many authors assume, nor was it compatible with the non-linear relationship that JW posits. The key argument that JW assert to be empirically testable was examined. The cross-sectional model that relates the dividend announcement effects, the market value of the firm, dividends, liquidity demand by current shareholders, and the cash demand from the firm were tested. The regression analysis showed that stock price reactions to Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 167 dividend announcements were not explained by the JW model. The likelihood ratio tests also indicated that the cross sectional model of JW is inconsistent with the data or empirically may be misspecified. While scores of studies focus on formulating theoretical models of signalling, empirical research is a must for the testing of the validity of these models. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. BIBLIOGRAPHY Acharya, S.. 1988, "A Generalized Econometric Model and Tests of a Sionallinc Hypothesis with Two Discrete Signals," Journal o f Finance 43:413-429. Aharony, J., and I. Swary, 1980, * Quarterly Dividend and Earnings Announcements and Stockholders' Returns: An Empirical Analysis," Journal o f Finance 35:1-12. Akerlof, George A., 1970, "The Market for Lemon: Quality Uncertainty and the Market Mechanism," Quarterly Journal o f Economics (August), 488-500. Ambarish, R., K. John, and J. Williams, 1987, "Efficient Signaling with Dividends and Investment," Journal of Finança 42:321-344. Asquith, P., and D. Mullins, 1983, "The Impact of Initiating Dividend Payments on Stockholders' Wealth," Journal o f Business 56:77-96. ______, 1986, "Equity Issues and Offering Dilution," Journal o f Financial Economics 15:61-89. Bamber, L.S., 1986, "The Information Content of Annual Earnings Releases: A Trading Volume Approach," Journal o f Accounting Research 24:40-56. Banz, R., 1981, "The Relationship between Return and Market Value of Common Stocks," Journal of Financial Economics 9:3-18. Barclay, Michael J., 1987, "Dividends, Taxes and Common Stock Prices: The Ex- Dividend Day Behavior of Common Stock Prices Before the Income Tax," Journal o f Financial Economics 19:31-44. Barnea, Amir, Robert A. Haugen, and Lemma W. Senbet, 1985, Agency Problems and Financial Contracting, Englewood Cliffs, NJ: Prentice-Hall. Barry, C., and S. Brown, 1984, "Differential Information and the Smail Firm Effect," Journal o f Financial Economics 13:283-294. Bar-Yosef, Season, Jeffrey L Callen, and Joshua Livnat, 1987, "Autoregressive Modelling of Eamings-lnvestment Causality," Journal o f Finance 42:11-28. , and Lucy Huffman, 1986, "The Information Content of Dividends: A Signalling Approach," Journal of Financial and Quantitative Analysis 21:47-58. Baskin, Jonathan, 1989, "An Empirical Investigation of the Pecking Order Hypothesis," Financial Management (Spring), 26-35. Beaver, W.H., 1968, "The information Content of Annual Earnings Announcements," Journal o f Accounting Research Supplement 6:67-92. Belsley, D.A., E. Kuh, and R E. Welsch, 1980, Regression Diagnostics: Identifying Influerrtial Data and Sources of Multicollinearity, New York: Wiley. 168 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 169 Bhattacharya, Sudipto, 1979, "Imperfect Information, Dividend Policy and the Bird In the Hand Fallacy," B eilJoum tl o f Economics 10:259-270. , and Jay R. Ritter, 1983, "Innovation and Communication: Signalling with Partial Disclosure," Review of Economic Studies 5:331-346. Black, Fischer, 1976, "The Dividend Puzzle," Journal of Portfolio Management (Winter), 5-8. Black, Fischer, and Myron Scholes, 1974, "The Effects of Dividend Yield and Dividend Policy on Common Stock Prices and Returns," Journal o f Financial Economics 1:1-22. ______, 1973, "The Behavior of Security Returns around Ex-dividend Days," Working Paper, Universitv of Chicago. Blanchard, O.J., 1989, "A Traditional Interpretation of Macroeconomic Fluctuations," American Economic Review 79:1146-1164. Blazenko, George W., 1987, "Managerial Preference, Asymmetric Information, and Financial Structure," Journal of Finance 42:839-862. Bolster, Paul J., Lawrence B. Lindsey, and Andrew Mitrusi, 1989, "Tax-Induced Trading: The Effect of the 1986 Tax Reform Act on Stock Market Activity," Journal of Finance 44:327-344. Booth, L.D., and D.J. Johnston, 1984, "The Ex-Dividend Day Behavior of Canadian Stock Prices: Tax Changes and Clientele Effects," Journal o f Finance 39:457- 476. Brennan, Michael J., 1970, "Taxes, Market Valuation and Corporate Financial Policy," National Tax Journal 23:417-427. , and Alan Kraus, 1987, "Efficient Financing Under Asymmetric Information," Journal of Finance 42:1225-1243. , and Anjan V. Thakor, 1990, "Shareholder Preferences and Dividend Policy," Journal o f Finance 45:993-1018. Brown, Keith C., and Scott L. Lummer, 1984, "The Cash Management Implications of a Hedged Dividend Capture Strategy," Financial Management (Winter), 7-17. Brown, Stephen, and Jerold B. Warner, 1985, "Using Daily Stock Returns: The Case of Event Studies," Journal o f Financial Economics 14:3-31. Clements, M.P., and G.E. Mizon, 1991, "Empirical Analysis of Macroeconomic Time Series," European Economic Review, 887-932. Campbell, John Y., and Robert J. Shiller, 1987, "Cointegration and Tests of Present Value Models, " Journal of Political Economy 95:1062-1088. Campbell, T., 1979, "Optimal Investment Financing Decisions and the Value of Confidentiality, " Journal of Finandal Quantitativa Analysis 14:913-924. Crockett, Jean, and Irwin Friend, 1988, "Dividend Policy in Perspective: Can Theory Explain Behavior?" Review of Economics and Statistics (November), 603-613. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 170 Crutchley, Claire E., and Robert S. Hansen. 1989, "A Test of Agency Theory of Managerial Ownership, Corporate Leverage, and Corporate Dividends,”Financial Management (Winter), 36*47. DeAngelo, Harry, and Ronald W. Masulis, 1980, "Leverage and Dividend Irrelevancy Under Corporate and Personal Taxation,” Journal o f Finance 35:453-467. Denis, David J., 1990, "Defensive Changes in Corporate Payout Policy: Share Repurchases and Special Dividends,” Journal o f Finance 45:1433-1456. Dhrymes, P.J., and M. Kurz, 1964, "On the Dividend Poiicy of Electric Utilities,” Review o f Economics and Statiitics (February), 76-81. , 1967, "Investment, Dividend, and External Finance Behavior of Firms,” In R. Ferber, ed.. Determinants of Investment Behavior, New York, 427-467. Diamond, D., 1984, "Financiai intermediation and Delegated Monitoring,” Review o f Economic Studies 51:393-414. Dyl, E.A., and J.R. Hoffmeister, 1986, "A Note on Dividend Policy and Beta,” Journal of Business Finance and Accounting 13:107-115. Eades, Kenneth M., Patrick J. Hess, and E. Han Kim, 1984, "On Interpreting Security Returns During the Ex-Dividend Period,” Journal of Financial Economics 13:3- 34. Easterbrook, F., 1984, "Two Agency Cost Explanations for Dividend Payments,” American Economic Review 74:680-689. Eckbo, B., V. Maksimovic, and J. Williams, 1990, "Consistent Estimation of Cross- Sectional Models in Event Studies,” Review of Financial Studies 3:343-365. Elton, Edwin J., and Martin J. Gruber, 1970, "Marginal Stockholder Tax Rates and the Clientele Effect,” Review of Economics and Statistics 52:68-74. ______, 1978, "Taxes and Portfolio Composition,” Journal o f Financial Economics 6:399-410. ______, and Joel Rentzler, 1984, "The Ex-Dividend Behavior of Stock Prices; A Re-Examination of the Clientele Effect: A Comment,” Journal o f Finance 39:551- 556. Engle, Robert P., and Byung Sam Yoo, 1987, "Forecasting and Testing in Co-Integrated Systems,” Journal of Econometrics 35:143-159. , and Clive W. J. Granger, 1987, "Cointegration and Error Correction: Representation, Estimation, and Testing,” Econometrics 55:251-276. Fame, Eugene F., 1974, "The Empirical Relationships Between the Dividend and Investment Decisions of Firms,” American Economic Review 64:304-318. , 1980, "Agency Problems and the Theory of the Firm,” Journal o f Political Economy 88:288-307. , and H. Babiak, 1968, "Dividend Policy: An Empirical Analysis,” Journal o f American Statistical Association 63:1132-1161. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 171 , and Michael C. Janaen, 1983, "Agency Problems and Residual Claims," Journal o f Law and Economies 26:327*349. , Lawrence Fisher, Michael C. Jensen, and Richard Roll, 1969, "The Adjustment of Stock Prices to New Information," Intarnational Economic Raviaw 10:1*21. Fedenia, Mark, and Theoharry Grammatikos, 1991, "Portfolio Rebalancing and the Effective Taxation of Dividends and Capital Gains Following the Tax Reform Act of 1966," Journal o f Banking and Finança 15:501*519. Feldstein, M., and J. Green, 1963, "Why Companies Pay Dividends," Amarican Economic Review 73:17*30. Frankfurter, George, and William Lane, (1990), "The Rationality of Dividends," Unpublished working paper, Louisiana State University. Friend, I., and J. Hasbrouck, 1966, "Determinants of Capital Structure," Research in Finance 7:1*19. Fuller, W. A., 1976, Introduction to Statistical Tima Saries, New York; John Wiley. , and G. E. Battese, 1974, "Estimation of Linear Models with Crossed Error Structure," Journal of Econometrics 2:67*76. Garcia, Beatrice E.. 1966, "Trading to Nab Dividends Captures Investors' Fancy," Waii Street Journal, May 19, 1966. Ghosh, Chinmoy, and J. Randall Woolridge, 1991, "Dividend Omissions and Stock Market Rationality," Journal o f Business Finance & Accounting 16:315*330. Goldfeld, S. M., and R. E. Quandt, 1976, Nonlinear Methods in Econometrics, New York: North*Holland Inc. Gordon, Myron, 1963, "Optimal Investment and Financing Policy," Journal o f Finance (May), 264*272. Graham, Benjamin, and David Dodd, 1951, Security Analysis; Principles and Technique, New York: McGraw-Hill. Grammatikos, Theoharry, 1969, "Dividend Stripping, Risk Exposure, and the Effect of the 1964 Tax Reform Act on the Ex-Dividend Day Behavior," Journal of Business 62:157*173. Green, J., 1960, "Taxation and the Ex-dividend Day 6ehavior of Common Stock Prices," NBER Working Paper 496. Grinblatt, Mark S., Ronald W. Masulis, and Sheridan Titman, 1964, "The Valuation Effects of Stock Splits and Stock Dividends," Journal o f Financial Economics 13:461*490. Grundy, B., 1965, "Trading Volume and Stock Returns around Ex-dividend Dates," Working Paper University of Chicago. Harris, L., 1966, "Cross-Security Tests of the Mixture of Distributions Hypothesis," Journal of Financial and Quantitative Analysis 21:39*46. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 172 Harris, Milton, and Artur Raviv, 1986, ”A Sequential Signalling Model of Convertible Debt Call Policy,” Journê! o f Finança 40:1263*1281. Hess, Patrick J., 1982, "The Ex Dividend Day Behavior of Stock Returns: Further Evidence on Tax Effects,” Journal o f Finança 37:445-456. Higgins, R.C., 1972, "The Corporate Dividend-Saving Decision,” Journal o f Financial and Quantitative Analysis 7:1527-1541. Holtz-Eakin, D„ W. Newey, and H. S. Rosen, 1988, 'Estimating Vector Autoregressions with Panel Data,” Econometrics 56:1371-1395. Jensen, M.C., 1986, "Agency Costs of Free Cash Flow, Corporate Finance and Takeovers,” American Economic Raviaw 76:323-339. , and W. Meckling, 1976, "Theory of the Rim: Managerial Behavior, Agency Costs, and Ownership Structure,” Journal o f Financial Economics 3:305-360. John, K., and J. Williams, 1985, "Dividends, Dilution, and Taxes : A Signaling Equilibrium,” Journal o f Finance 40:1053-1070. , and A. Kalay, 1985, 'Informational Content of Optimal Debt Contracts,” In E. Altman and M. Subrahmanyam, eds.. Recent Advances in Corporate Finance, Homewood, IL: Irwin, 133-161. , and L.H.P. Lang, 1991, "Insider Trading around Dividend Announcements: Theory and Evidence,” Journal of Finance 46:1361-1389. Kalay, Avner, 1982, "The Ex-Dividend Day Behavior of Stock Prices: A Re-Examination of the Clientele Effect,” Journal of Finance 37:1059-1070. , 1984, "The Ex-Dividend Day Behavior of Stock Prices: A Re-Examination of the Clientele Effect: A Reply,” Journal of Finance 39:557-561. Kalay, A., and U. Loewenstein, 1985, "Predictable Events and Excess Returns,” Journal o f Financial Economics 14:423-449. Karpoff, Jonathan M.. and Ralph A. Walkling, 1988, "Short-Term Trading around Ex- Dividend Days: Additional Evidence,” Journal o f Financial Economics 21:291- 298. ______, 1990, "Dividend Capture in NASDAQ Stocks,” Journal o f Financial Economics 28:39-65. Keating, J. W., 1990, "Identifying VAR Models Under Rational Expectations,” Journal of Monetary Economics 25:453-476. Keim, Donald 8., 1985, "Dividend Yields and Stock Returns: Implications of Abnormal January Returns,” Journal o f Financial Economics 14:473-489. Kmenta, Jan., 1986, Elements o f Econometrics, New York: Macmillan. Kumar, P., 1988, "Shareholder-Manager Conflict and the Information Content of Dividends,” The Review of Financial Studies 1:111-136. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 173 Laderman, Jeffrey M., 1985, ”Capture>the-Oividend: The Street's Hottest Game,’ Business Week, December 2, 1985, 129-130. Lakonishok, Josef, and Theo Vermaelen, 1983, 'Tax Reform and Ex-Dividend Day Behavior,” Journal o f Finance 38:1157-1179. ______, 1986, 'Tax-Induced Trading around Ex-Dividend Days,' Journal o f Financial Economics 16:287-319. Lamoureux, Christopher G., 1990, 'Dividends, Taxes, and Normative Portfolio Theory,' Journal o f Economics and Businass 42:121 -131. Lang, Larry H.P., René M. Stulz, and Ralph A. Walkling, 1991, 'A Test of the Free Cash Flow Hypothesis: The Case of Bidder Returns,' Journal o f Financial Economics 29:315-335. Lang, Larry H.P., and Robert H. Litzenberger, 1989, 'Dividend Announcements: Cash Flow Signalling vs. Free Cash Flow Hypothesis?' Journal o f Financial Economics 24:181-192. Lastrapes, William D., and Faik Koray, 1990, 'International Transmission of Aggregate Shocks under Fixed and Flexible Exchange Rate Regimes: United Kingdom, France, and Germany, 1959 to 1985,' Journal of Intarnational Money and Finance 9:402-423. Laub, M„ 1972, 'Some Aspects of the Aggregates Problems in Dividend Earnings Relationship,' Journal of Amarican Statistical Association, 792-805. Lehn, K., and A. Poulsen, 1989, 'Free Cash Flow and Stockholder Gains in Going Private Transactions,' Journal o f Finance 44:771-787. Leland, Hsyne E., and David H. Pyle, 1977, 'Informational Asymmetries, Financial Structure, and Financial Intermediation,' Journal o f Finance 32:371-387. Lintner, J., 1956, "Distribution of Incomes of Corporations among Dividends, Retained Earnings and Taxes,' American Economic Review 46:97-113. Litterman, R.B., 1986, 'Forecasting with Bayesian Vector Autoregressions: Five Years of Experience,' Journal o f Business and Economic Statistics 4:25-38. Litzenberger, Robert H., and Krishna Ramaswamy, 1979, 'The Effect of Personal Taxes and Dividends on Capital Asset Prices,' Journal of Financial Economics 7:163- 195. ______, 1980, "Dividends, Short Selling Restrictions, Tax-Induced Investor Clienteles and Market Equilibrium,'Journal o f Finança 35:469-482. ______, 1982, 'The Effects of Dividends on Common Stock Prices: Tax Effects or Information Effects?” Journal o f Finance 37:429-443. Long, John S., Jr., 1977, 'Efficient Portfolio Choice with Differential Taxation of Dividends and Capital Gains,” Journal o f Financial Economics 5:25-53. Lupoletti, W.M., and R.H. Webb, 1986, ”Defining and Improving the Accuracy of Macroeconomic Forecasts: Contributions from a VAR Model,” Journal of Business 59:263-285. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 174 Markowitz, H. M„ 1952, "Portfolio Selection," Journal o f Finance 7:77-91. , 1959, "Portfolio Selection, Efficient Diversification of Investments," Cowles Foundation Monograph 16. , 1987, Maan-Variance Analysis in Portfolio Choice and Capital Markets, New York: Basil Blackwell. , and A.F. Perold, 1981, "Sparsity and Piecewise Linearity in Large Portfolio Optimization Problems," In Sparse Matrices and Their Uses, New York: Academic Press, 89-108. Masulis, R., and A. Korwar, 1988, "Seasoned Equity Offerings: An Empirical Investigation," Journal o f Financial Economics 15:91 -118. McCabe, G.M., 1979, "The Empirical Relationship Between Investment and Financing: A New Look," Journal of Financial and Quantitative Analysis 14:119-135. McMillin, W. Douglas, and Randall E. Parker, 1990, "Federal Debt, Tax-Adjusted q, and Macroeconomic Activity," Journal o f Money, Credit, and Banking 22:100-109. Michaely, Roni, 1991, "Ex-Dividend Day Stock Price Behavior: The Case of the 1986 Tax Reform Act, Journal o f Finance 48:845-859. Michel, A., 1979, "Industry Influence on Dividend Policy," Financial Management (Autumn), 22-28. , and I. Shaked, 1988, "Country and Industry Influence on Dividend Policy: Evidence from Japan and the U.S.A.," Journal o f Business Finance and Accounting 13:365-381. Mikkelson, W., and M. Partch, 1988, "Valuation Effects of Security Offerings and the Issuance Process," Journal o f Financial Economics 15:31 -80. Miller, Merton H., 1988, "Behavioral Rationality in Finance: The Case of Dividends," Journal o f Business (October), 451-488. Miller, M., 1987, "The Information Content of Dividends," In Dornbusch, Fisher and Bossons (éd.). Macroeconomics and Finance: Essays in Honor o f Franco Modigliani, Cambridge, MA: MIT Press, 37-58. Miller, Merton H., and Franco Modigliani, 1981, "Dividend Policy, Growth, and the Valuation of Shares," Journal o f Business 34:411 -433. , and K. Rock, 1985, "Dividend Policy under Asymmetric Information," Journal o f Finance 40:1031 1051. , and Myron S. Scholes, 1978, "Dividends and Taxes," Journal o f Financial Economics 8:333-384. ______, 1982, "Dividends and Taxes: Some Empirical Evidence," Journal of Political Economy 90:1118-1141. Modigliani, Franco, and Merton H. Miller, 1958, "The Cost of Capital, Corporation Finance and the Theory of Investment," American Economic Review 48:281- 297. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 175 Mundlak, Yair, 1978, "On the Pooling of Time Series and Cross Section Data," Econometric» 46:69-85. Myers, B.C., 1964, "The Capital Structure Puzzle," Journal o f Finance 39:575-592. , and N.S. Majluf, 1964, "Corporate Financing and Investment Decisions When Firms Have Information that Investors Do Not Have," Journal o f Financial Economics 13:167-221. Narayanan, M.P., 1966, "Debt Versus Equity Under Asymmetric Information," Journal o f Financial and Quantitative Anaiysis 23:39-51. Ofer, Aharon R., and Daniel R. Siegel, 1967, "Corporate Financial Policy, Information, and Market Expectations: An Empirical Investigation of Dividends," Journal of Finance 42:669-911. , and Anjan V. Thakor, 1967, "A Theory of Stock Price Responses to Alternative Corporate Cash Disbursement Methods: Stock Repurchases and Dividends," Journal o f Finance 42:365-394. Partington, G.H., 1965, "Dividend Policy and Its Relationship to Investment and Financing Policies: Empirical Evidence," Journal o f Business Finance and Accounting 12:531-542. Patell, J., 1976, "Corporate Forecasts of Earnings Per Share and Stock Price Behavior: Empirical Tests," Journal o f Accounting Research 14:246-276. , and M. Wolfson, 1964, "The Intraday Speed of Adjustment of Stock Prices to Earnings and Dividend Announcements," Journal o f Financial Economics 13:223- 252. Pechman, Joseph A., 1967, Federal Tax Policy, Washington, D C.: Brookings Institution. Penman, S., 1963, "The Predictive Content of Earnings Forecasts and Dividends,” Journal of Finance 36:1161 -1199. Pettit, R. R., 1972, "Dividend Announcements, Security Performance, and Capital Market Efficiency," Journal o f Finance (December), 993-1007. Pincus, M., 1963, "Information Characteristics of Earnings Announcements and Stock Market Behavior," Journal o f Accounting Research 21:155-163. Pogue, G. A., 1970, "An Extension of the Markowitz Portfolio Selection Model to Include Variable Transactions' Costs, Short Sales, Leverage Policies and Taxes,” Journal o f Finance 25:1005-1027. Poterba, James M., and Lawrence H. Summers, 1964, "New Evidence That Taxes Affect the Valuation of Dividends," Journal o f Finance 39:1397-1415. Richardson, G., S. E. Sefcik, and R. Thompson, 1966, "A Test of Dividend Irrelevance Using Volume Reaction to a Change in Dividend Policy," Journal o f Financial Economics 17:313-333. Riley, J., 1979, "Informational Equilibrium," Econometrics 47:331-359. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 176 Robin, Ashok .1., 1991, "The Impact of the 1986 Tax Reform Act on Ex-Dividend Day Returns,” Finêneiê! M anagtm tnt (Spring), 60-70. Ross, S.A., 1973, "The Economic Theory of Agency: The Principal's Problem," American Economic Review 63:134-139. , 1977, "The Determination of Financial Structure : The Incentive-Signalling Approach," BettJournal of Economics 8:23-29. Rozeff, M.S., 1982, "Growth, Beta and Agency Costs As Determinants of Dividend Payout Ratios," Journal o f Financial Research 5:249-259. Runkle, David E., 1987, "Vector Autoregressions and Reality," Journal o f Business & Economic Statistics 5:437-454. Sharpe, William P., 1963, "A Simplified Model for Portfolio Analysis," Management Science 9:277-293. Shaw, Wayne H., 1991, "An Examination of Ex-Dividend Day Stock Price Movements: The Case of Nontaxable Master Limited Partnership Distributions,"Journal of Finance 46:755-771. Sims, C., 1980, "Macroeconomics and Reality," Econometrica 48:1-49. Smirlock, M., and W. Marshall, 1983, "An Examination of the Empirical Relationship between the Divid^rd grd Investment Decisions: A Note " ■i'».r'nal o f Finance 38:1659-1667. Spence, M., 1973, "Job Market Signaling," Quarterly Journal o f Economics 87:355- 379. , 1974, "Competitive and Optimal Response to Signals : An Analysis of Efficiency and Distribution," Journal of Economic Theory 7:296-332. Spencer, David E., 1989, "Does Money Matter? The Robustness of Evidence from Vector Autoregressions," Journal of Money, Credit, and Banking 21:442-454. Talmor, Eli, 1985, "Personal Tax Considerations in Portfolio Construction: Tilting the Optimal Portfolio Selection," Quarterly Review o f Economics and Business 25:55-71. Thakor, Anjan V., 1989, "Strategic Issues in Financial Contracting: An Overview," Financial Management (Summer), 39-58. Wallace, T.D., and Ashiq Hussain, 1969, "The Use of Error Components Models in Combining Cross Section with Time Series Data," Econometrica 37:55-72. Walters, James E., 1956, "Dividend Policy and Common Stock Prices," Journal o f Finance (March), 29-41. Wansley, J.W., and W.R. Lane, 1987, "A Financial Profile of the Dividend Initiating Firm," Journal o f Business Finance and Accounting 14:425-436. Watts, R., 1973, "The Information Content of Dividends," Journal o f Business (April), 191-211. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 177 White, Halbert, 1980, "A Heterotkedasticity-Consistent Covariance Matrix Estimator and a Direct Test for Heteroskedaaticitv," Econonwtrie* 48:817 838. Williams, J., 1988, "Efficient Signallino with Dividends, Investment, and Stock Repurchases," Journal of Finança 43:737-747. Zeghal, D., 1983, "Firm Size and the Informational Content of Financial Statements," Journal o f Financial Economics 12:299-310. Zivney, Terry L , and Michael J. Alderson, 1986, "Hedged Dividend Capture with Stock Index Options," Financial Management (Summer), 5-12. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. APPENDIX A: THE RELATIONSHIP BETWEEN DIV AND FCASH In order to examine the contemporaneous relationship between free cash flows and dividend payments, the simple regression is first assumed as follows’: Y * X,^, + Xj^, + e, (A.1) where Y is DiV; X, is EARN; X, is FCASH; and e ~ N(O.o'). A two-step procedure is taken to estimate the coefficient of X,. The first step is to regress X% on X,. The estimated residual terms from this step is: Û = Xj - X, ♦ = X; - X,(X,'X,) ’X,'X, = ( I - X,|X,'X,) ’X,' I X, - M, X, , (A.2) where $ - |X,'X,) ’X,'X, M, - I ! - X,|X,'X,) ’X,' 1. The second step is to regress Y on the residual terms obtained from the first step. The resulting coefficient estimate of 0 from the second step is equal to the coefficient estimate of X, in (A.1). - (û'û)’û' Y - [ |M,X,)'|M,Xa) r' (M,Xj)' Y - ( X,'M, M,X, )•’ (X,'M,) Y. IA.3) ^The m a s u r e of free cash flows is estimated by the operating i n c o m minus i n c c m taxes, minus interest expenses, minus total amounts of dividends paid (see Section B.2 of Chapter 3). 178 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 179 Substituting X, > X, • Y into (A.3), ht - ( Xj'M, M,X, )•' (X, - V)' M, Y - [ X,'M, M,Xj )•’ X,' M, Y - ( X,'M, M,X, ) ’ Y'M, Y. (A.4I Since M, is an idempotent matrix, M,* » M,; and X,'M, - X /l I • X,(X,'X,) 'X,' I - X,' - X,'X,(X,'X,| ’X,' - X,' - X,' - 0 . Thus, ht - • (X,’M,X,1’ Y' M,Y < 0 . (A.5) Now, let X, - X, - Y - X„ where X, may be interest expenses, and substitute this constraint into (A.3). ht - I X,'M, M,X, ]•' (X, - Y - X,)' M, Y = (Xj'M, ' X,' M, Y - (Xj'M, M,Xjl ’ Y'M, Y • (X/M, M,X,1’ X,'M, Y. = - (X,'M, M,X,I ’ Y'M, Y - [X,'M, M,X,) ’ X,'M, Y. (A.6) The sign of the first term in equation (A.6) is always negative, but the sign of the second term depends on the correlation between X, and Y. Then, it is not easy to predict the sign of Therefore, it is evident that, as more terms such as taxes are added to the constraint, the sign of h i will be affected by the complex correlations between the dependent variable and those elements in the constraint. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. APPENDIX B: THE EFFECTS OF 1986 TRA A taxable firm aligibla for tha 85% dividend income deduction could engage In short-term trading by buying cum-dividend and selling ex-dividend if the following tax- induced trading return is positive. -fgu. +D[l-(l-0.85)T^] -Ccd-Tq,), (S.l) where Ted ie the marginal tax rate on corporate dividends received, T„ is the marginal tax rate on corporate capital gains, and C, is the round trip transaction costs. Before the 1986 Tax Reform Act, the short-term trading return for these taxable corporations could be expressed as ♦(Pe^-P^)x0.28 -Cgll-O.Ze) . (B.2) Under the 1986 Tax Reform Act, the corporate dividend deduction percentage was reduced to 70%, and corporate dividend and capital gains tax rates were equalized to 34% (see Table 4.1). Thus, the trading return for short-term traders could be changed to (1-0-70) XO. 34] 34-0,(1-0.34) . (B.3) Now, the tax-induced short-term trading returns between the pre-1986 tax reform period and the post-tax reform period are compared. It is noted that the coefficients of capital gains increased from 0.28 to 0.34, whereas those of dividends 180 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 181 decreased from 0.931 to 0.898. In addition, the weights that transaction costs take out of the unit trading return decreased from 0.72% to 0.66%. Thus, by introducing transaction costs, it seems to be inconclusive whether the 1986 Tax Reform Act offered taxable corporations more incentives to exploit short-term trading around the ex-dividend day. Security dealers and brokers also have similar incentives to exploit short-term trading. Denier^ and brokers are subject to the same tax rates on dividends and capital gains. They could undertake short-term trading by buying cum-dividend and selling ex- dividend if the following trading return is positive. *(PoM-P,ri r„-Ce(l-r„) , (B-4) where T„ is the corporate income tax rate. During the pre- and post-1986 tax reform period, the short-term trading return for dealers could be written as ( 1 -0.46) + (f X O .46-Cg( 1 -0.46) . (B.5) ( 1 -0 .34) * ( XO . 34-C,( 1 -0.34) . (B.6) While the coefficient of dividends increased from 0.54 to 0.66, the factor of capital gains decreased from 0.46 to 0.34. Also, the burden incurred by transaction costs has increased from 0.54 to 0.66. Thus, as in the case of taxable corporations, it is not clear whether the 1986 Tax Reform Act provided securities dealers with incentives for tax-induced short-term trading. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. APPENDIX C: SIGNALLING MODEL' A representative firm is modeled. Corporate management in the best interests of current shareholders can make an investment of I, and pay a dividend of D. Funds for both investment. I, and dividends, D, are obtained from either internally retained cash, C, or net new shares of stock, N, which are issued at the ex dividend price per share, p„. The constraint for a firm's sources and uses of funds is: D + I - C + p„N. (C.1I Suppose that Q is the number of shares outstanding before N shares are sold. At the ex dividend day, current shareholders receive the dividend per share, (1-tlD/Q, where t is a single tax rate on dividends. In equilibrium, the cum-dividend price per share, p, is equal to the sum of the ex-dividend price por share, p^, and the after tax dividend per share, (1-t)D/Q: P * P., + <1-t) D/Q. (0.2) Assume that management has strong incentives to signal for the firm's stock to command a higher price in the market. This signalling is generally induced when the firm issues new shares or current shareholders sell some fraction of their outstanding shares. Suppose that the liquidity demand from the firm by its current shareholders is L. Current shareholders can sell some portion of their shares either cum or ex-dividend to satisfy their consumption needs. They are supposed to sell M outstanding shares to new investors at the ex-dividend day. The liquidity supplied to these current shareholders is measured by; L « D p„ M. (C.3) Management's objective is to maximize current shareholders' wealth. This can be achieved by maximizing current shareholders' cash inflow, (1-t)D + p,M. Since the ^Appendix c proves the signalling model of John and Williams (1985). 182 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 183 current shareholders' equity in the firm is diluted after the sale of shares on both personal and corporate account, the fractional equity of current shareholders is (Q-M) / (Q + N), where M outstanding shares and N new shares are sold to new investors. Therefore, the true present value of current shareholders' remaining equity is X(Q-M)/(Q+N), where X is the present value of the future cash inflow (X is inside information, known to insiders but not observed directly by outsiders). Consequently, management's job is to select the optimal dividend, DIX), and net new funds, p, N(X), which maximize the firm's present value to its current shareholders. Q-M M ax{ |1-t)D + p „M + X }. (0.4) D, p„N Q + N Substituting the above constraints, (C D through (C.3), into the maximand (C.4), we get the equivalent maximand: From (C.3), p„M - L • D. From (C.l), P „ N - D + I -C. From (C.2), pQ = p„Q + (1-t)D, P * p„Q + (1-t)D, p«,Q ■ P • (1-t)D. p„Q - p„M “ IP -(1-t)D ] - [ L-D] = P- D + tD-L+D “ P + to - L. P«Q + P«N “ IP -(1 -tD ) ] + I D +1 -C l = P - D +tD +D +1 -C “ P + tD + I - C. P«Q • P«M p„ (Q-M) P + tD - L p„Q + p„N p« (Q+N) P + tO + I - C Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 184 Thus, Q-M U ID, P, X) - |1-t)D + p„M + Q + N P + to - L |1-t)D + (L-D) + ------X P + tD + I - C P + tD - L L-tD + ------X. (C.5) P + tD + I - C In short, management's problem simplifies to deciding the optimal dividend DIX). Max U |D,P,X). IC.6) D >0 If the firm's private attribute, X, were known to outsiders, then a market price, P, would be equal to its true value, V. Substituting the value V into the maximand IC.5), we get V + tD - L V = I L- tD) + ------X V + tD + I - C V * I V + tD + I - C ) - IL - tD ) * I V + tD + I -C) + I V + tD - L) X V» + I tD + I -C) V I L-tD) V + IL-td) * I tD + I - Cl + V X + I tD - L) X V* + [ iD + I - C - X - IL-td) 1 V - IL-tD)* ( tD + I -C-Xl V* - [ IL-tD) + IC + X-l-tD) ] V + IL-tD)* I C + X-l-tDJ - 0 { V - IL-tD) } { V - 1C + X - 1 -tD )) = 0 Thus, V ID.XI » PIDIXI) - C + X - 1 - tD. IC.7) Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 185 Oifferentiating the maximand (C.5) with respect to the aggregate, cum-dividend price, P, we obtain IP+tD-U'IP+tD + l-C) ■ (P+tD-U(P + tD + l-C)' ( P + tD + I - O * P + tD + l - C - P-tD + L ( P + to + I - O* l-C + L Up * ------— — X . (C.8) I P + to + I - o * This means that, whenever C < L + I, U, > 0 . That is, management's maximand increases in the price P. When liquidity demand from the firm on personal account, L, and corporate account. I, is greater than the firm's internal supply of cash, C, a stock price, P, increases, benefiting current shareholders who sell their shares. Differentiating the maximand (C.5) with respect to the dividend, D, we obtain (P(D)+tD-Ll'(P(D)+tD + l-C) - (P(D)+tD-U(P(D) + tD + l-C)' Un ■ -t + ---- P(D) + tD + I - C ]* [P'(D)+t](P(D)+tD + l-C] - (P(D) +tD-LI[P'(D) + t) [ P(D) + tD + I - C )* [ P' + 1 1 [ I - C + L ] Uo = -t + ------X ( P + tD + I - C P When Uo = 0, L + I -C t . ( t + P' ) X. (C.9) ( P + tD + I - C )* This implies that management determines dividend such that the marginal benefit of dividend to current shareholders may be equal to its marginal cost. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 186 Now, we substitute (C.7) into the first-order condition (C.9) for the announcement effect. From (C.9), L + l-C L + l-C t . t ------X + P '------X (P + tD + l-O* (P+tD + l-O* L + l-C L + l-C P'------X - t [1 ------X ] (P+tD + l-O* (P+tD + l-O* P " (L + I-OX = t [ (P + tD + l-O* • (L + l-C) X 1 Since (P+tD + l-C) = X in (C.7), P " (L + l-C) X - t [ X*-(L + l-C) X ) - fX*[X-(L+l-C) 1 P'*(L +1-0 - 11 X-(L + l-C) I - t [ (P + tD +1-0-( L + l-C) ) » t ( P + tO - L 1 ( P + tD - L ) P' - t ------(L + l-C) Thus, the dividend announcement effect is PID(X)1 + tD(X) - L P'[D(X)1 - t ------. (C.10) Finally, for the dividend signalling function, we substitute both (C.7) and the derivative of (C.7) with respective to X into (C.l). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 187 From (C.7), PID(X)j - C + X - 1 -tD. P'[D(X)1* D IX) - 1 - tO'IX). 1 - tD'IX) P'(D(X)) ------□MX) From (C.10). 1 ■ tD'(X) [C + X-l-tD(X)] + tO(X) - L D'(X) I ■ C + L (1-tD') (l-C+ L) = tD'* ( (C + X-l-tD) +tD-L] (1-tD') (l-C + L) - t D'* [ C + X - 1 - L ) ( l-C + L ) - t D * ( (l-C + L) + (C + X-l-L) ) - tD * X. l- C + L D' . (C.11) tX Integrating (C.l 1) subject to the condition (C.9) gives the optimal dividend condition. 1 D(X) = — Max (l-C + L, 0) Ln(X). (C .l2) t Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. VITA Jaisik Gong has been a doctoral candidate in Finance at Louisiana State University from 1988 to 1992. His dissertation directed by Or. George M. Frankfurter encompasses three leading issues in the anomaly of dividend policy: (1 ) the determinants of dividend policy; (2) dividends, taxes, and portfolio choices; and (31 empirical tests of dividend signalling. While pursuing his Ph.D. in Finance, he taught undergraduate courses including investments and worked on financial databases at LSU. Immediately prior to joining LSU, he served in the Korea Development Bank, Seoul, as bank economist from 1984 to 1987 and conducted research in Euro-bonds, syndicated loans, and international banking. He earned his M.B.A. in Accounting from Seoul National University in 1984 and received the B.A. degree from Seoul National University in 1981. 188 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. DOCTORAL EXAMINATION AND DISSERTATION REPORT Candidate: Jalslk Gong Major Field: Business Administration (Finance) Title of Diaaertation: Three Essays in Dividend Policy Dean of the Grafluaiiuate School ITTEE: Date of Examination: May 20, 1992 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.