Open Collections

O W N E R S H I P S T R U C T U R E A N D C O R P O R A T E D I V I D E N D P O L I C Y By Savita Verma A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF T H E REQUIREMENTS FOR T H E D E G R E E OF DO C T O R OF PHILOSOPHY in T H E FACULTY OF GRADUATE STUDIES C O M M E R C E AND BUSINESS ADMINISTRATION We accept this thesis as conforming to the required standard T H E UNIVERSITY OF BRITISH COLUMBIA 1990 © Savita Verma, 1990 In presenting this thesis in partial fulfilment of the requirements for an advanced degree at the University of British Columbia, I agree that the Library shall make it freely available for reference and study. I further agree that permission for extensive copying of this thesis for scholarly purposes may be granted by the head of my department or by his or her representatives. It is understood that copying or publication of this thesis for financial gain shall not be allowed without my written permission. Department of Commerce and Business Administration The University of British Columbia 1956 Main MaU Vancouver, Canada Date: Abstract This study investigates the potential role of ownership structure as a determinant of the corporate dividend policy. A firm's dividend policy is modelled as the outcome of a voting game among groups of asymmetrically informed shareholders, who also have different marginal tax rates for dividend income. The outcome of the voting game is determined by the relative voting powers of these shareholder groups. Voting power is denned as the probability that a particular block of shares will be pivotal in determining the outcome of the voting game. Using Shapley values as instruments for shareholder groups' voting powers, data on firms which traded on the Toronto Stock Exchange during the 1976-88 period are em-ployed to test the model's predictions. Table of Contents Abstract ii List of Tables vi List of Figures x Acknowledgement xi 1 I N T R O D U C T I O N 1 2 C O R P O R A T E D I V I D E N D POLICIES: T H E O R I E S A N D E V I D E N C E 4 2.1 Dividend policy under differential taxation of dividends and capital gains 6 2.1.1 Dividends, taxes, and firm valuation 6 2.1.2 Existence of tax-based clienteles 9 2.2 Dividend policy as a signal of firm value under asymmetric information 12 2.2.1 Dividend signalling models 13 2.2.2 Empirical evidence on dividend signalling 15 2.3 Dividend policy as a vehicle for minimizing agency costs 18 2.4 Summary 19 3 P O T E N T I A L D E T E R M I N A N T S OF O W N E R S H I P S T R U C T U R E 20 3.1 Value of Control 21 3.2 The Mechanics of Shareholder Voting 23 ii i 3.2.1 Potential problems with voting by proxy 25 3.3 Ownership structure and firm valuation 28 3.4 Shareholder voting behaviour: empirical evidence 30 3.5 Proxy contests 32 3.6 Summary 33 4 A M O D E L OF T H E C O N S E N S U S D I V I D E N D P O L I C Y 34 4.1 The consensus dividend policy 37 4.2 Properties of the Dividend Equilibrium 44 4.3 Summary 48 5 D A T A , M E T H O D O L O G Y A N D E M P I R I C A L R E S U L T S 49 5.1 Data sources and sample characteristics 49 5.1.1 Sample size 49 5.1.2 Sample distribution by year, industry and corporate ownership . 50 5.1.3 Ownership of voting rights by managers 52 5.1.4 Constraints on dividend payments 53 5.1.5 . Asset size and capital structure 54 5.1.6 Aggregate dividend yields 54 5.2 Test strategy 56 5.2.1 Seemingly Unrelated Regressions analysis 57 5.2.2 Accounting for Dmax and Z?m , n 58 5.2.3 Linear structural model (LISREL) formulation 64 5.3 Empirical results: SUR analysis 69 5.3.1 Total sample 69 5.4 Empirical results: LISREL formulation 71 5.4.1 Total sample 71 iv 5.4.2 Cases with absolute voting power 75 5.4.3 Cases with non-absolute voting power 78 5.5 Summary and conclusions 80 6 C O N C L U S I O N S 120 A p p e n d i c e s 124 A T A X A T I O N O F I N V E S T M E N T I N C O M E 124 A . l The Canadian Tax System 124 B S H A P L E Y V A L U E I N O C E A N I C G A M E S 128 B . l Distribution of Voting Power Among F i rm Shareholders 128 B.2 Assumptions Underlying Shapley Value 132 B.3 Shapley value and the dividend voting game 134 B i b l i o g r a p h y 136 v List of Tables 5.1 Data sources and sample size 81 5.2 Annual distribution of sample firms classified by country of control, 1976-1988 82 5.3 Distribution of sample firms by industry classification as of 1988 83 5.4 Corporate ownership of voting rights in sample firms under domestic con-trol, 1976-88 84 5.5 Corporate ownership of voting rights in sample firms under foreign control, 1976-88 85 5.6 Managers' ownership of voting rights in sample firms, 1978-82 86 5.7 Constraints on dividend payments imposed by debt covenants in the sam-ple firms 87 5.8 Capital structure of sample firms under domestic control, 1976-88 88 5.9 Capital structure of sample firms under foreign control, 1976-88 89 5.10 Annual dividend yields on firms under domestic control, 1976-86 90 5.11 Annual dividend yields on firms under foreign control, 1976-86 91 5.12 SUR estimates of regressions of cash (DIVC) and non-cash (DIVS) div-idend yields on percent ownership of voting rights by large corporate shareholders (IOWN), managers (MOWN), IOWN-sqrd, MOWN-sqrd and IOWNxMOWN 92 5.13 SUR estimates of regressions of cash (DIVC) and non-cash (DIVS) divi-dend yields on voting powers of the large corporate shareholders (ISHAP) and managers (MSHAP) 93 vi 5.14 Distribution of the 1188 sample observations by zero- and non-zero divi-dend yields and by absolute- and non-absolute voting power 94 5.15 Descriptive statistics for the two voting power variables ISHAP and M S H A P , the two dividend yield variables DIVC and DIVS, eleven financial variables NE, SE, TA, CA, C L , LD, P E , STDR and M V E , and their correlations. 95 5.16 Pearson correlation coefficients for dividend yield variables, financial vari-ables, and voting power variables 96 5.17 LISREL measurement model (design matrix A) — Total sample and sub-sample of cases with non-absolute voting power 97 5.18 LISREL measurement model (design matrix A) using ISHAP — Cases with absolute voting power 98 5.19 LISREL measurement model (design matrix A) using M S H A P — Cases with absolute voting power 99 5.20 SUR estimates of regressions of cash (DIVC) and non-cash (DIVS) divi-dend yields on voting powers of the large corporate shareholders (ISHAP) and of the managers (MSHAP) and the financial variables N E , L D I V C , SE, TA, P E , STDR and M V E 100 5.21 OLS regression estimates of parameters for calculating Dmax and D m , „ , based on 353 absolute voting power cases i.e. cases for which either ISHAP=1 or MSHAP=1 101 5.22 Coefficients of OLS regressions of DIVC on estimated Dmax, estimated Dmin, and shareholder voting powers (ISHAP and M S H A P ) for the 170 cases of non-absolute voting power (neither ISHAP nor M S H A P has value 1) 102 5.23 Estimates of the LISREL measurement model (design matrix A) for the total sample 103 vii 5.24 E s t i m a t e s of the L I S R E L s t r u c t u r a l equat ions (design m a t r i x T) for the t o t a l sample 104 5.25 E s t i m a t e of the L I S R E L var iance-covar iance m a t r i x of the unobservable a t t r ibu tes (design m a t r i x $ ) for the t o t a l sample 105 5.26 E s t i m a t e s of the n o r m a l i z e d residuals S — XJ for L I S R E L for the t o t a l sample . 106 5.27 E s t i m a t e s of the L I S R E L measurement m o d e l (design m a t r i x A ) us ing I S H A P for cases w i t h absolute v o t i n g power 107 5.28 E s t i m a t e s of the L I S R E L s t r u c t u r a l equat ions (design m a t r i x T) us ing I S H A P . for cases w i t h absolute v o t i n g power 108 5.29 E s t i m a t e of the L I S R E L var iance-covar iance m a t r i x of the unobservable a t t r ibu tes (design m a t r i x $ ) us ing I S H A P for cases w i t h absolute v o t i n g power 109 5.30 E s t i m a t e s of the n o r m a l i z e d residuals S — £ for L I S R E L us ing I S H A P for cases w i t h absolute v o t i n g power 110 5.31 E s t i m a t e s of the L I S R E L measurement m o d e l (design m a t r i x A ) us ing M S H A P for cases w i t h absolute v o t i n g power I l l 5.32 E s t i m a t e s of the L I S R E L s t r u c t u r a l equat ions (design m a t r i x T) u s ing M S H A P for cases w i t h absolute v o t i n g power 112 5.33 E s t i m a t e of the L I S R E L var iance-covar iance m a t r i x of the unobservable a t t r ibu tes (design m a t r i x $ ) us ing M S H A P for cases w i t h absolute v o t i n g power 113 5.34 E s t i m a t e s of the n o r m a l i z e d residuals S — £ for L I S R E L us ing M S H A P for cases w i t h absolute v o t i n g power 114 5.35 E s t i m a t e s of the L I S R E L measurement m o d e l (design m a t r i x A ) for cases w i t h non-abso lu te v o t i n g power 115 v m 5.36 Estimates of the LISREL structural equations (design matrix T) for cases with non-absolute voting power 116 5.37 Estimate of the LISREL variance-covariance matrix of the unobservable attributes (design matrix for cases with non-absolute voting power. . 117 5.38 Estimates of the normalized residuals S — S for cases with non-absolute voting power 118 B.39 Examples of Shapley Values for Given Ownership Structures 131 ix List of Figures 4.1 Shareholder Marginal Tax Rates 38 4.2 The dividend voting game 40 x Acknowledgement I wish to express my sincere thanks to Professor B. Espen Eckbo for the encouragement and guidance he has provided throughout the course of this work. I have benefitted greatly from the challenges he has posed and from the criticisms he has offered. I would also like to thank Professor Alan Kraus for the many helpful suggestions and comments, for prompt feedback, and for his doing so with a very supportive touch. Professor Piet de Jong has helped keep the perspective. His comments and feedback on the editorial as well as the statistical aspects of this work, are greatly appreciated. Finally, I would like to acknowledge my appreciation of the stimulating work environ-ment provided by colleagues in the doctoral program and the Finance Seminar, in the Faculty of Commerce and Business Administration at the University of British Columbia. Savita Verma March, 1990 xi Chapter 1 I N T R O D U C T I O N This study investigates the potential link between a firm's dividend policy and its owner-ship structure. The proposed link is a simple one: Firm policies over which shareholders disagree, get resolved at meetings through the outcome of a vote. K they face different marginal tax rates for dividend income, then shareholders will favour different dividend policies and will vote for the one which maximizes their net-of-taxes income. The extant literature suggests several explanations as to why firms pay dividends in spite of the apparent tax penalty to their shareholders: the existence of shareholder tax clienteles, information signalling through dividends under asymmetric information, and dividends as vehicles for minimizing agency costs. The explanation of dividend payments proposed here assumes the coexistence of different shareholder tax-clienteles in a firm's ownership structure and arrives at an equilibrium dividend policy which is determined by the outcome of a vote held among the firm shareholders. This is in contrast to the explanations available in the literature which treat shareholders alternatively as unanimous, or as passive investors who either do not have a say in setting firm policies, or choose not to exercise it. In the proposed model, some shareholders of a firm hold sizeable blocks of votes, with the remaining votes are distributed among a large number of individual shareholders. It is assumed that some of the large block-holders do not pay any taxes on dividends. The rest of the shareholders — the remaining large block-holders as well as the small shareholders, are taxed on their dividend receipts although at different rates. Also, it is assumed that 1 Chapter 1. INTRODUCTION 2 the distributions of relative tax rates of shareholders are only imperfectly known. Based on their relative tax rates for dividends and capital gains firm shareholders can be divided into two groups: one group prefers dividends to capital gains and the other group prefers capital gains to dividends. The distribution of votes among the two shareholder groups also implies a certain distribution of voting power. Voting power here means the ex ante probability of winning a vote for a group of shareholders. Given the voting power distribution for a particular firm, it is shown that in equilibrium its dividend policy has to coincide with the one that would emerge if a vote were to be held among the firm shareholders. This leads to the central proposition that corporate dividend policy is in part determined by the relative voting powers of the shareholder groups in different tax brackets. The above proposition and other implications of the proposed model are tested using data on firms which traded on the toronto Stock Exchange during the 1976-88 period. The period covers part of the tax regime in Canada when taxes on capital gains were determined as set down by the 1972 tax reform legislation. Prior to 1972, capital gains were not taxed at all. Since 1985, individuals have a life time tax exemption on the first $100,000 of their capital gains. Thus in the period prior to 1972 and since 1985, the implied tax penalty of receiving dividends rather than capital gains would seem to be higher compared to that during the 1972-85 period. Decisions by investors and policies of firms which may be sensitive to differential taxation of dividends and capital gains during the 1972-85 period, are likely to be even more so during the other two regimes. The rest of this study is organized into five chapters. Chapter 2 contains a discussion of the existing theories of corporate dividend policies and the available empirical evidence. As this literature overview suggests, neither the theoretical models nor the available empirical evidence have put the dividend puzzle to rest. Chapter 1. INTRODUCTION 3 The model being proposed in this study treats a firm's ownership structure as exoge-nous. The potential determinants of the ownership structure are nevertheless of interest in their own right.1 In Chapter 3, the various theories and available evidence on the determinants of ownership structure, and how they relate to the proposed model, are discussed. In Chapter 4, the proposed model linking ownership structure with corporate dividend policy is presented. The assumptions underlying the model and its implications are discussed and the hypotheses to be tested axe set up. Chapter 5 contains the details of the data gathered, the sample characteristics and the results of the empirical analysis. A measure employed to proxy for shareholder voting power is Shapley value. Tables containing data description and analysis are included at the end of the chapter. Chapter 6 — the last chapter, contains conclusions drawn from the results and details of the related future work planned. Two appendices are included. Appendix A contains a description of the Canadian tax code with a focus on the differences in the tax regimes in effect over the years. Appendix B contains a discussion of oceanic games and Shapley value as a measure of shareholder voting power in these games. 1In a full equilibrium characterization, an investor might arrive at his optimal portfolio by simulta-neously controlling for his fractional ownership of the firm (and his implied voting power) and the firm's financial policies. The financial policies would then in turn, account for the shareholders' voting powers to incorporate the possibility of disagreement among them. The proposed partial equilibrium model is offered as a first step towards this ultimate goal. Chapter 2 C O R P O R A T E DIVIDEND POLICIES: T H E O R I E S A N D E V I D E N C E The determinants of corporate dividend policies constitute one of the central unresolved issues in financial economics. While corporate managers tend to view the firm's payout decision as economically important,1 it remains a puzzle why firms continue to pay divi-dends, given the relative tax penalty on cash distributions. Miller and Modigliani (1961) have shown that in a perfect capital market with no transaction costs, agency costs, taxes or information asymmetries, the firm's choice of dividend policy does not affect the value of the firm. Their central insight is that, given these market conditions, individual shareholders can costlessly generate their own preferred dividend policy (say — by selling off shares), precluding a positive price for dividends generated at the corporate level. Following the Miller and Modigliani "dividend irrelevance proposition", theoretical as well as empirical research has appealed to various forms of market imperfections and the interdependence between dividends, capital structure and the firm's investment policy to explain the dividend puzzle. For instance, the tax codes of most countries favour capital gains over dividends as a source of income to individual investors. Not only is the nominal capital gains tax rate typically lower than the nominal rate applied to dividends; investors have the option to defer capital gains taxes indefinitely. Thus, paying cash dividends almost certainly imposes a relative tax penalty on a large number of individual shareholders.2 Nevertheless, while a substantial amount of research has been devoted 1Lintner (1956) presented some first systematic evidence that corporate managers set target yield ratios as a function of expected earnings. 2This conjecture is supported by Feenberg (1981) who, based on a sample of 86,000 of the universe 4 Chapter 2. CORPORATE DIVIDEND POLICIES: THEORIES AND EVIDENCE 5 to the question of whether equilibrium security prices reflect this tax disadvantage,3 empirical tests do not reliably support the existence of tax effects in expected stock returns.4 Similarly, studies of the ex-dividend day price drop do not conclusively support the hypothesis that the marginal investor in the firm's stock faces a relative tax penalty on dividends.5 The literature also explores the role of agency costs,6 costs of financial distress and information asymmetries7 to explain the observed patterns in firm dividend policies. Dividend policy in these models is treated alternatively as a signal of the firm's future prospects or as a means by which managers minimize their own access to funds generated by the firm. By voluntarily contracting (via the announced dividend policy) to pay out the funds as dividends, the managers supposedly curb their own temptation to make low or negative net present value investments on behalf of the firm. While the above and other theories provide plausible explanations of the dividend de-cision at a detailed level, no consensus has emerged as to how they combine to explain the overall dividend picture. There does, however, appear to be widespread agreement that unanticipated changes in dividends convey new information to the market and therefore affect market prices.8 In the following, we look at the theories and evidence referred to above in greater of 86 million U.S. federal tax returns in 1977, shows that the total dividend reported for taxation was 27 billion dollars, resulting in over 8 billion dollars in taxes. 3 E.g., Brennan (1970), Long (1977), Litzenberger and Ramaswamy (1980), Shaefer (1982), Constan-tinides and Ingersoll (1984), and Dybvig and Ross (1986). 4 E.g., Black and Scholes (1974), Litzenberger and Ramaswamy (1979,1982), Morgan (1980), Miller and Scholes (1982), and Hess (1983). 5See, e.g. Elton and Gruber (1970), Kalay (1982), Hess (1982), Booth and Johnston (1984), Eades, Hess and Kim (1984), Barone-Adesi and Whaley (1986), and Barclay (1987). 6 E.g. , Jensen and Meckling (1976), Smith and Warner (1979), Easterbrook (1984), Schleifer and Vishny (1986). 7 E.g., Ross (1977), Bhattacharya (1979), Miller and Rock (1985), Ambrish, John and Williams (1987). 8 E.g., Charest (1978), Asquith and Mullins (1983), Grinblatt Masulis and Titman (1984), Eades, Hess and Kim (1985), Kalay and Loewenstein (1985), and Handjinicolaou and Kalay (1984). Chapter 2. CORPORATE DIVIDEND POLICIES: THEORIES AND EVIDENCE 6 details. The focus is on the insights provided by them as well as the aspects in which they fall short of resolving the dividend puzzle. This critical survey of the literature is conducted under the following three headings: • Dividend policy under differential taxation of dividends and capital gains. • Dividend policy as a signal of firm value under asymmetric information. • Dividend policy as a vehicle for minimizing agency costs. 2.1 Dividend policy under differential taxation of dividends and capital gains We start with a look at those investigations in the literature which focus on differen-tial taxation of dividend and capital gains incomes and different marginal tax rates for investors as the determinants of dividend policy. These investigations can be broadly classified into two groups. One group consists of those investigations which hypothesize that investors require a premium from high dividend yield stocks as compensation for the tax penalty incurred. The other group includes explorations which take the view that dividend policy may be irrelevant to firm valuation, but that existence of tax-based clienteles may nevertheless govern firms to set dividend policies in a non-trivial manner. 2.1.1 Dividends, taxes, and firm valuation Brennan's (1970) general equilibrium model was the first one to be set in an economy with taxes and in which dividend income is taxed at a rate higher than the capital gains tax rate. In such an economy, investors designing portfolios to maximize expected utility of wealth require a higher rate of return from higher dividend yield stocks for a given level of risk. Brennan' model is thus an extended C A P M with dividend yield constituting a second factor, the first one being the systematic risk of the security. Chapter 2. CORPORATE DIVIDEND POLICIES: THEORIES AND EVIDENCE 7 In an early test of Brennan's model, Black and Scholes (1974) found that the risk-adjusted returns of two portfolios — one of high dividend yield stocks and the other of low yield stocks, do not differ significantly from each other. In other words, they reject the hypothesis that investors require differential returns from stocks with different dividend yields. To explain why this could happen despite differential taxation, Miller and Scholes (1978) argued that firms or investors ought to be able to neutralise the tax penalty on dividends through the use of appropriate investment strategies. The firms could do so by paying dividends through stock repurchases rather than as cash payouts. And the investors may neutralize the dividend tax penalty through "dividend laundering". Dividend laundering involves (i) levering the purchase of equities so as to offset taxable dividends with interest deductions, and (ii) removing unwanted risk in this levered position by purchasing tax-deferred insurance. An assumption underlying this strategy is that interest payments on borrowings for investment can be used to reduce taxable income from other investments. In practice however, it may not be so straight forward to undo the firm dividend policy in this manner either by the firm itself or by its shareholders. Firms would need to repurchase stock on the open market9 in order to keep the tax authorities from treating the proceeds to sellers as taxable dividends. This would force the amount paid out in dividends to be determined by the prevailing market price of shares and therefore uncertain from period-to-period. In turn, this would take away from the amounts paid to shareholders through share repurchases, the one key feature identifying them as dividends — namely, the predictability of timing and amounts of dividend payments. Indeed the fact that managers can, and very likely do, control their size and timing 1 0 rules out 9 Any pro-rata repurchases of shares on a regular basis directly from the shareholders would be deemed taxable as ordinary dividends. 1 0 Evidence of favourable stock price reactions when firms make repurchase announcements, is discussed in the next section along with other models of dividends as signals of firm value. Chapter 2. CORPORATE DIVIDEND POLICIES: THEORIES AND EVIDENCE 8 repurchases as a mere means of making predictable, periodic distributions to shareholders. Furthermore, this would be true even if there were no tax-related complications associated with using repurchases as a vehicle for paying dividends. Shareholders, like firms, may not be able to undo a firm's dividend policy either because they may not be able to borrow at an interest rate no higher than the rate of return from the tax shelter.11 Besides, there is the question of how elastic the supply of tax-sheltered investment will be in equilibrium. In another test of the tax-adjusted C A P M , Litzenberger and Ramaswamy (1979) reported significant premia being charged for dividends in equilibrium stock returns. In a similar test using data on Canadian firms, Morgan (1980) found that prior to 1972, there were significant premia being charged for dividends, but that these premia disappeared for the 1972-80 period. 1 2 Morgan concluded that dividends and capital gains became substitutes, albeit imperfect, when capital gains taxes were introduced. Tests such as the above involving measurement of dividend yields and their premia have been strongly criticised however, by Miller and Scholes (1982) for their "inappropri-ate" yield measurements. Miller and Scholes note that the dividend yields are measured using the cash dividend amounts paid during the period over which the returns are calculated. These short-run dividend yields are then linked in these tests with the risk-adjusted returns which, in reality, may be compensation for the dividend announcement effect rather than the dividend tax burden, if any. Miller and Scholes suggest that yield calculations using only the dividends declared in advance, measure the appropriate ex-pected long-run dividend yield. This way, the dividend announcement effect is eliminated uFeenberg (1981) argues that less than 3 percent of the taxpayers in his 1977 sample of the US taxpayers could have made use of the dividend laundering scheme suggested by Miller and Scholes (1978). 1 2 Capital gains taxes were introduced in Canada in 1972. They were subsequently eliminated in 1985 for the first $100,000 in capital gains for individuals. For details of taxation in Canada, see Appendix A. Chapter 2. CORPORATE DIVIDEND POLICIES: THEORIES AND EVIDENCE 9 and only the long-term tax differential effect is measured. Thus modifying the dividend yield measurement, Miller and Scholes still do not find any premia for dividends. Confounding the debate further on whether dividends are undesirable or simply ir-relevant, is the study by Long (1978) of the Citizens Utilities case. Under a special IRS ruling, the firm has two classes of shares outstanding which are identical except in their dividend payouts. One class of shares pays only stock dividends which are not taxable as ordinary income under the IRS ruling. Shares in the other class pay only cash dividends. Shares in the former class are freely convertible into the latter type on a one-to-one basis, but not the other way around. Historically, the stock dividends have been higher than the cash dividends. Long shows that in this case the cash dividend shares seem to be preferred by investors in the sense that the required rate of return is lower for them than for the stock dividend shares. 2.1.2 Existence of tax-based clienteles The question of whether differential taxation affects investors' valuation of firms is unre-solved, as seen from the evidence so far. As Miller and Modigliani (1961) argued however, even if dividends were to be irrelevant to firm valuation in reality, this would not be incon-sistent with the existence of differential taxation-induced investor clienteles. Investors' relative tax rates would determine whether they hold high dividend yield or low-yield stocks. As far as firm valuation is concerned though, one clientele would be as good as any other. In equilibrium, some firms would pay higher and others lower dividends, but no firm would be able to increase its value relative to another firm by merely changing its dividend yield. H the aggregate dividends demanded changed due to some exogenous change such as a relative tax rate change, equilibrium would be reattained by a resorting of individuals among the various clientele groups rather than through a change in relative values of firms. Chapter 2. CORPORATE DIVIDEND POLICIES: THEORIES AND EVIDENCE 10 Empirical evidence supports the assertion noted above that the aggregate supply of dividends is affected by the differential in tax rates for dividends and capital gains. Khoury and Smith (1977) show that Canadian firms significantly increased their dividend payouts when a capital gains tax was introduced in Canada in 1972. The hypothesis that clienteles re-sort among themselves when firms change their dividend policies, is also supported by Richardson, Sefcik and Thompson's (1986) finding that for firms which announce dividends for the first time, both the trading volume and firm value go up significantly around the announcement date, and that the pattern of abnormal trading volume around the announcement date persists even after the announcement effect has been accounted for. In particular, this suggests that investors can have a preference for one or another dividend policy, its irrelevance to firm value notwithstanding. Elton and Gruber (1970) present a model in which the marginal tax rates of some traders would allow them to make arbitrage profits through trading around the ex-dividend dates.1 3 To eliminate these profits, the ex-dividend day drop in price for a stock must be of the magnitude which makes the marginal trader indifferent between selling the stock cum-dividend and ex-dividend. From the data on actual drops in prices when stocks go ex-dividend, the tax rates of the marginal trader can be inferred. Existence of different implied marginal tax rates for different dividend yield stocks then implies the existence of tax based dividend clienteles as suggested by Miller and Modigliani. Using data on US firms for the period April,1966-March, 1967, Elton and Gruber find that the implied marginal tax rate decreases when dividend payout increases. They conclude that this is evidence in support of the existence of tax-based clienteles. Elton and Gruber's interpretation is based on the assumption that the cum- and 1 3Certain jurisdictions such as UK, explicitly restrict tax arbitrage by individual investors and tax-exempt corporations through progressive taxation of trading proceeds, the closer the trading dates are to the ex-dividend dates. Based on the UK data, Poterba and Summers (1984) find premia being charged for dividends. Chapter 2. CORPORATE DIVIDEND POLICIES: THEORIES AND EVIDENCE 11 ex-dividend prices axe set by investors, who have decided to buy and sell shares around the dividend payment date for reasons other than dividend receipts and taxes. The tax-adjusted prices then emerge so as to make these investors indifferent between buy-ing/selling their shares cum-dividend and ex-dividend. Kalay (1982b) argues however, that additional information is required before the marginal tax rates can be inferred from calculations such as suggested by Elton and Gruber. Kalay (1982b) and Lakonishok and Vermaelen [(1983), (1986)] test the alternative hypothesis that the ex-dividend day prices are determined by the short-term traders whose tax rates are identical for dividend and capital gains incomes. They argue that these traders may make arbitrage profits by trading around the ex-dividend dates even though they axe taxed at the same rate for dividends and capital gains. Their evidence suggests that the short-term traders do have an impact on the ex-dividend day prices, thereby ruling out the possibility of inferring the marginal tax rates of the presumed tax clienteles.14 Barclay (1987) compares ex-day price behaviour of stocks before and after the enact-ment of the federal income tax in U S . 1 5 He finds that in the before-tax period, ex-day prices fell by the full amount of the dividend on the average, and that investors valued dividends and capital gains as perfect substitutes. This is different from the after-tax enactment evidence which suggests that investors discount taxable dividends more than they discount capital gains. The following overall picture thus emerges from the differential taxation-based models of dividends: If all investors are assumed to have the same marginal tax rate and if 1 4 Lakonishok and Vermaelen (1983) find that the ex-dividend day decline in price is smaller in Canada compared to the drop in US. For US firms, Barone-Adesi and Whaley (1983) find the ratio of the price decline to the amount of dividend paid, to be equal to unity. For Canadian firms, Booth and Johnston (1984) find it to be between zero and unity. Also they find the ratio to be monotonically increasing in dividend yield and attribute this link to another finding that the ex-dividend trading volume increases as dividend yield decreases. 1 5 Barclay covers the years 1900-1910 for pre-tax enactment and the years 1962-85 for the post-tax enactment data. The federal income tax was fully enacted in the US in 1913. Chapter 2. CORPORATE DIVIDEND POLICIES: THEORIES AND EVIDENCE 12 dividends axe taxed at a higher rate than capital gains, then the required rates of return reflect this tax penalty. If however, the investors are in different tax brackets and there are opportunities for tax arbitrage, then the classical tax clienteles emerge with low tax rate clienteles holding high dividend yield firms and high tax clienteles holding low dividend yield stocks. There are so far no pricing implications for the assets, though. But as Dybvig and Ross (1986) show, once short selling restrictions are introduced which curb tax arbitrage, then heterogenous marginal tax rates across investors can result in two or more distinct tax clienteles coexisting among a firm's shareholders. Full implications of this simultaneous existence of investors with different marginal tax rates among a firm's shareholders are yet to be explored in the literature. In the next section, we look at the literature which explores dividend policy as a potential signalling tool. 2.2 Dividend policy as a signal of firm value under asymmetric information The existence of the dividend announcement effect noted earlier suggests that dividends, and specially changes therein, have to be looked upon as more than just a means available to firms for distributing their cash flows based on the shareholders' tax rates. The early empirical model of dividends proposed by Lintner (1956) was based on his survey results that firm managers set dividends to gradually attain a target payout ratio over time, and that there is reluctance among them to cut dividends below their prevailing levels. The reluctance stems from the belief that a cut may be interpreted by investors as unfavourable information. More recently however, Kalay (1980) has argued that in the presence of bond covenants restricting dividend payments, cutting of dividends cannot be a decision made entirely at the discretion of the managers, and therefore cannot be interpreted as a signal. Kalay does not find a large enough percentage Chapter 2. CORPORATE DIVIDEND POLICIES: THEORIES AND EVIDENCE 13 of forced dividend cuts though, in order to be able to refute their information content. Further, in a subsequent study of the debt covenants of a random sample of firms, Kalay (1982a) finds that most firms pay dividends which are well below the maxima allowed by the covenant restrictions. Thus the debt covenants do not appear to impose a binding constraint on firm dividend payments. If dividends are in fact being used as signalling devices by firms, then Kalay's evidence noted above suggests that the observed dividend policies are possibly the outcomes of the cost-benefit trade-off from this signalling. We next look at the models in the literature which formalize this notion, followed by the available empirical in support thereof. 2.2.1 Dividend signalling models Formal models which establish corporate financial decisions as signals of the future prospects of a firm in the event of asymmetric information, are proposed by Ross (1977), Bhattacharya (1979) and Miller and Rock (1985). The model by Ross employs capital structure as the signal; it is easily altered to use dividend policy as the signalling device (Kalay (1980)). In the two models by Bhattacharya, and Miller and Rock dividends are employed as signals by managers to communicate their private information to the mar-ket. Offsetting the resulting benefits from signalling axe the various direct and indirect costs of paying dividends — providing thereby a theory of an optimal dividend policy. Hakansson (1982) develops sufficient conditions which can make dividend policy an in-formative and therefore a valuable signal. These conditions require either that investors have different probability assessments of dividend payouts, or that they have different preferred allocations of consumption over time, or that the financial markets are incom-plete. Ambrish, John and Williams (1987) propose a model in which multiple signals using dividends and investment axe employed by a firm. Their model is able to explain the positive announcement effect of dividends on stock prices and the negative effect of Chapter 2. CORPORATE DIVIDEND POLICIES: THEORIES AND EVIDENCE 14 net new stock issue. The beginnings of a realization that the shareholders of a firm may not be unani-mous regarding dividend payouts can be seen in the approach taken by Brennan and Thakor (1988). They model the choice of corporate cash disbursement method when the objective is to maximize the wealth of a majority of shareholders. They note that unlike cash dividends, share repurchases are not (and cannot be) pro-rata in order to qualify for favourable tax treatment. In view of the associated information asymmetries, repurchases can result in a wealth transfer from uninformed to informed shareholders. When the costs of information acquisition are fixed, the large shareholders are likely to be better informed than the small shareholders and further, as the size of the distribu-tion increases there is better incentive for the large shareholders to become informed. In view of their informational disadvantage then, the small shareholders axe likely to prefer cash distributions to share repurchases in general, and to prefer open market repurchases to self-tenders within the repurchase option. Brennan and Thakor conclude that if the choice outcome is determined by the potentially informed shareholders, then cash divi-dends will be preferred for small distributions, open market operations for medium-sized distributions, and self-tenders for the largest distributions. Any further refinement of this conclusion — specially to a testable hypothesis, would require the firm ownership structure to be explicitly incorporated into the model. But Brennan and Thakor leave their conclusions stated in terms of "an unspecified majority of shareholders" only. Masulis and Trueman (1988) also recognize the possibility of heterogeneity among shareholders on preferred dividends, and propose that the preference of some firm share-holders to defer cash dividends because of the tax penalty is tempered by the costs of doing so at the firm level. The funds retained by the firm must either be invested in real assets — at diminishing returns, or else must be put in financial assets resulting in double taxation. Other uses of the retained funds such as debt redemption may move Chapter 2. CORPORATE DIVIDEND POLICIES: THEORIES AND EVIDENCE 15 the firm from some other optima and would give rise to costs of their own. One conse-quence of all this is an agreement among shareholders to counter these costs and receive cash dividends, their tax penalty to some shareholders notwithstanding. The differential personal rates of taxation for different shareholders and the implied benefits of dividend deferral do however cause disagreement among them regarding the investment strategy. Shareholders in low tax brackets for instance, derive less benefit from the deferral and therefore prefer less investment specially when it is financed by retained funds. Masulis and Trueman also show that this disagreement disappears when investment is externally financed. 2.2.2 Empirical evidence on dividend signalling Empirical validation of the information content of dividends is provided by Eades (1982) who tests a dissipative signalling costs version of the signalling-through-dividends hy-pothesis. Further, Kalay and Loewenstein (1986) find information content in the timing of the dividend announcements as well. They find evidence of positive market reaction to earlier-than-expected announcements and negative or insignificant reaction to the later-than-expected ones. The proportion and magnitude of dividend reductions associated with the late announcements are larger than those on the average. One issue with a direct bearing on dividend policy is the manner in which dividends are paid. Most dividend paying firms do so by paying cash dividends at regular intervals — quarterly, semi-annually or annually. Once in a while there may be an "extra" or a "special" dividend which may not be repeated. Other methods of paying dividends include stock dividends and share repurchases. Theoretically a stock dividend is akin to a stock split, and a share repurchase its opposite. The former increases and the latter decreases the total number of shares outstanding without changing the underlying characteristics of the firm. Chapter 2. CORPORATE DIVIDEND POLICIES: THEORIES AND EVIDENCE 16 Brickley (1982) analyses stock returns, earnings and dividend patterns around the announcements of "extra" and "special" labelled dividends as well as the unlabelled, regular dividends. Judging by the stock price reactions, he suggests that increases in the unlabelled, regular dividends convey the most positive information among all dividend announcements. Grinblatt, Masulis and Titman (1984) analyze price reactions to announcements of stock splits and stock dividends. After adjusting for firm-specific announcement effects, they find that stock split and stock dividend announcements result in significant positive price reactions on ex-days. They also note that the reaction is stronger for dividends than for splits. Evidence on stock splits indicates that investors view them as signals of future increases in firm cash flows (Fama, Fisher, Jensen and Roll (1969)). Based on their comparison of firms which split stock with those that did not during the 1963-82 period, Lakonishok and Lev (1987) find that stock splits typically follow a period of abnormal growth in share prices. Also, they find that stock dividends are generally distributed by firms which pay lower than average cash dividends, and that the distribution of stock dividends appears to be on the decline. Share repurchases, like stock splits, are interpreted by investors as favourable news as shown by the share price appreciation on repurchase announcements [Dann (1981), Masulis (1980) and Vermaelen (1981)]. But favourable news notwithstanding, Barclay and Smith (1988) suggest that the observed dominant use of cash dividends over share repurchases must indicate higher costs (net of benefits) associated with the latter. These costs would include (i) the out-of-pocket expenses — higher for self-tenders because of the underwriter's involvement and lower for cash dividends and open market operations, and (ii) those arising from increase in the information asymmetry around repurchase events, since managers have control over the timing and size of repurchase. Barclay and Smith observe that the bid-ask spreads widen around repurchases and offer this as evidence Chapter 2. CORPORATE DIVIDEND POLICIES: THEORIES AND EVIDENCE 17 supporting their argument. Further evidence of the potential for this adverse selection in the repurchase timing and size by firm managers, is offered by Eckbo and Masulis (1988). They observe that stock repurchases typically do follow a significant run-up in stock price and that there is a drop in price subsequent to the repurchase announcement which is proportional to the run-up. Macdonald and Lindstrom (1986) investigate the financing hierarchy proposition, namely that firms with low free cash flows institute financing policies so as to mini-mize the probability of future equity issues and that those with high free cash flows use dividends only as a marginal source financing. They do so by comparing the dividend policies of firms at the time they are issuing equity with when they are not doing so. They show that in the latter case dividends depend only on the lagged stock returns. On the other hand, firms smooth dividend payouts when they are issuing equity. Thus dividends appear to serve as signals not only of future firm performance but also of the firm's current and expected future capital requirements. Marsh and Merton (1987) obtain similar evidence similar at the aggregate economy level. They show that the aggregate real dividend changes are driven by the one-period lagged real changes in stock prices, and that no other contemporaneous or lagged variables significantly contribute to these changes. They conclude that in view of this observed systematic time-series behaviour of aggregate real dividends, something more than the strictly firm-specific theories such as signalling, are needed to explain the dividend puzzle. Marsh and Merton do not however offer suggestions as to what these, possibly economy-wide, determinants of the dividends behaviour might be. Chapter 2. CORPORATE DIVIDEND POLICIES: THEORIES AND EVIDENCE 18 2.3 Dividend policy as a vehicle for minimizing agency costs The one shortcoming of the signalling explanation of dividends as it stands today is that it does not explain the variation in payouts across firms. One explanation for this variation is the differences in agency costs for different firms. For instance, Easterbrook (1984) argues that maintaining regular dividends requires a firm to go to capital markets frequently for raising funds and that this brings the firm under frequent scrutiny. The frequent scrutiny has the effect in turn of reducing the monitoring and other agency costs to the external shareholders of such a firm. This benefit when traded off against the costs of raising external financing then prescribes an optimal dividend policy even in the absence of taxes. The study by Rozeff (1981) of non-regulated US firms demonstrates that dividend payouts are negatively related to a firm's growth rate, it's stock beta and the percentage of the firm owned by the managers, and positively related to the number of outside shareholders. Thus, a growing firm makes better use of its cash flows by re-investing them rather than pay them out as dividends and at the same time, saves the costs of frequent trips to the capital markets — as does a firm with high variance (high stock beta) in its cash flows. Further, the fewer the outside shareholders the less the need to monitor and therefore lower the dividends. Similarly, the greater the number of outside shareholders the more diffuse the ownership and thus higher the dividends in lieu of the greater need to monitor. In what is possibly a first explicit reference to dividends as a tacit compromise reached between different shareholder groups, Schleifer and Vishny (1986) suggest that dividends act as subsidies paid to large shareholders from small shareholders towards the formers' monitoring costs. Those shareholders who own too small a fraction of the firm for mon-itoring to be worth their while and who pay taxes on dividends agree to receive some Chapter 2. CORPORATE DIVIDEND POLICIES: THEORIES AND EVIDENCE 19 dividends despite the tax penalty to them. Other shareholders who own a large enough fraction of the firm for monitoring to be worthwhile then have an added incentive to mon-itor if they receive dividends tax-free, with their tax savings subsidising their monitoring costs. It is not clear in the Schleifer and Vishny model why the large shareholders have to look to small shareholders at all to receive this tax-free dividend subsidy. All they need do is exercise their supposedly greater say in firm decision making in general, and the dividend decision in particular, since they control proportionately greater number of votes in the firm. 2.4 Summary The available models of a firm's dividend decision variously describe it as arising from differential personal taxation of dividends and capital gains, as a signalling device, and as a vehicle for curbing agency problems. Empirical evidence of the relevance of dividends to firm's valuation is mixed. These models tacitly assume shareholders to be unanimous regarding firm policies. This is true even of the models which derive relevance of dividends based on the existence of tax-clienteles which may have conflicting interests, but which do not try to influence dividend policy. Thus the potential role of ownership structure of a firm and, in partic-ular, that of the distribution of votes among diverse shareholder groups, on various firm policies has thus far been ignored. Chapter 3 P O T E N T I A L D E T E R M I N A N T S OF O W N E R S H I P S T R U C T U R E In the present study the ownership structure of a firm — specifically, the distribution of voting power among groups of shareholders with divergent interests, is assumed to be given. Its effects on dividend policy axe then explored under the assumption that the dividend policy does not affect firm value and therefore does not affect ownership structure in turn. 1 This raises the question of exactly what factors do determine the dis-tribution of voting power among the firm's shareholders and whether these factors affect the dividend decisions directly or indirectly in view of the assumption made of dividend irrelevance to firm valuation. Extant literature identifies these factors to be typically the ones which make control over the decision-making process in the firm valuable. Also, the focus in this study on the distribution of voting power aspect of ownership structure sets it apart from the prevailing approaches to investigate the effect of owner-ship structure on firm behaviour. Typically, in the existing studies the proxies employed for distinguishing between different ownership structures axe the various ownership con-centration indices such as the fractions of a firm owned by the five or the ten largest shareholders. Further, in the model proposed in this study, the equilibrium dividend is arrived at by shareholder groups playing a voting game. In practice however, a firm's dividend policy may not be (and typically is not) and, more importantly, need not be set by shareholders 1This is in contrast to the equilibrium models such as Brennan's (1973) in which required returns from investment in a firm include premia charged for receiving dividends. Thus the equilibrium investor portfolios are affected by dividend policies, thereby giving rise to different firm ownership patterns than if dividends did not enter asset pricing relations. 20 Chapter 3. POTENTIAL DETERMINANTS OF OWNERSHIP STRUCTURE 21 through the actually casting of their votes in a highly structured game. Firm managers can rationally anticipate the outcome of such a vote, given the distribution of voting rights among shareholders, and accordingly implement the equilibrium dividends. This process requires however that the shareholder voting process be instituted in a manner so as to ensure the rational anticipation and implementation of equilibrium outcome decisions by firm managers. There are practical problems in instituting the process however, which give rise to costs to shareholders of participating in the voting process. Most markets have devised regulations which attempt to facilitate participation and reduce costs. In section 3.1 below, we review the available literature on the potential determinants of a firm's ownership structure. Next, in section 3.2 the procedure which firm managements must follow so as to ensure shareholder participation in the voting process as prescribed by the regulatory agencies in Canada, is described. Those aspects of the procedure which keep it from from achieving its ultimate objective of full shareholder democracy and thus give rise to costs of participation in a vote for them, are pointed out. The final section in the chapter — section 3.3, describes the evidence thus far available on determinants of ownership structure, and the nature and outcomes of shareholder participation in firm decision making through the voting mechanism. 3.1 V a l u e of C o n t r o l Traditionally, models of firm behaviour treat shareholders alternatively as identical in-dividuals, as individuals who may not be identical but are unanimous regarding firm policies, or as passive investors who either have no say in setting firm policies or choose not to exercise it. In this traditional view, individuals who do not find some firm policy acceptable, sell their shareholdings in the firm, i.e. they follow the "Wall Street Rule" and vote "with their feet". This is a reasonable strategy, as long as the resulting need to Chapter 3. POTENTIAL DETERMINANTS OF OWNERSHIP STRUCTURE 22 rebalance portfolios does not impose costs on the selling shareholders. Conditions under which shareholder unanimity occurs are well known [e.g. Diamond (1967), Hart (1979), Baron (1979) and Grossman and Stiglitz (1980)]. For instance, perfect information and complete markets or markets with spanning and competitivity ensure value-maximization by firms and unanimous agreement among shareholders over firm policies. When shareholders are unanimous, complete separation of ownership and corporate control occurs trivially since under unanimity control over policy-setting has no value (the assumption of perfect information implies zero agency costs). In addition to value-maximization by firms, efficiency gains from specialization by individuals in management and risk-bearing functions, result under complete separation (Fama and Jensen (1983)). If conditions for complete separation and shareholder unanimity are not satisfied, then "control" of the firm's decision making process may become valuable. Control is valuable to an individual shareholder if exercising the control function allows him to extract cash flows from the firm not generally available to other shareholders. Among the factors which can give rise to a positive value to control, are agency costs [Jensen and Meckling (1976)] and asymmetric information between management and shareholders as well as between various groups of shareholders [Stiglitz (1972), Grossman and Hart (1981), Baron (1983)]. 3.2 The Mechanics of Shareholder Voting In the normal course of a firm's operations, the shareholders delegate policy-making authority to the board of directors. Typically, some of the top-level firm managers are also on its board. We will refer to these managers and the other directors collectively as the incumbent management. The incumbents in turn, appoint officers to carry out the Chapter 3. POTENTIAL DETERMINANTS OF OWNERSHIP STRUCTURE 23 day-to-day operations of the firm in the best interests of the firm shareholders. Also, the incumbents retain the authority to remove or replace these officers. The composition and tenure of the incumbent management itself is overseen by the firm shareholders. Under the present securities legislation, the management is required to get shareholder approval for all material changes in the firm's business structure. Material information and changes2 include: • Actual or proposed changes in the control of the company. • Actual or proposed acquisition or disposition of material assets. • Proposed take-overs, mergers, consolidations, amalgamations or re-organizations. • Any material discoveries, changes or developments in the company's resources, technology, products or contracts, which would materially affect the earnings of the company upwards or downwards. • Proposed changes in the capital structure, including stock splits or stock dividends. • Indicated changes in earnings upwards or downwards of more than recent average size and changes in dividends. • Any other material change in the affairs of the company which could reasonably be expected to affect materially the value of the security. In particular, changes in the control apparatus referred to in the first item above, in-clude decisions relating to the election or removal of directors, appointment of auditors, amendments to the articles of incorporation and the by-laws.3  2Source: Johnston (1977), p270 3Various provisions in the CBCA. For details, see Crete (1986) p210-212. Chapter 3. POTENTIAL DETERMINANTS OF OWNERSHIP STRUCTURE 24 A l l material information is communicated to the shareholders in an information cir-cular which accompanies or precedes the management's notice of the shareholder meeting at which the proposals are to be voted upon. Also accompanying the information cir-cular is a prescribed proxy form, which when signed to the effect and returned by the shareholder, entitles the management to vote on the former's behalf at the meeting. The notice of the meeting is sent to all those shareholders who are shown on company records on a date specified in the information circular. Regulation requires a minimum of ten and a maximum of sixty days between the date of the meeting notice and the meeting date. At the meeting, proposals which get a pre-specified majority of all votes cast, whether in person or by proxy, are then declared carried. The pre-specified majority can be either simple or a super-majority depending on the type of the proposal, as set down in the articles of incorporation. The proposals to be voted upon in the meetings are selected by the incumbent man-agement. Shareholders wishing to have their proposals included in the meeting agenda, have to convince the management to include these as the latter's own proposals. This fea-ture of the Canadian legislation distinguishes it from its U . S. counterpart which permits shareholders other than the management to have their own proposals put to vote.4 3.2.1 Potential problems with voting by proxy While proxies are normally solicited by management, any other shareholder or a group thereof can do so as well. This is done through a mailing of information circular and proxy form, similar to the ones by management. The circular identifies the soliciting individual or the group and the concerns about the management actions which may have precipitated the dissident action. Dissident information and proxy circulars are also mailed to firm shareholders of record, as per the lists supplied to the dissidents by the 4Crete (1986), p580 Chapter 3. POTENTIAL DETERMINANTS OF OWNERSHIP STRUCTURE 25 fcm management. Once received, the proxies are valid for up to sixty days and are used at the shareholder meeting for which they are issued. They can be revoked by the issuing shareholder at any time by the simple act of issuing a new proxy and authorizing a different agent to vote on his behalf. Proxy solicitation is costly.5 These costs further escalate when a contest occurs. Escalation comes from the advertising campaigns mounted by the management and the dissidents to win over other shareholders to their respective camps. Also, court battles can ensue between the two camps when information is alleged to be misrepresented or withheld by one or the other party. Whether contested or not, the management's costs of solicitation are borne by the firm. Proxy regulation is not clear however, on who should bear the cost of proxy solicitation by dissidents.6 Courts seem to support the reimbursement of dissidents' costs in the event of a successful dissident campaign.7 Proxy contests are considered to be the least viable of the corporate governance tools available to shareholders. Berle (1962) and Manne (1964) note that in an ideal, democratic political process, a vote works as it is supposed to, because everybody is equally well-informed and everybody casts his vote, so that the outcome truly represents the preferences of the majority. They suggest however that in practice, there is not enough information disclosure by the firms, and not enough participation by shareholders, in spite of the elaborate regulatory set-up requiring disclosure and encouraging participation in shareholder voting. They see the problem partly as one inherent in situations of collective choice and partly caused by the underlying proxy system of shareholder voting. 5Based on a survey of companies, Crete (1986) reports these costs to be anywhere between $3,000 to $30,000 per shareholder meeting for the proxy mailings only. Administrative costs are incurred on top of these. 6Crete (1986), pp295-311. 7 Crete (1986), p307. This is with reference to the US courts. In Canadian law, there is no reported decision regarding this issue. Chapter 3. POTENTIAL DETERMINANTS OF OWNERSHIP STRUCTURE 26 The problem of collective choice arises when shareholders individually hold small por-tions of the outstanding votes and the residual claims to the firm. The individuals do not have the incentive to participate in the vote, since they perceive that by themselves they cannot influence the vote outcome (Berle and Means (1967)). Further, if partic-ipating in the vote is costly then the so-called free-rider problem (Grossman and Hart (1980)) arises. Each individual would like every other individual to bear the cost and vote. Of course, the benefits of a favourable vote outcome will be shared by everybody. With everyone reasoning along these lines, no shareholder casts his vote. The free-rider problem is well illustrated when a two-tier tender offer is made for a firm's shares.8 In two-tier tender offers the target shareholders are faced with the lands of decisions the prisoners in the classic prisoners' dilemma are faced with. The individual making the offer likely has the technology to improve the firm's fortunes. So it is in the interest of each shareholder to hold on to his shares, so that he can benefit from the change in control. But for the tender offer to be successful, some shareholders must tender their shares. Each shareholder would thus like all other shareholders to tender their shares, possibly resulting in no shares being tendered. Voting by proxy, if costless, can alleviate the above collective choice problems.9 A small shareholder can ask a shareholder with large holdings and with objectives similar to his own, to vote on his behalf. A shareholder with large holdings has the economic incentives to be informed and to participate in the vote, even if it is costly to do so. Further, armed with enough proxy votes he can also be reasonably sure of influencing the vote outcome. 8In a two-tier tender offer, the bidder specifies a maximum number of shares to be accepted. The two-tier bid is accompanied by the bidder's intention to acquire the remaining shares in a subsequent offer if the conditions for the initial offer are met. These conditions include the receipt of a minimum number of tenders sufficient to guarantee a subsequent merger vote. 9In voting by proxy, all a shareholder need do is authorize a preferred agent to vote on his behalf, by putting the latter's name on the proxy circular, and mailing it. Several companies have introduced phone-in proxies, further facilitating shareholder participation. Chapter 3. POTENTIAL DETERMINANTS OF OWNERSHIP STRUCTURE 27 There are problems inherent in the proxy system as well, however.10 They arise firstly because of the less-than-accurate available information and secondly, because of the differential access of various parties to it. For instance as noted earlier, the period between the announcement of a shareholder meeting and the actual meeting can be anywhere between ten to sixty days. Also as noted before, only shareholders as of the specified record date receive the meeting announcement. Those shareholders who sell their holdings after the record date, have no interest in casting their votes. The buyers — the new shareholders on the other hand, do not automatically get the title to vote. The ability of a shareholders to sell their holdings at short notice, so desirable for their liquidity and efficient share pricing, slows down the reallocation of voting rights in response to the trades thus works to the detriment of shareholder participation in voting. Further, for frequently traded firms a sizeable number of shares are recorded as being held in a "street name" — the name of the broker or his nominee, a bank or some other intermediary rather than in the name of the beneficial owner. In this case whether the proxy materials reach the beneficial owner in time or at all, is determined largely by the diligence shown by the intermediary. These are the obstacles faced by the incumbent management in ensuring timely dis-closure of material information to shareholders. They are compounded further when dissidents attempt to solicit proxies. Even if management is not itself hampered in its proxy soli citations, it can seriously jeopardise the dissidents' timely access to shareholder name and address fists and to other policy proposals related information, without seeming to violate any regulations. 10See Crete (1986) for a detailed discussion of these. Also, Easterbrook and Fischel (1984) suggest reforms to the proxy system mainly in the areas of information disclosure and solicitation-cost sharing between the firm and the dissident shareholders. Chapter 3. POTENTIAL DETERMINANTS OF OWNERSHIP STRUCTURE 28 3.3 O w n e r s h i p s t ruc ture a n d firm va luat ion The theoretical model of this study presumes that there is no correlation between firm value and equilibrium ownership structure. Partial support for this assumption is found in Demsetz and Lehn (1985), who examine US corporations with different ownership concentration levels.1 1 They find no systematic link between concentration of ownership and firm profitability. No study has yet performed a similar analysis on Canadian data. According to the Report of the Securities Committee on Take-Over Bids (1983) con-centration levels in Canadian manufacturing are higher than those in other industrialized nations.1 2 For instance in 1983, for 137 (48.4%) out of 283 companies in the TSE 300 Composite Index, 50% or more of their voting stock was held by one or a small group of shareholders; for 85 (30%) of them, 20-49.9% of the voting stock was held by one individual or a small group, and only 61 (21.6%) of the 300 companies were widely held. Comparative figures for the US Standard and Poor 500 Index were 6 (1.2%), 68 (13.6%), and 426 (85.2%) companies respectively. Assuming the absence of a systematic link be-tween ownership concentration and profitability in Canada as well, an implication of the above would be that the trade-offs between benefits and costs of control in the Canadian firms are qualitatively different from those faced by the US firms. Ownership translates into control through the shareholders' right to vote on major firm policies. By voting to affirm or reject them, shareholders can force firm manage-ments to pursue value maximizing policies. Alternatively, in the event of concentrated ownership a major shareholder can steer the management to pursue policies that benefit him, possibly at the expense of other shareholders. The issues relating to shareholder voting behaviour are complex, and theoretical and empirical investigations into these 1 1 For a comprehensive survey of recent studies on ownership structure and related issues, see Jensen and Warner (1988). 1 2 Source: Khemani (1987). Chapter 3. POTENTIAL DETERMINANTS OF OWNERSHIP STRUCTURE 29 have begun relatively recently. Grossman and Hart(1988) and Harris and Raviv (1988) demonstrate that voting rules affect firm value. They show that the combination of a simple majority rule and one share-one vote ensure the election of the best management team. However their model also implies that this combination does not maximize the value of the securities in the event of a corporate control contest, and that the share value should increase when a firm switches from one share-one vote to shares with differential voting rights. Empirical evidence on this issue is mixed. Partch (1987) finds no effect of dual class recapitalizations on share prices. Jarrell and Poulson (1988) on the other hand find a significant negative price reactions to such recapitalizations. The difference between the findings of these two studies may be explained by the fact that Partch ex-amined recapitalizations prior to the 1986 N Y S E relaxation of listing standards, whereas Jarrell and Poulson employ the post-relaxation data. Negative price reactions to issuance of limited voting common stock are also observed by Lease, McConnell and Mikkelson (1984). But their small sample (6 firms) limits them from interpreting their results for the effect of such issuance on shareholder wealth. Rydqvist (1986) investigates whether a premium is paid by investors for the right to vote, over and above the price paid for the economic value of a firm's share. He compares the price differentials for shares with differential voting rights, trading on the Stockholm Exchange. He shows that the differentials become larger around the times of take-overs, when the voting rights are likely to be valued more. 3.4 Shareholder voting behaviour: empirical evidence Exploration of shareholder characteristics as potential determinants of voting behaviour has begun relatively recently. Voting behaviour of shareholders on issues such as anti-take over amendments is investigated by Brickley, Lease and Smith (1987). They show Chapter 3. POTENTIAL DETERMINANTS OF OWNERSHIP STRUCTURE 30 that large, outside blockholders vote differently than large, inside blockholders and that institutional shareholders' voting is likely to be influenced by whether or not they have direct business dealings with the management. Morck, Schleifer and Vishny (1987) explore the relationship between ownership by managers and firm performance as measured by Tobin's Q and profitability.1 3 They find that the relation between the mangers shareholdings and firm performance is non-monotonic, that in fact Q rises as managers' shareholdings increase from 0 to 5%, it falls as ownership increases from 5 to 25% and then increases again. They interpret this result as implying that managers' shareholding have to increase to the 25% level before they can exercise enough control to pursue their own interests, possibly at the expense of other shareholders. Holderness and Sheehan (1988) investigate the link between ownership structure and firm financial decisions. They analyze matched samples of firms with majority and dis-persed ownerships. Their results show that on the average, firms with a corporation as a majority owner pay higher dividends than firms with dispersed ownership. By con-trast, firms with an individual as a majority owner pay lower dividends than firms with dispersed ownership. Thus ownership structure appears to have links with firm financial decisions in general, and in particular with dividend policy. Brickley, Lease and Smith (1988) test the hypothesis that large shareholders vote against firm value decreasing proposals of the management. They analyze the voting patterns of large blockholders on anti-takeover amendments proposed by the manage-ments in 201 large U.S. firms in the year 1984. Anti-takeover amendments are chosen for the tests since if adopted they will help entrench the management. Once entrenched, the management will be free to undertake projects which may not be in the best interests of 13Tobin's Q is denned as the ratio of firm's market value to the replacement cost of plant and inventories. Chapter 3. POTENTIAL DETERMINANTS OF OWNERSHIP STRUCTURE 31 the shareholders. The evidence suggests that large blockholders are indeed more likely than non-blockholders to vote against anti-takeover amendments. Also among the insti-tutional shareholders, mutual funds, foundations, and public-employee funds are found to be more likely to oppose management than banks, insurance companies and trusts. The authors suggest that one reason for this may be that institutions in the first set are less likely to benefit from their business relations with the incumbent management than the ones in the second set. There is evidence that the stock market treats proxy statement disclosures and share-holder meetings as informative events. Brickley (1986) finds significant, positive abnormal returns for a random sample of stocks around their shareholder meeting dates, although proxy disclosures and shareholder meetings are predictable events. This is consistent with the finding of Kalay and Lowenstein (1985) that risk and expected return can increase around predictable events such as dividend announcements. It also suggests that share-holder participation in the firm decision making process, as at a shareholder meeting is possibly seen by the market as a "relevant" event. What remains to be understood is the manner in which this "relevance" is manifest. 3.5 Proxy contests The efficiency of the proxy system of voting and the dissidents' success in the event of a contest depends in great measure on the vote of large shareholders. What determines a large shareholder's voting behaviour is not yet fully understood, though. Theoretical and empirical investigations of this behaviour have begun relatively recently and several hypotheses have been proposed and tested. In one view, since large shareholders have the the incentives to monitor firm perfor-mance, they can recognize and then vote against, those management proposals which will Chapter 3. POTENTIAL DETERMINANTS OF OWNERSHIP STRUCTURE 32 reduce firm value. By contrast, large passive shareholders would just sell their holdings rather than vote against the management. At the other extreme, large institutional share-holders can be influenced in their voting by their existing or potential business relations with the incumbent management, to the detriment of the shareholders they represent. For instance, an insurance company may act as a corporation's insurer while holding its stock, or a fund manager's vote may be influenced by the possibility of his/her managing the same corporation's pension fund sometime in the future. The only investigations to-date of proxy contests are by Dodd and Warner (1983) and Pound (1988), both based on US data. Dodd and Warner find that the occurrence of a proxy contest is accompanied by significant share price appreciation. This appreciation is permanent when the dissidents subsequently win. Thus their evidence is consistent with the proxy contests being useful for management oversight. Pound investigates the link between the outcomes of proxy contests and the firm ownership compositions. His sample consists of 100 major contests which took place during the 1981-84 period. He finds that the dissidents are more likely to succeed in firms with lower institutional ownership and less dispersed share ownership. Also, he finds that proxy contests are more successful in placing outside directors on the boards than in transferring full control of the firm. 3.6 Summary The ownership structure of a firm is assumed to be given and is independent of the dividend policy. The potential determinants of the ownership structure are of interest in their own right, however. Literature suggests these determinants to be generally those which make control over the decision making process in a firm valuable. The literature on shareholder voting behaviour is not very extensive. One insight from the available evidence is that the size of holdings in a firm critically affects a shareholder's Chapter 3. POTENTIAL DETERMINANTS OF OWNERSHIP STRUCTURE 33 voting behaviour. The proxy system of voting allows shareholders to participate in a firm's decision making process at supposedly lower costs than if they had to be physically present at all shareholder meetings. However, while reducing some types of costs the proxy system introduces other types of costs to shareholder participation in voting. One of these is the differential access to information for different shareholders, which may be relevant to their voting decisions. Chapter 4 A MODEL OF THE CONSENSUS DIVIDEND POLICY In this chapter, we present an approach to the issue of how personal taxes affect corporate dividend policy, which is different from that in the literature. Here, the dividend decision process itself is studied rather than the pricing implications of dividends which may arise from their differential taxation or from their informational role. An important insight from the literature cited in the discussions so far is that, in the presence of short selling restrictions and heterogeneous marginal tax rates across investors, firms with different dividend payouts attract different shareholder tax clienteles. The restrictions on short-selling are necessary to rule out perfect tax arbitrage between investors in different tax brackets. In the absence of any such constraints, investors in high tax brackets would short-sell high dividend yield shares cum-dividend and then buy them back ex-dividend to cover the sale. Investors in low tax brackets would do the opposite: short-sell low dividend yield shares cum-dividend and then buy them back ex-dividend. Thus with short-selling permitted, effect of differential taxation of investors will be arbitraged away. By contrast, in this study the essence of a tax clientele is that all members unan-imously agree on and support (through a vote) the firm's dividend policy expected to be implemented over the next period. Thus from the perspective of this study, the tra-ditional tax clientele argument implies that the distribution of votes across the firm's shareholders has no impact on the firm's dividend policy. Suppose, then, that non-tax considerations induce investors in high tax brackets to 34 Chapter 4. A MODEL OF THE CONSENSUS DIVIDEND POLICY 35 hold high dividend yield stocks in spite of the relative tax penalty. These non-tax con-siderations might include preferred risk-(pre-tax)return profile of an otherwise not-so-attractive stock because of its high dividend yield. High dividend yield stocks would feature naturally in a reasonably diversified portfolio. Indeed, a high dividend yield stock may be attractive to an investor who places a high value on liquidity, despite the higher tax penalty. The periodic cash receipts imply savings in transaction costs which would be incurred if the investor were to periodically sell part of his portfolio in order to generate liquidity. Thus, for one or more of these reasons, one may expect to find dif-ferent tax-clienteles coexisting as shareholders in the same firm. However, one question then naturally arises: if investors in high and low tax brackets hold shares in the same company, how do they reach a consensus on the firm's dividend policy? Intuitively, one would now expect the consensus policy to depend on the distribution of votes across the various tax clienteles holding the firm's shares. We formalize the above intuition in a game-theoretic setting. "Voting power" is defined as the probability that a particular block of shares will be pivotal in determining the outcome of a vote (if it were to be held) on the firm's dividend policy. In contrast to the traditional clientele argument, this model implies a positive correlation between the voting power of the shareholder with the lowest tax-bracket and the level of the firm's expected dividend yield. While the model unavoidably casts the complex voting process in simplistic terms, its assumptions nevertheless reflect several stylized facts. For example, firms often have both corporate/institutional owners as well as individual shareholders, including managers. Since intercorporate dividends are tax free, this suggests that obtaining a consensus on the firm's dividend policy is indeed a non-trivial task even in the absence of agency costs. Our model is cast in terms of a voting game between a large corporate shareholder, which in Canada has a tax-induced preference for dividends over capital gains, and managers who Chapter 4. A MODEL OF THE CONSENSUS DIVIDEND POLICY 36 own voting stock. There is also a third group consisting of a large number of individual shareholders whose marginal tax rates we assume are unknown. K the corporate owner and the management cannot reach an agreement, this third group will participate in a costly vote to determine the period's dividend. The mere possibility of such a vote "disciplines" the two major players much in the same way that litigation costs affect a plaintiff's litigation and settlement decision.1 Furthermore, we endow the management with an informational advantage in estimat-ing the outcome of the shareholder vote, if held, as well as an absolute cost advantage in organizing the vote. This is consistent with the fact that management often keeps track of the identity of the firm's shareholders, an extremely difficult task for outside shareholders.2 These assumptions produce a consensus dividend policy which typically will not require a shareholder vote, a result which is also consistent with observed div-idend decision processes.3 The consensus dividend represents a compromise in that it will he somewhere between the minimum preferred by managers (e.g., given information signalling arguments) and the maximum feasible given the firm's bond covenants (which is the corporate shareholder's preferred dividend). The precise location of the dividend in this interval is shown to depend on the relative voting power of the two major players. 4.1 The consensus dividend policy Let I, M and OC denote the corporate shareholder, the management, and the re-maining body of anonymous individual shareholders, respectively. Following Canadian ^ee, e.g., Bebchuk (1984). 2This point is emphasized by Pound (1988) in his discussion of the relative costs of mounting a proxy contest. Johnston (1977) and Crete (1986) argue that dissident shareholders in Canada axe at a substantial cost disadvantage vis a vis management in preparing for proxy contest. 3In Canada, a "material" change in the firm's dividend policy must be reported to the Securities Commission. There are instances where shareholders have reacted to such information circulars by mounting a proxy contest to defeat the proposed dividend change. Chapter 4. A MODEL OF THE CONSENSUS DIVIDEND POLICY 37 tax laws, / pays capital gains tax but receives dividends tax-free. Thus, J prefers divi-dend income to capital gains income. In contrast, individual M's marginal tax rate on dividends exceeds the marginal tax rate on capital gains. M therefore prefers capital gains over dividends. Finally, as pictured in Figure 4.1, the marginal tax rates of mem-bers of the group OC are assumed to be distributed somewhere between the marginal tax rates of I and M . 4 In Figure 4.1, ff and tf are the ith. shareholder's marginal tax rates for dividend and capital gains incomes, respectively. In group OC, some individuals (such as ii) prefer dividends over capital gains, while others (such as ^3) prefer capital gains over dividends. If the tax rates of all the members of the group OC axe known, then the exact outcome of the vote is also known ex ante, and the dividend selected by the (known) majoxity will be adopted without a vote. Thus, for the model to be interesting, it must be assumed that the marginal tax rates of the members of group OC are unknown to the two other players. Of course each member of the group OC, acting individually, does not know any other member's marginal tax rate either. This lack of knowledge on every player's part about the types of other players, has the effect of making the game more complex than is immediately evident. The uncertainty in the outcome of the game, coupled with costs of resolving the uncertainty, drives the players to search for compromise outcomes. The basic structure of the game is as follows (Figure 4.2): • Shareholder I proposes a dividend D to management M. • M accepts or rejects the offer. Before making this decision, however, M receives private information on the probability p that I will win the vote, if held. The private signal is drawn from a distxibution function F which is common knowledge. 4This assumed distribution of tax rates is consistent with the Canadian tax code which allows the first $1,000 of dividend to be deducted from investor's taxable income, and which provide a dividend tax credit for small, individual investors. Chapter 4. A MODEL OF THE CONSENSUS DIVIDEND POLICY 38 marginal tax rates for dividends marginal tax rates for capital gains I ii 12 iz M Shareholders Figure 4.1: Shareholder Marginal Tax Rates Chapter 4. A MODEL OF THE CONSENSUS DIVIDEND POLICY 39 • H M rejects the initial proposal, I will call a vote at a cost C j . The cost to M of participating in the vote is CM < Cj. The vote is between Dmax proposed by I and Dmin proposed by M. Note that the players are only allowed to play their pure strategies at each of the two stages and that no randomization between these pure strategies is permitted. Thus, when I offers D to M in the first round, M's acceptance or rejection of D is governed only by the signal p received by M. Similarly if M rejects the initial offer D and if I decides to ask for a vote, then they respectively propose Dmax and Dmin. The outcome of the vote is then implemented as the dividend policy and the game is over. The game thus does not allow the players to make Bayesian revisions to their offers once they have the chance to infer some information from their opponents' play. The costs Ci and CM play a particularly important role in this model: With zero costs of participating in the vote, it follows that M will accept Ps dividend proposal only in those states where M's private information indicates that I would have been better off calling the vote. This is the essence of M's informational advantage. If calling the vote were costless for I his dominant strategy would be to propose Dmax, thereby always forcing the vote. A positive marginal cost Cj modifies Ps behaviour (i.e., causes I to propose D < Dmax) because Cj can now be avoided through an agreement with M . On the other hand a positive cost CM incurred by M in the event of a vote, counterweighs Ps incentive to reduce the proposed D since it increases the number of states in which M will accept Ps proposal. By assuming Cj > CM, we guarantee that 7's optimal proposal is less than Dmax. As stated earlier, this assumption is consistent with stylized facts about the relative costs to outside shareholders and management of preparing and participating er 4. A MODEL OF THE CONSENSUS DIVIDEND POLICY I proposes D Yi(D),YM(D) Figure 4.2: The dividend voting game. Chapter 4. A MODEL OF THE CONSENSUS DIVIDEND POLICY 41 in a general vote. In order to establish 7's equilibrium dividend proposal, let Yi(D) denote the value to shareholder i of receiving a cash flow of D dollars from the firm in the form of dividend rather than capital gain. Thus, for I, Yi(D) > 0 (a relative tax benefit) since vj < fj, while for M , YM(D) < 0 (a relative tax penalty) since tj^ > r ^ . Focusing first on M' s decision problem, we have that M will accept 7's initial proposal of D provided his relative tax penalty YM(D) is less than or equal to the conditional expected tax penalty resulting from calling the vote: YM(D) > (1 - p)YM(Dmin) + PYM{Dmax) - CM (4.1) Rearranging, P > Y f { f - Y f ' - ^ ^ *D) (4.2) Thus, M will accept a proposed dividend of D if his private information indicates the Ps probability of winning the vote (if held) exceeds the threshold value q(D). In the following, we refer to p as 7's voting power, and to 1—p as M's voting power. Expressions ( 4.1) and ( 4.2) say that M's accept/reject criterion depends on his private information about 7's voting power.5 Turning to 7's decision problem, the probability that M will reject D is equal to F(q(D)), where it is assumed that dF(q(D))/dD > 0 and d2F(q(D))/dD2 > 0. Thus, the expected probability that 7 will win a vote on Dmax, given that M rejects the initial proposal of D, is given by _ Sfm xf(x)dx 6 - mo)) <4-3) 5Thus, the concept of voting power is analogous to a probability whose value depends on the distri-bution of votes across all three groups I, M and OC. A related concept (the Shapley value) has been developed by Milnor and Shapley (1978) in the context of cooperative games with a few large and a large number of small shareholders. Recently, Hart and Moore (1988) have shown that the Shapley value is an appropriate measure of the value to a player of participating in a non-cooperative games as well, a result which is useful in our subsequent empirical analysis. Chapter 4. A MODEL OF THE CONSENSUS DIVIDEND POLICY 42 where / is the density corresponding to F, and a is the lowest value I puts on his own voting power. As a result, the expected value to 7 of proposing D is given by E(Yr) = [1 - F(q(D))}Yj(D) +F(q(D))[6YI{Dmax) + (1 - S)Y!(Dmin) - &] (4.4) Ps objective is to maximizes E(Yx) with respect to D. The first-order condition for this maximization problem is E'(Yr) = [1 - F(q(D))]Yj{D) - ^ j^V/CP) + Cz - Yj(Dmin)] +[x/(A»««) - F /(/J>m,„)] g(/J>)/( g( JD)) g '(I?) = [l-F(q(D))]YI'(D) (0 ^ n ^ ^ f . ^ ( g ) ~ g ( J ) ^ ( ^ m a x ) ~ (1 ~ q(Z?))r/(gnun) + C/] = 0 (4.5) where (i) is the marginal benefit of increasing D given that M accepts the proposal, (ii) represents the marginal increase in the probability that M will reject the proposal, and (iii) represents Fs opportunity cost if M rejects the proposal and a vote takes place when the private signal p equals q(D). Let D* denote the solution to this first order condition. The existence of D* is ensured by the fact that E„{Yl) = _2*zismmD) + <zi2^ (4.6) ^ ^ d D ^ {Yl[D) + C l ~ 9 ( D ) y ' v A » " ) " ( X " Q(D))Yi{Dmin)] (4.7) < 0. (4.8) In other words, 0 < Dmin < D* < Dmax and, ceterus paribus, the greater the voting power of 7, the higher is the probability that I will win the vote, if held. Similarly, Chapter 4. A MODEL OF THE CONSENSUS DIVIDEND POLICY 43 since M accepts the proposed dividend D* with probability 1 — F(q(D*)), the greater the probability that M will accept Fs proposal, the higher is the amount D*. 1 — F(q(D*))(> 0) is the probability with which a consensus dividend policy for the firm will emerge without being put to a vote. Note that the first order condition ( 4.5) above can be solved to yield an expression for D* if additional structure is imposed. Clearly, when the probability p of I winning the vote is 1 and when this fact is common knowledge — as it would be when I holds majority of the voting rights, then Dmax is the solution to the game. Similarly, when p is 0 and this fact is common knowledge (M holds majority of the voting rights), then D m i „ is the equilibrium dividend. The prediction that the equilibrium consensus dividend policy is a function of the dis-tribution of voting power among J, M, and OC is empirically testable and is explored in the rest of this investigation. Interestingly, the model also provides a possible alternative explanation for Kalay's (1982) observation that firms often pay a dividend which is be-low the maximum possible dividend allowed under the firm's debt covenants. Moreover, the model prediction is consistent with the observation of Booth and Johnston (1984) that the ex-dividend day trading volume is decreasing in the dividend yield. High yield firms in our model are firms with large corporate shareholders. Since these shareholders favour high dividends, they are less likely to participate in any tax-induced trading. If we assume then that the managers of a firm do not routinely trade around their own firm's dividend payment date, it implies a lower float available and thus lower incidence of ex-dividend trading volume in high-yield stocks. Chapter 4. A MODEL OF THE CONSENSUS DIVIDEND POLICY 44 4.2 Properties of the Dividend Equilibrium As noted in the previous section, the equilibrium dividend D* can be calculated if addi-tional structure is imposed on the model. For instance, the density function /(•) for the signal p as well as the the value functions Yj and Ym need to be specified before D* can be specified as a function of the parameters such as Dmax, Dmin and the dividend and capital gains tax rates of I and M. While appropriate assumptions could be readily made regarding the density function, specifying the value functions does not appear to be so straight forward. A potential specification for the value functions may be the after-tax proceeds for I and M , calculated as if they would sell their holdings after receiving the dividends. The problem though, with this approach would be the need to specify cum-and ex-dividend firm value acceptable to both I and M.6 The dividend payout D* is a Nash equilibrium in the voting game played by the shareholders M and I: Ps expected payoff is a maximin of the payoffs M will allow him to receive. We next look at the various assumptions that lead to this equilibrium and explore the extent to which it is self-enforcing. To start with, recall that the shareholder I makes a one-shot offer D* to M. Ii M rejects D*, then I asks for a vote. It would appear that this set-up endows I with first-mover advantage. However on closer scrutiny, it becomes clear that this advantage is neutralised by the fact that M possesses information which I does not. Indeed if I knew exactly what signal M receives regarding 7's probability of win, then I would adjust his offer D* accordingly and capture all of the benefits arising from this information. Thus the assumptions that I moves first and that M possesses private albeit imperfect information, are critical for I and M to play the voting game to begin with. 6In a game more complex than the one proposed here, I and M could possibly play two rounds rather than just one as in the set-up proposed above. In the first round, they would arrive at some mutually agreeable firm valuation criterion, and in the second one they would design the firm dividend policy. Of course, the first round would then utilize dividend policies to be proposed later in a strategic manner. C h a p t e r 4. A MODEL OF THE CONSENSUS DTVIDEND POLICY 45 T h a t M possesses p r iva te i n f o r m a t i o n is reasonable to expect i n p rac t i ce . R e c a l l tha t M represents those shareholders who have sizeable shareholdings i n the f i rm and who are t axed on the i r d i v i d e n d receipts at thei r o r d i n a r y i n c o m e t ax rate. So M represents i n p a r t i c u l a r , f i rm managers — managers w h o also h o l d shares i n the firm. T h u s M has access to the i n f o r m a t i o n regard ing the i d e n t i t y of a l l f i r m shareholders . B a s e d on this i n f o r m a t i o n , M can infer a lbei t i n an imperfec t manne r , the p r o b a b i l i t y w i t h w h i c h a shareholder w o u l d side w i t h h i m or w i t h 7, i f a d i v i d e n d payout is pu t to vote. Shareholder I can get access to this i n f o r m a t i o n o n l y at great costs. T h i s assumed ease of access to i n f o r m a t i o n on the m i n o r , oceanic shareholders for M a n d i ts l ack for I is subs tan t i a t ed by e m p i r i c a l obse rva t i on . 7 T h e exis tence of costs C j and CM to I a n d M of o rgan i s ing and p a r t i c i p a t i n g i n a vote , ru le out the p o s s i b i l i t y tha t a vote w i l l be he ld i n a l l states of na ture . R e c a l l tha t b o t h I a n d M axe r i sk -neu t r a l so that , i n the absence of costs they w o u l d a lways opt for a vote ra ther t h a n t r y to "negotiate" and axxive at a compromise q u a n t i t y such as D*. T h e costs CM m a y be thought of as the ones expended by M over and above those i n c u r r e d i n keeping the shareholders i n fo rmed regard ing major firm pol ic ies and changes there in . T h e y w o u l d i nc lude the costs of resources expended i n o rgan i s ing spec ia l shareholder meet ings i n a d d i t i o n to the rou t ine a n n u a l meet ings , and the o p p o r t u n i t y costs of los ing w h e n the vote is he ld . T h e costs C j w o u l d s i m i l a r l y consist of the t ime , m o n e y a n d effort expended b y I to organise a diss ident vote. Fu r the r , i n re la t ive te rms the costs Ci are l i ke ly to be h igher tha t CM i f on ly because of the d i f f icul ty of access to relevant i n f o r m a t i o n faced b y I. Once aga in , this is borne out b y e m p i r i c a l observa t ion . T h e a s s u m p t i o n tha t I asks for a vote w i t h p r o b a b i l i t y 1 i f his i n i t i a l offer is rejected b y M, is a c r i t i c a l one i n the proposed d i v i d e n d v o t i n g game. It is this a s s u m p t i o n w h i c h 7See Chapter 3 for the institutional details which govern the access to critical information for various groups of shareholders. Chapter 4. A MODEL OF TEE CONSENSUS DIVIDEND POLICY 46 forces M to act faithfully according to his signal p. If M were to reject D* even though his signal p > q(D*), he would face a vote with probability 1 which in turn would result in the imposition of Dmax with non-zero probability. The threat of an actual vote being held with probability 1 and M's wrongful rejection of J's offer being found out with non-zero probability, becomes a credible threat and keeps M from deviating from the equilibrium. The threat has the effect of making D* the firm dividend payout with probability 1 — F(q(D*)) and a dividend vote being called with probability F(q(D*)) as the unique equilibrium in the game. The assumption that / always asks for a vote if his initial offer is rejected, is easily jus-tified when the act of voting by I is appropriately interpreted. In practice, a shareholder such as I does not need to participate explicitly in a vote to express his disapproval of some management policy. All he need do is sell his shareholdings in a firm — vote with his feet so to say. Empirical evidence based on firms which make a significant change in their dividend policies, suggests that some shareholders do vote with their feet when the policy change is announced. Recall that the game as it is played, does not allow I to revise his initial offer D if M rejects it. If revisions to D were allowed, I would do so till M accepts his offer, thereby revealing his private information on p. In the proposed one-shot game, I has to make the "best" possible use of his only allowed move and force M to use his signal truthfully. I does so by employing a credible threat, namely, the rule of the game that M's rejection of D would result in a vote being called with probability 1. The quantities Dmax and Z)m,„ denote the upper and the lower limits on the payable dividends. There is no explicit assumption in the model which restricts Dmax since the firm can issue equity and use the proceeds to pay dividends. In practice, restrictions on the size of Dmax are likely to arise either due to bond covenants which may explicitly impose upper limits on maximum payable dividends, or from the firm's need to keep Chapter 4. A MODEL OF THE CONSENSUS DIVIDEND POLICY 47 its payouts in line with the future earnings capacity of the assets in place and its debt obligations. In either case, whether imposed by bond covenants or devised voluntarily by the firm, the quantity Dmax reflects the firm's asset size and composition, the earnings profile, whether and how much equity is sold by the firm, the existing debt burden, and its working capital needs given the firm's investment decision.8 Thus Dmax is governed by constraints which limit the size of the debt- and in vestment-financed dividend inventories. The quantity D m t n , which may take on any non-negative value including zero, in-corporates the information value of paying and maintaining a certain level of dividends. It is the minimum that must be paid out to communicate to the shareholders, and to the market at large, the absence of any unfavourable news. It embodies the reluctance on part of the management to cut dividends below their prevailing level. 9 The level of Dmin can proxy for the firm's ability, or lack of it, to provide shareholders with returns comparable with or better than their alternatives. For instance, a very low level of Dmin would indicate that the firm can earn better returns by retaining funds than shareholders can. Further, in the event that the dividends are equity-financed, Dmin can account for the costs to the firm of raising capital through the sale of equity. The above interpretations of Dmax and - D m i r j are employed to select their proxies in the model's tests in the following chapter. 4.3 Summary The proposed model, derives an equilibrium dividend policy for a firm as the outcome of a vote held among shareholders in different tax groups for dividends and capital gains. It is shown that dividend payouts are determined by the relative voting powers of 8See Kalay (1982) for a discussion of direct and indirect constraints on dividends arising from bond covenants. 9See Lintner (1965) and Kalay (1980). Chapter 4. A MODEL OF THE CONSENSUS DIVIDEND POLICY these different tax-clienteles; this is an empirically testable proposition. Chapter 5 DATA, METHODOLOGY AND EMPIRICAL RESULTS This chapter presents an empirical investigation of the main prediction of the theoretical model from chapter 4, namely that a firm's dividend payouts are in part determined by the relative voting powers of shareholder groups in different tax brackets. The empirical tests require information on corporate ownership structure, capital structure characteristics, and dividends. Data sources and sample characteristics are described in section 5.1. Section 5.2 describes the test procedures and presents the formal hypotheses to be tested. Sections 5.3 and 5.4 discuss the empirical results. The overall conclusions from the empirical analysis are summarized in section 5.5. 5.1 Data sources and sample characteristics 5.1.1 Sample size Table 5.1 provides a summary of the data gathered, their sources, the applicable periods and the total number of sample firms. The total sample consists of 822 different firms for the 13 year period 1976-1988. A l l sample firms are simultaneously found on (1) the University of Laval data file containing monthly stock returns to firms listed on the Toronto Stock Exchange, and (2) Statistics Canada's Intercorporate Ownership data base.1 The Laval data base contains approximately 60 percent of the stocks which are XI thank Prof. B. Espen Eckbo for providing me with access to his Intercorporate Ownership data base. 49 Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 50 listed on the Toronto Stock Exchange.2 Under the Corporations and Labour Unions Returns Act, Canadian corporations with assets greater than $500,000 or sales greater than $500,000 are required to report on an annual basis, the names of and the percentages owned by corporations holding 10% or more of any share class to Statistics Canada. These reports are supplemented by Statistics Canada with information from other sources such as Federal and Provincial Gazettes, F IRA submissions, Moody's, Financial Post Surveys, etc. Of the 822 firms identified as above, 320 axe listed in the Moody's International Manual. For these firms, the Manual provides information on bond covenants which may have a bearing on a firm's ability to pay dividends in a given year. For the 320 firms, data on fractions of voting rights owned by managers are obtained from the OSC Bulletin. For data typically contained in financial statements, the Report on Business database is employed. 5.1.2 S a m p l e d i s t r i b u t i o n by year, i n d u s t r y and corporate ownersh ip Table 5.2 shows the annual distribution of the sample firms. The sample period starts in 1976, which is the first year for which Statistics Canada makes the intercorporate ownership information available in machine-readable format. The last year for which selected data are available is 1988. One dimension of firm ownership likely related to taxation, is the country in which majority of the voting rights to the firm are held. If the preferences of corporate share-holders regarding dividends axe indeed driven by their tax brackets as the proposed model implies, then the tax code applicable to foreign corporate shareholders can conceivably 2Those mining and oil stocks listed on the TSE whose prices have never exceeded $5.00 are excluded from the Laval data base. These would be included should they reach that level and they would then be kept in the data base. Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 51 enhance or weaken these preferences. In the Intercorporate Ownership database, Statis-tics Canada also codes whether the firm is under domestic or foreign (typically U.S.) control, given that the controlling shareholder is a corporation. Table 5.2 shows the distribution of sample firms under domestic and under foreign control Approximately 15 percent of the sample firms are under foreign control. The sample contains typically about 700 firms per year in the "domestic control" group and approximately 160 in the "foreign control" group, with little intertemporal variation.3 Table 5.3 shows the distribution of the sample firms classified by their major 6-digit SIC code for the year 1988. The earliest year for which this classification is available is 1983.4 For both the domestically and as well as the foreign controlled firms, the manufacturing industries (SIC 10-39) have the largest representation, followed by Mining (SIC 6-9) and Finance and Insurance (SIC 70-74). The table is based on a total of 628 observations for the year 1988. Note that, although firms do not frequently change their major 2-digit SIC classification, for some firms the 1988 industry classification may differ from the classification in an earlier sample year. Tables 5.4 and 5.5 summarize the annual distribution of average corporate ownership of voting rights. The "total" column is the average total corporate ownership per firm, while the "individual" column is this total divided by the number of corporate sharehold-ers. For firms under domestic control (Table 5.4), the mean total corporate ownership is approximately 60 percent. The average number of corporate shareholders per firm is approximately 1.3, with a maximum of 10. The average ownership per individual corpo-rate shareholder is typically between 40 and 50 percent, with the median between 30 and 3Once in the sample, a firm drops out if it is delisted from the TSE or if the firm's ownership structure changes so that it is no longer required to report its ownership structure to Statistics Canada. 4The classification is based on the 1980 Standard Industrial Classification system — referred to as the SIC-C system by Statistics Canada. Classification of a firm under SIC-C may differ from that under the pre-1980 system called the SIC-E system. For convertibility between SIC-C and SIC-E classification structures and other relevant details, see Canadian Standard Industrial Classification for Companies and Enterprises, 1980, published by Statistics Canada, Standards Division, 1986. Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 52 40 percent. Thus, in the sample of domestically controlled firms, corporate shareholders are large but typically do not have absolute control. In the sample of firms under foreign control (Table 5.5), the mean total corporate ownership is approximately 75 percent, i.e., higher than in the sample of domestically controlled firms. The average number of corporate shareholders per firm is approximately 1.2, with a maximum of 9. Furthermore, the average ownership per individual corporate shareholder is typically between 60 and 65 percent. Thus, given that the firm is under foreign control, the controlling shareholder is typically another (foreign) corporation. 5.1.3 Ownership of voting rights by managers Table 5.6 shows the percent of the sample firms' voting rights typically held by managers. Data sources are various share trading reports as well as the Insiders' Reports regularly submitted to the Ontario Securities Commission by corporate managers and directors. The Manager Reports include information on trading in the firm's shares by firm man-agers and other board members. Summaries of these reports are published on a monthly basis in the OSC Bulletin. From these monthly reports, year-end share ownership by managers can be constructed. Share ownership is converted into percent ownership of voting rights for managers, using data on common stock outstanding from Report on Business data base. For one-share one-vote firms, these numbers directly provide the managers' percent ownership of voting rights. For firms restricted-voting shares, the data from Manager Reports are appropriately transformed to obtain the managers' percent ownership of voting rights. Data on the total number of shares oustanding (which are needed to compute the per-centage ownership by managers) are taken from the Report on Business. For the purpose of collecting these data, sample years are restricted to the sub-period 1978-82. In the sample of firms under domestic control, percentage ownership by managers Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 53 ranges from 0 to 93 percent, with a mean value of approximately 8 percent and a median value of 1.3 percent for the 1978-82 period. For the 61 firms under foreign control, the range of manager ownership is 0 to 67 percent, with mean values around 4 percent and a much smaller median around 0.3 percent. 5.1.4 Constraints on dividend payments Out of the 822 firms in the total sample, 320 are listed in the Moody's International Reports starting with their first issue in 1981. From these reports, information on the various classes of common stock outstanding for sample firms and details on their voting rights are obtained. Also, information regarding restrictions implied by debt covenants on cash dividends or other payments that may be made to the firm shareholders is obtained for the 320 firms. Table 5.7 shows the distribution of firms with the various restrictions — either self-imposed or implied by the firm's debt covenants. For 214 of the 320 firms listed in the Moody's, the debt consists of notes payable, short-term bank loans, long-term secured bank loans, and mortgages. The only conditions accompanying these loans are their repayment schedules and the interest rates. Another 69 firms have sinking fund provisions prescribed by their debt. For 25 firms, there are direct constraints on the amounts that may be paid in dividends, along with sinking fund provisions. The direct constraints define the maximum dividends payable so as not to exceed cumulative net earnings, cumulative proceeds from sale of equity, some fixed amounts, or some combination of these criteria. For 3 firms, the sinking fund provision is accompanied indirect dividend constraints which permit payment of dividends only if certain liquidity levels and working capital ratios are maintained. Of the remaining 9 firms, 7 have only direct constraints, 1 has only indirect constraints and 1 features sinking fund provision together with direct as well as indirect constraints. Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 54 5.1.5 Asset size and capital structure Table 5.8 shows the mean and median sizes of the total assets and total debt in the sub-sample of domestically controlled firms included in the Report on Business data base. The typical firm has total assets worth approximately 600 million dollars. The book value of total debt outstanding is on average 400 million dollars. The typical debt/equity ratio is 2.6 (median 1.3). The corresponding information on the sample firms under foreign control is listed in Table 5.9. On average, the total asset size is similar to that of domestically controlled firms, with a somewhat lower debt/equity ratio and a lower debt/asset ratio. 5.1.6 Aggregate dividend yields The empirical analysis in this study focuses on two different dividend yields: cash divi-dend yields and non-cash dividend yields. Dividend yields arising from cash distributions to shareholders are separated from those arising from stock dividends and other non-cash distributions such as rights issues. This is done in view of the different tax treatments these distributions are subjected to. While cash distributions are received tax-free by corporate shareholders, their non-cash receipts or at least parts thereof may be subject to capital gains taxes. By contrast, large non-corporate shareholders in high tax brackets get taxed at ordinary income tax rates on cash dividends whereas their non-cash receipts axe subject to capital gains taxes at lower rates. The overall trends in dividend yields in the sample firms over time are evident in some composite yields, namely yields to the equally- and the market value-weighted indices of the sample firms (see Tables 5.10 and 5.11). Each of these indices is calculated separately for the sample firms under domestic control and those under foreign control. During the 1976-86 period, the total annual dividend yields ( cash plus non-cash ) Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 55 of firms under domestic control range from 1.5% to 6% on the equally-weighted index (EWDT) and from 3.3% to 5.9% on the value-weighted index (VWDT). By comparison, the yields during the same period on firms under foreign control range from 0.3% to 7.5% on the equally-weighted index (EWFT) and from 2.5% to 7.9% on the value-weighted index ( V W F T ) . Thus there is greater variation in the dividend payouts among firms under foreign control, compared to the payouts among firms under domestic control. In either case, whether under domestic control or under foreign control, firms with larger market values appear to have higher total dividend yields on the average. This is suggested by the higher yields on the market value-weighted index compared to that on the equally-weighted index. Another set of composite yield indices looked at is the one in which the fractions of the firm market value controlled by non-corporate shareholders constitute the value weights. These are the ownership-and-value-weighted indices O V W D T and O V W F T for the total dividend yields on firms under domestic and foreign control, respectively. Their cash- and non cash-dividend yield counterparts are O V W D C , OVWDS, O V W F C and OVWFS. The ranges for O V W D T and O V W F T are respectively, from 2% to 4.1% and from 0.8% to 1.6%. A comparison of these ownership-and-value-weighted indices with their value-weighted counterparts suggests that dividend yields are smaller on the average for firms in which corporate shareholders control smaller fractions of voting rights.5 5 Note that the aggregate dividend yields to the indices of firms under domestic control and those under foreign control can not be meaningfully compared without reference to the tax codes of the countries of control. The higher observed corporate ownership of voting rights in the firms under foreign control would suggest — according to the proposed dividend model — that these firms should be paying relatively higher dividends. This tendency would be attenuated, however, by the withholding taxes charged at the sourcefor the non-Canadian recipients cf Canadian dividends. Whether the recipient shareholder gets a credit against these when paying taxes in his/her home country, would depend on whether Canada and that country have a tax treaty in effect. Indeed, the existence of such a tax treaty would neutralize the effect of the existence of a withholding tax at source, on the dividend yields of firms under foreign control. Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 56 5.2 Test strategy The proposed dividend model predicts a positive association of the voting power of cor-porate shareholders with cash dividend yields and a negative one with non-cash dividend yields, given the manner in which cash and non-cash dividends are taxed for the corpo-rate shareholders. The opposite is predicted for the association of dividend yields with the voting power of managers, i.e. positive association of managers' voting power with non-cash dividend yields and a negative one with cash dividend yields. These implica-tions of the proposed dividend model constitute the four main hypotheses that are tested in the subsequent analysis. To test the hypotheses, two techniques are employed. First, linear regressions are used to ascertain the nature of the association between ownership structure and firm dividend policy, under the assumption that the true association between these has a linear func-tional form. Specifically, the Seemingly Unrelated Regression (SUR) estimates for cash and non-cash dividend policies as dependent variables are obtained and analyzed. The explanatory variables include the voting powers of corporate shareholders and managers, and some firm financial variables. The inclusion of financial variables is warranted by the need to account for the two key features of the proposed dividend model, namely, the variables Dmax and Z? m , n . The available time series and cross-sectional data are pooled for this. Second, an attempt is made to account for the fact that the the variables Dmax and Dmin cannot be directly observed for a firm, that these two variables have some common proxies, and that these proxies may in turn be correlated with the voting power variables. Here, the Linear Structural Relations (LISREL) approach is employed to accomplish this. Further, we note that for a majority of the sample firms under foreign control, the controlling blocks of votes are held in the U.S.. The U.S. investors are levied a 15% withholding tax on dividends received in Canada. The U.S.-Canada tax treaty allows these investors to receive a tax credit against it towards their U.S. taxes. Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 57 O n c e aga in , the time-series and cross-sect ional d a t a are p o o l e d for analys is . F u r t h e r detai ls of the S U R fo rmu la t i on and the L I S R E L analys is unde r t aken are discussed next . 5.2.1 Seemingly Unrelated Regressions analysis T h e i n i t i a l e x p l o r a t i o n of the na ture of the l i n k between shareholder v o t i n g powers and d i v i d e n d payouts is conduc ted b y r u n n i n g Seeming U n r e l a t e d Regress ions ( S U R ) for the a n n u a l cash a n d non-cash d i v i d e n d yie lds D I V C and D I V S as dependent var iables . T h e e x p l a n a t o r y var iables i nc lude the Shap ley values I S H A P and M S H A P for the v o t i n g powers of the corpora te shareholders and the managers , r e spec t ive ly . 6 For the r e m a i n i n g exp l ana to ry var iables i n the S U R analys is , we look to the po t en t i a l de te rminan t s of D m a x and Dmin. 5.2.2 Accounting for Dmax and Dmin T h e S U R ana lys i s , i f done us ing on ly the vo t ing powers as e x p l a n a t o r y var iables , w o u l d not i nco rpo ra t e the role of the two exogenous var iables fea tur ing i n the m o d e l : the 6We note here that ISHAP and MSHAP are not the only possible explanatory variables which may be employed as proxies for shareholder voting power. An alternative functional form possible for the proposed link may be stated in terms of the proportions (IOWN and MOWN) of voting rights owned by the firms' corporate shareholders and their managers. Inclusion of the squared variables (IOWN-sqrd and MOWN-sqrd) and their crossproduct (IOWN x MOWN) would allow for the impact of their relative values on DIVC and DIVS to be incorporated. For the purpose of comaparing the relative performance of IOWN and MOWN with that of ISHAP and MSHAP as proxies for voting powers, SUR results using only the voting power variables are provided in Tables 5.12 and 5.13. For the analysis here, ISHAP and MSHAP are retained as the appropriate proxies for shareholder voting power based on the belief that these incorporate much more information than the quantities IOWN and MOWN do. For instance the information that when IOWN is 50% or more the corporate shareholders have absolute voting power even if MOWN may be as high as 49% is reflected in the value 1 for ISHAP and value 0 for MSHAP — something that IOWN and MOWN fail to capture. The quantities IOWN, MOWN, IOWN-sqrd, MOWN-sqrd and IOWN x MOWN cannot represent this scenario as ISHAP and MSHAP can, since the former do not incorporate the specified majority rule — a critical component of the voting power computation. Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 58 maximum payable dividend Dmax and the minimum dividend -Dm,„ (> 0) the firm must pay. While in the proposed model the investment and financing decisions of a firm are assumed not to directly affect the equilibrium dividend, their impact on the latter is directed through the limit quantities Dmax and Dmin. Neither of the two quantities Dmax and Dmin is directly observable however, so that proxies have to be employed in order to capture their effect on the equilibrium dividends. Once proxies for Dmax and D m , „ have been identified, their role in the determination of dividends in conjunction with the firm ownership structure can be explored. This is done through two alternative approaches as described next. The first approach consists of employing the proxies identified for Dmax and Dmin as explanatory variables in addition to the voting power variables in the DIVC and DIVS regression equations. The hypotheses that the SUR coefficients for the variables proxying for Dmax and Dmin are significantly different from zero, and that the coefficients of the voting power variables are as predicted by the model, can then be tested. Formally, these are stated as the set of hypotheses (HI.1,HI.2) following the discussion to identify the proxies for Dmax and Dmin. The second approach consists of using the sub-sample of absolute majority cases7 to estimate the values of the parameters which determine Dmax and Z?m,„. These estimates can then be used to calculate values of Dmax and Dmin for the non-absolute majority cases. These values together with the values of the voting power variables can then be input in the SUR analysis of dividend policy as determined by Dmax, Dmin, and the voting power variables, for the non-absolute majority cases. The following hypotheses can then 7Included in the 1188 observations available for the sub-period 1978-82, are 727 cases in which either the corporate shareholders or the managers have absolute voting power, Le. cases in which either IOWN or MOWN exceeds 50% ( or the prescribed majority), or equivalently, the cases in which either ISHAP or MSHAP has value 1. The distribution of the 1188 observations over absolute and less-than-absolute power, and over zero- and non-zero cash and non-cash dividends cases is given in Table 5.14 Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 59 be tested: that the parameter estimates for Dmax and Dmin obtained from the absolute majority cases are significantly different from zero, and that the SUR coefficients of Dmax, Dmin and the voting power variables ISHAP and MSHAP for the non absolute majority cases are significantly different from zero and of the predicted signs. Formally, these are stated as the sets of hypotheses (H2.1,H2.2) and (H3.1,H3.2) following the discussion to identify the proxies for Dmax and D m , „ . To implement the second approach, note that for the absolute majority cases the observed dividend payment has to be a measure of the quantity Dmax, or Z?m,„, or both (when Dmax=Dmin) according to the proposed dividend model. For instance, when corporate shareholders have absolute voting control then the observed dividends ought to be their most preferred payouts — Dmax. In some of these cases these may well be the minimum payouts Dmin the firm must maintain as well, for some exogenous reasons. For these cases then, a regression of the observed dividends on the proxies selected for Dmax for the absolute majority casesthen provides estimates of the parameter values (regression coefficients). These estimates can next be used to determine the values of Dmax for those cases in which neither the corporate shareholders nor the managers have absolute voting majority. Similarly, the absolute majority cases provide parameter estimates for £>m»n when actual dividends axe regressed on the proxies selected for D m t n . The parameter estimates in turn help estimate Dmin for the non-absolute majority cases. Given the values of Dmax and Dmi„ estimated as above, the dividend payments for the non-absolute majority cases can now be regressed on the limit variables Dmax and Dmin as well as the voting power variables. Once again, hypotheses regarding the signs of the voting power variables can be tested. Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 60 Proxies for Dmax: The potential proxies for Dmax would include a firm's net earnings (NE), the lagged cash dividend yield (LDIVC), the proceeds from the sale of equity (SE), total assets (TA), current assets (CA), current liabilities (CL) and the long-term debt (LD) in a given year. These would be based on the fact that these are typically the variables which determine the inventory available to a firm for payment of dividends in any given period, as implied by — say, the firm's debt constraints8, its liquidity situation, or its ability to raise funds which may be used for paying dividends. For instance, if the debt covenants require that dividends be paid only if certain min-imum working capital and liquidity conditions are satisfied, then the available dividend inventory becomes a function of the current assets and liabilities, and total assets. If on the other hand, the covenants allow the firm to pay dividends out of the inventory it has built up through accumulated earnings and sale of equity net of other debt obligations and prior dividends paid, then SE, LDIVC, TA and LD become some of the determinants of Dmax • Table 5.15 gives some statistics (mean, standard deviation, minimum, and maximum) on the above financial variables for the sample firms. Table 5.16 gives their correlation coefficients. In view of the very high correlations of current assets C A , current liabilities C L , and long-term debt LD with total assets TA (0.848, 0.871, and 0.923, respectively), only the variable TA is retained out of these four for further' analysis. The reason for retaining TA rather than any of the other three is the belief that TA would also control for size, while playing the stand-in for CA, CL and LD. In addition to the above, the standard deviation of monthly returns to the firm (STDR), and the average market value of equity for the year (MVE) are also used as 8Kalay (1982) formulates different dividend inventory constraints based on some standard features of bond covenants. Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 61 proxies for Dmax. Inclusion of STDR is based on the belief that a firm with relatively greater variation in its returns would set a relatively lower for the maximum dividends that can be/should be paid out in a given period. MVE is included as a variable which would be relevant in case decision to sell equity is made in a particular period. Thus, the variables NE, LDIVC, SE, TA, STDR, and MVE are selected as proxies for the latent variable Dmax. Proxies for Dmin: Selection of proxies is governed by the supposed information content inherent in the act of a firm making dividend payments. The signalling value of dividend payments arises from the recipients' belief that the a firm making, and maintaining, the announced level of dividend payments is in "good health". That is, the firm can afford to make the announced/expected dividend payments after taking care of its capital needs, and that it expects to be able to continue doing so. Accordingly, Dmin is the level that would convey to the recipients the very minimum dividend that the firm can comfortably support on an ongoing basis, given its current earnings ability and its ability to maintain this level of performance. It is the level which, on the one hand, firm managers would be very reluctant to cut, for fear of sending "negative" signals to shareholders about the firm's current profitability, and which on the other hand, has to convey information about the firm's future prospects. The proxies for -Dm, n have to, therefore, set a floor for the dividend payments as well as have information content for future expected dividends. The first proxy for D m 8 n then, is the dividend paid out in the last period (LDIVC). This is the minimum expected dividend payment in the future. In addition, the variables selected to proxy for the firm's ability to maintain this dividend payment include annual net earnings (NE), the asset size (TA), the average (monthly) price-to-earnings ratio (PE), standard deviation of the monthly rates of return (STDR), and the market value Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 62 of equity (MVE). Thus, Dmin is proxied by the variables NE, LDIVC, TA, PE, STDR, and MVE. We note that of these, the proxies NE, LDIVC, TA, STDR, and MVE are common between Dmax and Dmin • Test hypotheses: Having identified the proxies as above, the regression equations employed to account for D m a x and Dmin using the two alternative approaches discussed above can now be proposed. For the first approach, the SUR equations include these proxies as explanatory vari-ables for DIVC and DIVS in addition to the voting power variables ISHAP and MSHAP. Thus, DIVCjt = a0 + cuNEjt + a2LDIVCjta3SEjt + a4TAjt a5PEjt + a6ISHAPjt + a7MSHAPjta8STDRjt + a9MVEjt + e l j t DIVSjt = f3o + /3iNEjt + p\LDIVCjtp\SEjt + /34TAjt f35PEjt + p6ISHAPjt + p\ MS HAPjt p\ STDRjt + p\MVEjt + e2jt where subscripts j and t denote the firm and the time period, respectively. The set of hypotheses (H1.1,H1.2) then consists of the following: H l . l : a6 > 0, a7 < 0, and ca ^  0 for i = 1,2,3,4,5,8,9. H1.2: ft < 0, > 0, and f3,, ± 0 for 1 = 1,2,3,4,5,8,9. For the second approach, the first set of SUR equations is estimated based on the subsample of absolute majority cases as follows: Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 63 DLx = oo + ociNEjt + a2LDIVCjta3SEjt + a4TAjt a5PEjt + a8STDRjt + a9MVEjt + eljt Dmin = fr+p\NEjt + frLDIVCjtfrSEjt + /34TAjt PEit + p\ STDRjt + p\ MVEjt + e2jt The set of hypotheses (H2.1,H2.2) then consists of the following: H2.1: a,- ^  0 for i = 1,2,3,4,5,8,9. H2.2:#^0 for i = 1,2,3,4,5,8,9. Parameter estimates of the a's and the /?'s from these equations are used to estimate the values of Dmax and X)m,„ for the non-absolute majority cases, which are in turn used to estimate the second set of SUR equations based on this sub-sample of non-absolute majority cases: DIVCjt = a 0 + aiD*^ + a 2 D*.„ + a3ISHAPjt + a4MSHAPjt + eljt DIVSjt =P0+ p\D*ax + & Z & „ + p\ISHAPit + f34MSHAPjt + e2jt For these equations, the following hypotheses are tested: H3.1: a,- ^  0 for i = 1,2,3,4; a 3 > 0 and a 4 < 0. H3.2:##0 for i = 1,2,3,4; & < 0 and /34 > 0. 5.2.3 Linear structural model (LISREL) formulation The analysis proposed so far employs estimates of Dmax and Z) m ,„, the two variables themselves being directly unobservable. The unobservables Dmax and D m m have some Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 64 common proxies and, further, it is possible that some of these proxies affect the dividend policy through their effect on the ownership structure as well. Thus we have a scenario in which the three unobservable and correlated, attributes — Dmax, Dmin and the firm ownership structure (OWN), determine a firm's dividend policy. To explore this relation-ship in which the explanatory variables are at once unobservable as well as correlated but which are measurable with error through the use of proxies, the linear structural modelling approach (LISREL)9 is employed. In addition to incorporating unobservable attributes in the analysis, LISREL procedure allows the unobserved attributes and resid-uals to be correlated and to be estimated within the specified model, based on their observable proxies.10 Three LISREL runs are proposed: one with the total available sample, one with only the sub-sample of the absolute voting power cases, and the third one with only the sub-sample of the non-absolute voting power cases. As in the SUR analysis approach, it is of interest to see whether the relationship between dividend policy and ownership structure is qualitatively affected by whether, or not, either the corporate shareholders or the managers have absolute voting power. The details of the proposed runs are given below. Total sample The LISREL measurement model is specified as x = A£ + 8 where x is a 9 x 1 vector of observables (NE, LDIVC, SE, TA, PE, ISHAP, MSHAP, STDR and MVE) constituting the proxies for the 3 x 1 vector of the unobservable attributes 9See Dillon and Goldstein (1986) for a discussion of LISREL together with examples. 1 0 This is in contrast to the factor analysis approach which also allows unobservable attributes to be incorporated into observed relationships among a given set of variables. Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 65 (Dmax, Dmin and OWN). The matrix A is the 9x3 matrix of regression coefficients of x on £. The 9x1 vector 6 represents the random disturbances in the measurement of the observables. The measurement model, and specifically the matrix A represents beliefs regarding the nature of the relationships between the latent variables and the observables. The specification employed in the LISREL runs for the total sample is given in Table 5.17. Assignment of value 1 to a particular regression coefficient in A represents normalization. Assignment of value 0 on the other hand implies that the unobservable attribute is assumed not to be related to that particular observable. The entries A, — 15 in all, denote those coefficients which are allowed to be freely determined by the procedure. Thus, for the total sample, 11 restrictions are imposed on the matrix A. These are the minimum number of constraints necessary for identification of the model. Consistent with the specifications in SUR analysis, the unobservable Dmax is assumed to be determined by the financial variables NE, LDIVC, SE, TA and STDR — these being the proxies selected earlier for Dmax- hi addition, the two voting power variables ISHAP and MSHAP are allowed to load on Dmax- Similarly, the unobservable Dmin is assumed to be determined by NE, LDIVC, TA, PE and STDR — again, consistent with the earlier proxy selection for Dmin- hi addition, the voting power variable MSHAP is allowed to load on the unobservable D m , „ . The third and the last unobservable OWN — the firm's ownership structure, is proxied by TA, SE, STDR and MVE on the one hand, and the voting power variables ISHAP, MSHAP on the other. The variable TA is a stand-in for the firm size and the extent to which firm size may be a determinant of the ownership composition of a firm. SE features as a proxy since whenever equity is sold, it can lead to a change in this composition. The variable STDR accounts for a firm's riskiness and the variable MVE can affect the timing of the buy/sell decisions of the shareholders. Thus, both STDR and MVE axe relevant Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 66 to the ownership composition and are, therefore, chosen as proxies for OWN. Lastly, the variables ISHAP and MSHAP account for the voting composition within the ownership structure of a firm, which is directly relevant to the dividend decision. The structural model is specified as y = r£ + 6 where y is the 2x1 vector of the cash and non-cash dividend yields DIVC and DIVS, F — the 2x3 matrix of factor loadings, and e — the 2x1 vector of disturbance terms. The structural model specifies the relationship between the set of the observed dependent variables DIVC and DIVS and the set of the ubobservable attributes Dmax, Dmin and OWN. The measurement model and the structural model then together constitute the hypothesized model. This hypothesized model implies a certain correlation structure between all the 11 observables. A test of how close this implied correlation structure is to the actually observed one, thereby constitutes a test of the hypothesized model. Formally, the test hypothesis is stated as follows: H4: The variables ISHAP and MSHAP are significant determinants of DIVC and DIVS via their influence on the unobservable attributes Dmax, Z>m,„ and OWN, i.e. A,- 7^ 0 for i = 1,..., 16 in the design matrix A, the factor loadings 7 P g ^  0 for p, q = 1,2 in the matrix T, and coefficient of ISHAP in DIVC = 7 n A 9 + 712 A10 > 0. coefficient of MSHAP in DIVC = 712 A 1 2 < 0. coefficient of ISHAP in DIVS = 721A9 + 722 A a o < 0. coefficient of MSHAP in DIVS = 722 A 1 2 > 0. We denote by \P, 0 e and Qs respectively, the variance-covariance matrices of the vectors y, £, e and 8. The LISREL procedure allows for these matrices to be estimated Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 67 along with other parameters of the model, based on the correlation matrix for all the observables, namely, DIVC, DIVS, NE, LDIVC, SE, TA, P E , ISHAP, M S H A P , STDR, and M V E . Cases with absolute voting power The measurement matrix A for the observations in which either the corporate sharehold-ers or the managers have absolute voting power, has two possible specifications. Note that for these cases, the voting power variables ISHAP and M S H A P are perfectly cor-related, with a correlation coefficient equal to -1. Thus, only one of these two variables need be included in the analysis.1 1 The two alternative specifications are given in Tables 5.18 and Table 5.19. These 8 x 3 matrices now have only 11 constraints each, denoted by the values 1 and 0. Except for the deletion of the variable M S H A P in Table 5.18 and the variable ISHAP in Table 5.19, the design matrices are identical to that for the total sample. Further, the hypothesized links between the observables and the unobservables are also exactly as in the total sample case. 1 1 It is indeed necessary that only one of the variables be included, since inclusion of both would make the correlation matrix of the observables non-invertible, and thereby, make the LISREL procedure non-executable. Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 68 5.3 Empirical results: SUR analysis 5.3.1 Total sample A total of 523 observations on upto 320 firms, are available12 on all of the 11 variables — the two dividend yields DIVC and DIVS, the seven financial variables NE, LDIVC, SE, TA, PE, STDR and MVE, and the two voting power variables ISHAP and MSHAP. Recall that the effect of Dmax and £>m,„ on dividends is to be explored using two alternative approaches. The first of these approaches consists of using the financial variables as additional explanatory variables in SUR analysis, to serve as proxies for Dmax and Dmin. To this end, SUR coefficient estimates are obtained for DIVC and DIVS, using ISHAP, MSHAP, NE, LDIVC, SE, TA, PE, STDR and MVE as the explanatory variables (see Table 5.20). The null hypothesis that all regression coefficients are zero is rejected (F-statistic = 14.581 with p-value = 0.000). However, neither ISHAP nor MSHAP has a significant coefficient in the two regression equations, although their signs are as predicted — positive for ISHAP in DIVC and for MSHAP in DIVS, and negative for MSHAP in DIVC and for ISHAP in DIVS. Other significant explanatory variables are NE, LDIVC, STDR and MVE for DIVC ( t-values: 2.944, 11.881, -4.323 and -2.219, respectively), and SE and STDR for DIVS (t-values: 2.825 and 6.156, respectively). This leads to failure to reject the set of hypotheses (H2.1,H2.2): that at least some of the variables acting as proxies for Dmax and Dmi„ are significant determinants of DIVC and DIVS. Next, recall that the second approach to account for Dmax and Dm{n consists of using 1 2For 29 Of these 320 firms, the country of control is U.S.. For another 18 the controlling shareholders are residents of a country other than Canada or U.S.. Another 63 firms have been controlled in Canada for part of the sample sub-period 1978-82. In view of the small number of available observations, no attempt has been made to split the sample by country of control. As noted in footnote 5, ignoring the country of control should have a minimal effect on dividend policies of Canadian firms under foreign control as long as the dividend recipients' country gives a tax credit against the Canadian withholding taxes. Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 69 the absolute majority cases to estimate parameter values, and then employing these estimates to explore the link between dividends and voting power using the non-absolute majority cases. Among the 523 observations in the above analysis, 353 cases are the ones in which either ISHAP or MSHAP has value 1. Assuming that the observed DIVC = Dmax for these 353 cases, OLS regression coefficients are estimated using NE, LDIVC, SE, TA and STDR as explanatory variables. These are the proxies selected for D m a x as discussed earlier. Similarly, assuming that the observed DIVC = Dmi„, coefficients in the regression of DIVC on the selected D m i n proxies NE, LDIVC, TA, PE and STDR are estimated. Table 5.21 contains these parameter estimates. The null hypothesis that the regression coefficients of all of the selected proxies for Dmax are zero, is rejected based on the F-statistic (23.02 with degrees of freedom 5 and 348, p-value=0.000). None of the individual coefficients are significant, however, indicating the possibility of collinearity among the proxies chosen. The null hypothesis that the regression coefficients for all of the Dm{n proxies axe zero, is rejected as well based on an F-statistic of 26.98 (with degrees of freedom 5 and 348, p-value=0.000). In this case, LDIVC and STDR have significant coefficients (t-values of 8.475 and -3.056, respectively). Thus, we fail to reject the hypotheses (H3.1,H3.2): that for the sub-sample of cases with absolute majority, the hypothesized financial variables significantly determine D m a x and Dmi„. Next, given these parameter estimates the quantities Dmax and Dmin are calculated for the 170 cases out of the 523 in which neither ISHAP nor MSHAP has value 1. For the 170 cases, DIVC is now regressed on D m a x , D m i n and ISHAP and MSHAP. The results axe given in Table 5.22. The null hypothesis that all coefficients in the regression of DIVC on Dmax, D m i n , ISHAP and MSHAP axe zero, is rejected based on the F-statistic (88.754 with degrees of freedom 4 and 165, p-value=0.000). Neither ISHAP and MSHAP Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 70 enter the DIVC regression directly with significant coefficients. On the other hand, both D m a x and Dmin appear with significant and positive coeffi-cients (t-values: 16.615 and 2.287, respectively) in the regression for DIVC. Thus, while we fail to reject the hypotheses (H4.1,H4.2), the claim that voting power variables are significant determinants of DIVC and DIVS, together with D m a x and D m t n , is not borne out. 5.4 Empirical results: LISREL formulation The evidence accumulated thus far may be summed up as follows: the variables ISHAP and M S H A P have some explanatory power for firm dividend policy. However, when D m a x and D m i n are included as explanatory variables, the voting power variables are no longer significant determinants of the dividend policy. The analysis done so far does not account for the fact that the unobservables D m a x and Dmin have some common proxies, and the possibility that some of these proxies affect the dividend policy through their effect on the ownership structure, as discussed earlier. The scenario that the three unobservable, and correlated, attributes D m a x , D m i n , and the firm ownership structure OWN, determine a firm's dividend policy, is explored next using the proposed LISREL formulation. 5.4.1 Total sample Parameter estimates The measurement model employed with ISHAP and M S H A P as proxies for shareholder voting power is given in Table 5.17. Recall that a total of 11 restrictions are imposed on the factor loadings in the matrix A. These are indicated by the values equal to either 0 or 1 for the loadings, the value of 1 simply indicating normalization. Also, the measurement Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 71 errors 6 are assumed to be uncorrelated with each other, with the unobservables, and with the errors e in the structural equation. Thus 0 e and ®g are assumed to be diagonal matrices. These restrictions are the minimum necessary to achieve identification of the parameters of the model. LISREL estimates the parameters of the specified model by fitting the covariance matrix £ of the observables implied by the model to the sample covariance matrix S of the observables. This is done by minimizing the function log(defE) + i r ( S £ _ 1 ) - log(detS) with respect to the parameters of these matrices. The estimates of the parameters for the measurement model ( the matrix A are presented in Table 5.23 and those for the structural equations (the matrix T are in Table 5.24 — together with their t-values. In Table 5.25, the estimates of the variance-covariance matrix (<& of the unobservable attributes axe given. Table 5.26 shows the matrix of the normalised residuals S — £ . Estimates of the measurement model and the corresponding t-statistics show that out of the predicted variables NE, LDIVC, SE, TA, ISHAP and STDR, only TA loads significantly and in a positive way on Dmax — besides NE whose loading is normalised to have value 1. On the other hand, of the hypothesised variables NE, LDIVC, TA, PE (normalised), ISHAP, MSHAP and STDR, all but TA load significantly on D m i n . Further, out of the hypothesized variables SE, TA, ISHAP, MSHAP, STDR and MVE (normalised), none has a significant loading on the ownership structure dimension OWN. Among the unobservable attributes D m a x , Dmin and OWN, D m , „ is the only significant factor in both the structural equations — one for DIVC and the other for DIVS. The variable OWN appears in a significant way only in the DIVS equation. Thus, neither Dmax nor OWN appear to influence the cash dividend policy directly. The influence of the Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 72 shareholder voting power on cash distributions appears to manifest itself only indirectly, and only through its effect on Z?m,„. The nature of this influence can be traced through the relationship between ISHAP and Dmin (negative; see Table 5.23) on the one hand, and between -Dm,„ and DIVC on the other hand (negative; see Table 5.24). The overall effect of ISHAP on DIVC is thus significant and positive. The effect of ISHAP on DIVS can be similarly traced to be significant and negative via their link with Dmin (see Tables 5.23 and 5.24). For M S H A P , the association with DIVC and DIVS is exactly the opposite of that observed for ISHAP. Note the significant, positive loading of M S H A P on D m , „ to begin with. This, coupled with the already noted link of Z?m,„ with DIVC and DIVS, implies that M S H A P affects DIVC negatively and it affects DIVS positively. The associations noted above between the voting powers ISHAP and M S H A P and the dividends DIVC and DIVS, are consistent with the proposed hypotheses. R o b u s t n e s s For the overall goodness-of-fit of the measurement and the structural models, several indicators can be examined. For instance, for the measurement model, the squared multiple correlations are given by var(yi) var(y2) The closer these multiple correlations are to quantity 1, the closer is the hypothesized model to some true (unspecified) model. The quantities 0.518 and 0.015 thus indicate that the proposed model estimates DIVC (yi) better than it estimates DIVS (y2). Alternatively, the total coefficient of determination for DIVC and DIVS jointly is given by Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 73 1 * * ( * ) _ , 5 2 5 det(cov(y) Once again, a value of the coefficient of determination closer to 1 would indicate closer approximation of reality by the hypothesized model. For the overall model, the x 2 _ s fatistic has the value 207.27 (df=26). Based on this, the null hypothesis that the proposed model has a good fit for the observed data would be rejected. However, large sample size and departure from multivariate normality of the observed variables would tend to increase x2 beyond the values that may be expected to arise from specification error. Therefore, alternative measures independent of the sample size are looked at. The goodness-of-fit index (GFI) and the adjusted goodness-of-fit index (AGFI) measure the relative amounts of the variances and covariances for the 11 observed variables and jointly accounted for by the model. Both G F I and A G F I axe independent of the sample size and robust for departure from normality. Once again, values of these closer to 1 would indicate a better fit. For the model being analyzed here, these values are: G F I = i - ^ - ' s - p - j s , t r(S-!S) 2 AGFI = 1 - Mi±il(i _ G F J ) = . 8 3 6 ZiCL where k = total number of x and y variables = 11, and d = number of degrees of freedom = 26. Another measure — the root mean square residual ( R M R ) , is an average of the residuals S — X. For the model we have Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 74 While the GFI, A G F I and R M R measures suggest a reasonably good fit for the proposed model, potential problems are indicated by some higher than "acceptable" residuals in S — S 1 3 (see Table 5.26), model mispecification being one of them. Further, as noted earlier, LISREL allows the variances and covariances of the unob-servable attributes to be estimated with the model. The procedure does not however, impose restrictions to ensure that the resulting matrices are well behaved. For instance, the var-cov matrices obtained by the procedure are not constrained to be positive-definite, and can turn out to be non-positive definite if the model being tested is not well-specified. Indeed, the var-cov matrix obtained here for the random disturbances 6 in the mea-surement model turns out not to be positive definite — suggesting that the model being tested may be misspecified, and that the results obtained need to be interpreted with caution. The evidence regarding the hypothesis H5 is thus mixed at best. 5.4.2 Cases with absolute voting power Parameter estimates As in the SUR analysis, we next explore whether the link between dividends and voting power are driven by the absolute voting power of one or the other group of shareholders. This is done by looking at this link separately for the absolute and non-absolute voting power cases, and ascertaining if the essential nature of the link is significantly different. As described in the formulation set-up, the 353 cases with absolute voting power are analyzed with two alternative design matrices, one in which only the variable ISHAP out of the two voting power variables is used (see Table 5.18), and the other in which only M S H A P is used (see Table 5.19). The former has a total of 10 restrictions, whereas the latter imposes 11 restrictions. Except for the deletion of one of the voting power 1 3 Dillon and Goldstein (1986) suggest acceptability of normalized residuals smaller than 2 in magnitude as a rule of thumb. Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 75 variables, the design matrices in these approaches are otherwise identical to those used for the total sample. Model w i th ISHAP: Estimates of the design matrix A, the design matrix V, the correlation matrix $ and the normalized residuals S — £ when ISHAP is used, are given in Tables 5.27- 5.30. Unlike the total sample case, the estimates of the measurement model and the corre-sponding t-statistics show that none of the respective hypothesized variables load signifi-cantly on D m a x and on OWN. On the other hand, determinants of D m , n for the absolute majority cases are identical to those for the total sample, i.e the variables N E , LDIVC, ISHAP and STDR load significantly on - D m t n . In contrast to the total sample results once again, none of the unobservable attributes Dmax, D m i n and O W N have significant links with either DIVC or DIVS. Model w i th MSHAP: Next, estimates of the design matrix A, the design matrix T, the correlation matrix <fr and the normalized residuals S — S when the variable M S H A P is used, are given in Tables 5.31- 5.34. In contrast to the above approach when ISHAP is used, this time the results are iden-tical to those obtained for the total sample. That is, the only variable with a significant loading on Dmax is TA, the variables NE, LDIVC, M S H A P and STDR load significantly on D m i n , and none of the hypothesized variables load significantly on O W N . Further, Dmin is the only significant factor in both the structural equations — one for DIVC and the other for DIVS. The variable O W N appears in a significant way only in the DIVS equation. The significance of D m a x hi the DIVS equation is an improvement over the total sample case (note the t-value of 1.976 as opposed to the t-value of 1.860 for the total sample). Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 76 Robustness Model with ISHAP: Fo r the m o d e l us ing I S H A P , the squared m u l t i p l e corre la t ions are: 1 _ = .664, 1 - ^4 = 2.582(H) T h e t o t a l coefficient of d e t e r m i n a t i o n for D I V C and D I V S j o i n t l y is: 1 - , , v / . = .900 det(cov(y) W e note tha t the e s t ima ted m a t r i x \t is not pos i t i ve definite. T h i s together w i t h a squared m u l t i p l e cor re la t ion of 2.582 above impl i e s a serious p r o b l e m w i t h the fit of the m o d e l . W e repor t the r e m a i n i n g s ta t is t ics for the purpose of c o m p a r i s o n . Fo r the overa l l m o d e l us ing I S H A P , the x 2 - s ta t i s t ic has the value 130.83 (df=18) . T h e goodness-of-fit i ndex ( G F I ) and the adjusted goodness-of-fit i n d e x ( A G F I ) are: GFI = l - ^ S " 1 S - f = . 9 3 9 t r (£ - !S ) 2 AGFI = l _ (i _ G F 7 ) = . 8 1 4 2d where k = t o t a l n u m b e r of x and y var iables = 11, a n d d = n u m b e r of degrees of f reedom = 18. F i n a l l y , the root mean square res idua l ( R M R ) is Model with MSHAP: A s w i t h the pa ramete r es t imates , the robustness s tat is t ics are s imi l a r to those for the t o t a l sample case w h e n the v o t i n g power var iab le M S H A P ra ther t h a n I S H A P is used to ana lyze the absolute vo t ing power cases. T h e squared m u l t i p l e corre la t ions i n this case are: Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 77 ! _ ! ^ ! l l = . 5 8 4 l_^4 = .048 var(j/i) var(y2) The total coefficient of determination for DIVC and DIVS jointly is: 1 - =,6X1 det(cov(y) Among the various matrices estimated, the matrix 0$ is not positive definite. For the overall model using MSHAP, the x2-statistic has the value 76.10 (df=19). The goodness-of-fit index (GFI) and the adjusted goodness-of-fit index (AGFI) are: GFI = l - " - ( S - ' S - f =.960 AGFI = 1 - 1" ^ (1 - GFI) = .884 2d where k = total number of x and y variables = 11, and d = number of degrees of freedom = 19. Finally, the root mean square residual (RMR) is ™*=*£±%?$*=™ 5.4.3 Cases with non-absolute voting power Parameter estimates The measurement model used in this case is given in Table 5.17. Thus, a total of 11 restrictions are imposed on the factor loadings in the matrix A as in the case of the total sample. Estimates of the parameters for the measurement model are given in Table 5.35 and those for the structural equations (the matrix V are in Table 5.36 — together with their t-values. In Table 5.37, the estimates of the variance-covariance matrix (<fr Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 78 of the unobservable attributes are given. Table 5.38 shows the matrix of the normalised residuals S — S. Estimates of the measurement model show that out of the predicted variables NE, LDIVC, SE, TA, ISHAP and STDR, only STDR loads significantly and in a negative way on Dmax. Of the variables NE, LDIVC, TA, PE (normalised), ISHAP, MSHAP and STDR, all but MSHAP load significantly on £>m,„. Further, of the variables SE, TA, ISHAP, MSHAP, STDR and MVE (normalised), TA and STDR load positively and MSHAP loads negatively on OWN. Among the unobservable attributes Dmax, D m ,„ and OWN, both Dmax and OWN are significant determinants of DIVC — the former affecting DIVC in a positive manner and the other doing so in a negative manner. None of the attributes appears to have a link with DIVS, though. For the non-absolute majority cases then, the only way in which shareholder voting power appears to be influencing dividend policy is through MSHAP and its role in the firm's overall ownership structure as represented by OWN. Further, the overall effect of this indirect link between MSHAP and DIVC is negative — opposite to the hypothesized sign. Robustness For the sub-sample of non-absolute majority cases, the robustness statistics are similar in nature to those obtained for the total sample, showing in general a reasonably good fit for the proposed model. The squared multiple correlations for the model are: 1 -var(ei) = .984, 1 -var(e2) var(y2) = .173 var(y1) The total coefficient of determination for DIVC and DIVS jointly is: Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 79 1 - „ • ? < * > ,=.910 det(cov(y) The estimated matrix $ of correlations among the unobservables is not positive def-inite. The x2-statistic for the overall model has the value 107.29 (df=26). The goodness-of-fit index (GFI) and the adjusted goodness-of-fit index (AGFI) are: G F I = i - M s - ' s - p ' - j o g AGFI = 1 - k ( k n + , ^ (1 - GFI) = .767 2d where k = total number of x and y variables = 11, and d = number of degrees of freedom = 26. Finally, the root mean square residual (RMR) is 5.5 Summary and conclusions Analysis of data on the publicly traded Canadian firms has provided mixed evidence for testing the general assertion that a firm's dividend policy is affected by its ownership structure. The general hypotheses proposed and tested axe: (1) a firm's cash dividend payouts axe positively affected by the voting powex of its coxporate shareholders and negatively by that of its managers, and (2) the non-cash dividend payouts axe negatively affected by the corporate shareholders' voting power and positively by that of the firm managers. First, dividend yields axe regressed on Shapley values of corporate shareholders and managers, and additional financial variables suggested by existing dividend models, and Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 80 possibly proxying for the constructs Dmax and D m , „ in the proposed dividend model. In these regressions, the voting power variables do not demonstrate any explanatory power for cash or non-cash dividends. Indeed, estimated values of Dmax and Dmin (estimated as linear functions of the financial variables from those observations in which either the corporate shareholders or the managers have absolute voting power) turn out to be significant explanatory variables for cash and non-cash dividend yields. When analysis allows for Dmax and Z?m,„ to be determined endogenously as latent variables, by the financial as well as the voting power variables — as in the LISREL approach, and for Dmax and Z?m,„ to in turn affect the dividend yields, we fail to re-ject the proposed hypotheses. Both the corporate shareholders' voting power and that of the managers' load significantly on the Z ? m t n variable, which in turn has significant explanatory power for both cash as well as non-cash dividend yields — with predicted signs. Neither Dmax nor the shareholder voting powers appear to affect the dividend yields directly in general. Finally, there are some indications that the proposed model may explain cash dividend yields better than non-cash dividend yields. Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 81 Table 5.1 Data sources and sample size.1 Data Item Source No of firms Period Corporate ownership of voting rights Intercorporate ownership Reports 822 1976-88 Cash dividends and other distributions Laval data tapes 822 1976-86 Restrictions on dividend payments Moody's International Reports 320 1981 and 1982 issues Managers' ownership of voting rights OSC Bulletin and Manager Reports 320 1978-82 Common stock outstanding Report on Business and other financial data 320 1978-82 1 Not all firms appear in each of the years covered. Numbers in the "No of firms'' column indicate the maximum numbers of firms covered in the period indicated. Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 82 Table 5.2 Annual distribution of sample firms classified by country of control, 1976-1988. Total number Firms under Firms under of domestic control Foreign control Year firms (% of total) (% of total) 1976 758 566 (74.7) 192 (25.3) 1977 770 581 (75.5) 189 (24.5) 1978 780 596 (76.4) 184 (23.6) 1979 797 612 (76.8) 185 (23.2) 1980 809 639 (79.0) 170 (21.0) 1981 797 636 (79.8) 161 (20.2) 1982 757 609 (80.4) 148 (19.6) 1983 713 577 (80.9) 136 (19.1) 1984 696 561 (80.6) 135 (19.4) 1985 677 549 (81.1) 128 (18.9) 1986 662 533 (80.5) 129 (19.5) 1987 646 500 (77.4) 146 (22.6) 1988 628 477 (76.0) 151 (24.0) 1 The sample consists of a total of 843 firms which were listed on the Toronto Stock Exchange and which appeared at least once during the 1976-1988 period in the Intercorporate ownership database prepared by Statistics Canada. Under the Corporations & Labour Unions Returns Act (CALURA), firms with assets greater than $250,000 or sales greater than $500,000 are required to report the names of individuals and corporations holding 10% or more of any of their share classes, to Statistics Canada on an annual basis. In addition, StatsCan collects information on individual and corporate shareholdings below 10%, when deemed sig-nificant in determining the control of a firm. For the same reason, StatsCan may collect this information for firms which may be below their reporting thresholds under CALURA. Information on the country of the controlling shareholder is provided by the Intercorporate ownership database. A firm is under domestic (foreign) control if majority of its voting rights are held in Canada (foreign countries). Of the 843 firms in the sample, 377 remained under domestic control and 91 remained under foreign control for the entire thirteen year period. The remaining 375 firms either changed from domestic to foreign control, or vice versa, sometime during 1976-88 or were below the CALURA reporting threshold for at least one year during this period. Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS Table 5.3 Distribution of sample firms by industry classification as of 1988.1 83 Total number Firms under Firms under SIC SIC of domestic control Foreign Control Division Major Group Industry Description firms^ (% of total) (% of total) 628 (100 %) 477 (75.96) • 151(24.04) A 01-02 Agriculture & Related Services 44 24(3.82) 20(3.18) B 03 Fishing & Trapping - - -C 04-05 Logging & Forestry 1 1 (.16) 0 D 06-09 Mining, Quarrying & Oil Wells 130 94 (14.97) 36 (5.73) E 10-39 Manufacturing 151 97(15.45) 54(8.60) F 40-44 Construction 14 11 (1.75) 3(.48) G 45-57 Transportation & Storage 11 10 (1.59) 1(16) H 48-49 Communication & Other Utilities 21 19 (3.03) 2(.32) I 50-59 Wholesale Trade 36 28 (4.46) 8(1.27) J 60-69 Retail Trade 20 18 (2.87) 2(.32) K 70-74 Finance & Insurance 148 133 (21.18) 15 (2.39) L 75-76 Real Estate Operators & Insurance Agents 17 13 (2.07) 4 (.64) M 77 Business Services 19 17 (2.71) 2 (.32) N 81-84 Government Services - - -O 85 Educational Services 1 1 (.16) 0 P 86 Health & Social Services 1 1 (.16) 0 Q 91-92 Accommodation, Food & Beverage 8 5 (.20) 3 (.48) R 96-99 Other Services 6 5 (.80) 1(16) 1 The sample consists of 628 firms which were listed on the Toronto Stock Exchange and which appeared in the Intercorporate ownership database in 1988. The industry classification is based on the four-digit 1980 Standard Industrial Classification structure (SIC-C) and may differ from classifications based on the older SIC-E structure. For convertibility tables between SIC-C and SIC-E and other relevant details, see Standard Industrial Classification for Companies and Enterprises, 1980 published by Statistics Canada, Standards Division, 1986. Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS Table 5.4 Corporate ownership of voting rights in sample firms under domestic control, 1976-88. 84 Proportion of voting rights owned Number of by corporate shareholders Number of corporate shareholders total2 (%) each3 (%) Year Firms mean med min max mean med min max mean med 1976 566 1.06 1 L 1 7 53.01 50.12 .01 100 43.17 33.32 .01 100 1977 581 1.11 1 L 1 7 54.57 51.50 .01 100 44.12 32.97 .01 100 1978 596 1.19 1 L 1 8 57.51 54.29 .01 100 45.66 34.43 .01 100 1979 612 1.25 1 L 1 10 59.86 58.83 .01 100 47.42 36.37 .01 100 1980 639 1.27 1 I 1 10 60.93 58.99 .01 100 48.97 38.11 .01 100 1981 636 1.31 1 L 1 9 62.32 63.32 .01 100 50.15 40.33 .01 100 1982 609 1.33 ] L 1 9 61.89 62.28 .01 100 49.93 39.80 .01 100 1983 577 1.32 1 1 8 60.59 62.08 .01 100 47.88 37.11 .01 100 1984 561 1.34 3 I 1 8 61.76 62.96 .01 100 48.84 37.94 .01 100 1985 549 1.36 ] 1 7 64.10 65.06 .01 100 50.44 39.91 .01 100 1986 533 1.34 ] L 1 6 65.11 67.96 .01 100 51.15 41.11 .01 100 1987 500 1.34 1 1 6 67.17 72.36 .01 100 52.58 42.84 .01 100 1988 477 1.37 1 L 1 6 67.80 73.08 .01 100 53.04 42.82 .01 100 1 Data source for information on corporate ownership of voting rights in sample firms is the Inter-corporate ownership database. See footnote 1, Table 5.2. 2 Numbers under the "total" columns refer to aggregate ownership of voting rights in a firm by all corporate shareholders together. 3 Numbers under the "each" columns refer to an individual corporate shareholder's holdings of voting rights in a sample firm. Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 85 Table 5.5 Corporate ownership of voting rights in sample firms under foreign control, 1976-88. Proportion of voting rights owned Number of by corporate shareholders Number of corporate shareholders total2 (%) each3 (%) Year Firms mean med min max mean med min max mean med min max 1976 192 1.20 1 L 1 4 69.20 69.04 14.39 100 59.25 56.08 7.38 100 1977 189 1.23 1 I 1 5 70.29 70.40 .01 100 60.11 57.41 .01 100 1978 184 1.20 1 L 1 5 72.29 74.10 .01 100 63.46 63.87 .01 100 1979 185 1.21 ] I 1 5 74.32 74.93 .01 100 64.70 66.48 .01 100 1980 170 1.14 1 L 1 4 73.87 74.74 .01 100 64.73 65.25 .01 100 1981 161 1.12 1 L 1 4 73.11 73.63 .01 100 65.38 70.00 .01 100 1982 148 1.17 1 L 1 4 73.55 73.62 .01 100 65.05 68.87 .01 100 1983 136 1.18 ] 1 4 75.01 74.02 17.5 100 64.31 61.98 6.91 100 1984 135 1.29 1 1 4 76.33 76.49 7.16 100 63.02 61.23 7.16 100 1985 128 1.31 ] 1 9 77.79 81.79 3.82 100 64.31 61.03 3.82 100 1986 129 1.25 1 1 8 78.74 85.52 3.82 100 66.97 70.05 3.82 100 1987 146 1.21 1 1 4 80.75 90.97 3.82 100 69.92 72.55 3.43 100 1988 151 1.21 1 1 4 81.29 91.36 9.2 100 70.37 73.90 3.43 100 1 Data source for information on corporate ownership of voting rights in sample firms is the Inter-corporate ownership database. See footnote 1, Table 5.2. 2 Numbers under the "total" columns refer to aggregate ownership of voting rights in a firm by all corporate shareholders together. 3 Numbers under the "each" columns refer to our individual corporate shareholder's holdings of voting rights in a sample firm. Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 86 Table 5.6 Managers' ownership of voting rights in sample firms, 1978-82.1 Total number Proportion of voting rights owned of by managers (%) Year firms mean med min max Firms under domestic control: 1978 126 10.30 1.41 0 93.17 1979 128 9.15 1.32 0 93.19 1980 130 8.52 1.35 0 93.13 1981 130 7.41 1.35 0 93.44 1982 135 7.81 1.39 0 100 Firms under foreign control: 1978 59 4.20 .30 0 63.37 1979 59 5.36 .27 0 67.63 1980 60 4.46 .38 0 66.92 1981 59 3.84 .36 0 66.93 1982 61 24.41 .40 0 100 1 Data on the ownership of voting rights by managers - members of a firm's board of directors -have been reconstructed from the share trading reports submitted to the Ontario Securities Commission (OSC) and supplemented by information contained in Manager Reports — also submitted to the OSC by managers. Managers' voting rights are those directly as well as beneficially owned by them. Data on outstanding shares have been obtained from the Report on Business database. Sample firms consist of the TSE firms which are included in the Report on Business database and which have reported to Statistics Canada under CALURA at least once during the period 1978-82. Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 87 Table 5.7 Constraints on dividend payments imposed by debt covenants in the sample firms.1 Type of constraint Number of firms No constraint, explicit or implicit2 214 Sinking fund (SF) provision 69 Direct constraints only3 7 Indirect constraints only4 1 Direct constraints as well as SF provision 25 Indirect constraints as well as SF provision 3 Direct and indirect constraints and SF provision 1 Total 320 1 320 of the sample firms are covered by Moody's International Manuals starting with the year 1981. Information on debt covenants has been obtained from the 1981 and 1982 issues. 2 Typically, the debt in these firms consists of notes, bank loans and mortgages, with the only interest rates and repayment schedules in the loan terms. 3 Direct constraints specify limits on the maximum amounts that may be paid as the inventory consisting of cumulative net earnings and proceeds from the sale of equity, after other obligations such as preferred dividends have been paid. 4 Indirect constraints restrict dividend payments to only when certain debt-to-equity and liquidity ratios are maintained. Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 88 Table 5.8 Capital structure of sample firms under domestic control, 1976-88. Total assets2 Total debt2 Total number (millions) (millions) Debt/Equity2 Debt/Assets Year Firms Mean Med Mean Med Mean Med Mean Med 1976 109 366.58 116.68 270.89 59.14 2.79 1.31 .54 .57 1977 138 353.47 85.02 257.13 39.56 2.72 1.32 .55 .57 1978 202 308.94 59.67 227.73 30.07 3.46 1.44 .60 .59 1979 206 387.73 65.88 275.35 36.44 2.97 1.38 .57 .58 1980 210 469.62 86.11 334.17 45.18 2.68 1.44 .56 .59 1981 219 574.66 99.70 429.12 50.75 3.58 1.45 .57 .59 1982 224 622.03 106.88 469.11 56.65 1.50 1.39 .57 .59 1983 234 635.81 113.99 476.62 63.75 2.28 1.28 .58 .60 1984 228 745.49 132.84 558.30 67.21 5.24 1.33 .57 .59 1985 217 842.34 141.41 630.56 65.10 1.74 1.33 .56 .59 1986 208 920.05 160.17 711.80 79.40 2.28 1.24 .56 .57 1987 181 1,117.74 200.05 862.63 91.36 2.16 1.13 .53 .55 1988 7 1,083.92 276.79 953.44 123.94 9.48 .81 .52 .44 1 The sample consists of TSE firms which are included in the Report on Business database and which have reported to Statistics Canada under CALURA at least once during the period 1976-88. 2 These are book values of total assets and total debt. Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 89 Table 5.9 Capital structure of sample firms under foreign control, 1976-88. Total assets2 Total debt2 Total number (millions) (millions) Debt/Equity2 Debt / Assets Year Firms Mean Med Mean Med Mean Med Mean Med 1976 51 460.10 220.09 248.18 104.34 1.31 1.14 .51 .54 1977 52 519.66 234.35 280.55 119.43 1.31 1.20 .51 .55 1978 66 447.53 156.87 233.75 78.94 1.15 1.06 .48 .51 1979 68 537.95 163.87 291.38 93.14 1.15 1.06 .48 .52 1980 70 598.45 188.48 309.45 94.03 1.15 1.03 .48 .51 1981 70 680.87 243.79 371.37 124.96 1.36 1.13 .51 .53 1982 72 710.96 268.18 403.46 134.44 1.54 1.34 .52 .54 1983 72 758.02 238.02 417.21 114.21 1.15 1.04 .50 .51 1984 73 917.73 252.90 547.62 156.57 1.21 1.05 .50 .51 1985 70 1,223.29 289.47 811.09 152.37 1.59 1.01 .51 .51 1986 68 • 870.23 259.06 454.02 106.42 1.38 .86 .48 .49 1987 62 1,019.14 334.93 520.95 161.25 .97 .87 .47 .48 1988 1 The sample consists of TSE firms which are included in the Report on Business database, and which have reported to Statistics Canada under CALURA at least once during the period 1976-88. 2 These are book values of total assets and total debt. Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 90 Table 5.10 Annual dividend yields on firms under domestic control, 1976-86. Annual dividend yields Year R C A N 2 E W D C 3 E W D S 3 E W D T 3 V W D C 4 V W D S 4 V W D T 4 O V W D C 5 O V W D S 5 O V W D T 5 1976 0.0915 0.0264 0.0035 0.0299 0.0436 0.0043 0.0480 0.0337 0.0033 . 0.0372 1977 0.0757 0.0253 0.0039 0.0292 0.0452 0.0002 0.0457 0.0359 0.0002 0.0361 1978 0.0861 0.0326 0.0059 0.0388 0.0427 0.0127 0.0558 0.0320 0.0084 0.0406 1979 0.1187 0.0205 0.0028 0.0233 0.0439 0.0005 0.0444 0.0288 0.0003 0.0291 1980 0.1305 0.0206 0.0083 0.0290 0.0346 0.0126 0.0476 0.0226 0.0110 0.0336 1981 0.1920 0.0204 0.0120 0.0326 0.0365 0.0030 0.0395 0.0220 0.0023 0.0243 1982 0.1474 0.0248 0.0060 0.0310 0.0474 0.0114 0.0593 0.0300 0.0024 0.0325 1983 0.0958 0.0385 0.0221 0.0611 0.0321 0.0181 0.0506 0.0198 0.0090 0.0290 1984 0.0966 0.0223 0.0071 0.0296 0.0344 0.0077 0.0425 0.0227 0.0052 0.0280 1985 0.1002 0.0232 0.0006 0.0239 0.0336 0.0020 0.0356 0.0219 0.0007 0.0226 1986 0.0938 0.0137 0.0016 0.0151 0.0334 0.0004 0.0338 0.0199 0.0001 0.0200 1 For sample description, see Table 5.1. 2 RCAN = Quarterly interest rate on short-term Canadian government bonds. Data source is International Financial Statistics. 3 EWDC, EWDS and EWDT are the quarterly yields on the equally-weighted index of firms under domestic control, for cash dividends, stock dividends and total dividends, respectively. 4 VWDC, VWDS and VWDS are the analogues of EWDC, EWDS and EWDT respectively, for the value-weighted index of firms under domestic control. 5 OVWDC, OVWDS and OVWDT are the analogues of VWDC, VWDS and VWDT respectively, for the index in which the weight of a firm is the proportion of its value held by non-corporate shareholders. Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 91 Table 5.11 Annual dividend yields on firms under foreign control, 1976-86. Annual dividend yields Year R C A N 2 E W F C 3 E W F S 3 E W F T 3 V W F C 4 V W F S 4 V W F T 4 O V W F C 5 O V W F S 5 O V W F T 5 1976 0.0915 0.0313 0.0001 0.0314 0.0489 0.0001 0.0490 0.0132 0.0000 0.0132 1977 0.0757 0.0752 0.0000 0.0752 0.0386 0.0000 0.0386 0.0137 0.0000 0.0137 1978 0.0861 0.0258 0.0000 0.0259 0.0331 0.0000 0.0332 0.0124 0.0000 0.0124 1979 0.1187 0.0060 0.0000 0.0060 0.0260 0.0000 0.0262 0.0085 0.0000 0.0085 1980 0.1305 0.0067 0.0003 0.0069 0.0251 0.0006 0.0257 . 0.0071 0.0006 0.0078 1981 0.1920 0.0215 0.0002 0.0217 0.0797 0.0001 0.0798 0.0166 0.0001 0.0168 1982 0.1474 0.0049 0.0000 0.0050 0.0356 0.0000 0.0357 0.0123 0.0000 0.0124 1983 0.0958 0.0078 0.0000 0.0079 0.0295 0.0000 0.0295 0.0080 0.0000 0.0080 1984 0.0966 0.0089 0.0024 0.0114 0.0247 0.0008 0.0255 0.0078 0.0004 0.0083 1985 0.1002 0.0039 0.0000 0.0039 0.0367 0.0000 0.0367 0.0103 0.0000 0.0103 1986 0.0938 0.0034 0.0000 0.0034 0.0338 0.0000 0.0338 0.0088 0.0000 0.0088 1 For sample description, see Table 5.1. 2 RCAN = Quarterly interest rate on short-term Canadian government bonds. Data source is International Financial Statistics. 3 EWFC, EWFS and EWFT are the quarterly yields on the equally-weighted index of firms under foreign control, for cash dividends, stock dividends and total dividends, respectively. 4 VWFC, VWFS and VWFS are the analogues of EWFC, EWFS and EWFT respectively, for the value-weighted index of firms under foreign control. 5 OVWFC, OVWFS and OVWFT are the analogues of VWDC, VWDS and VWDT respectively, for the index in which the weight of a firm is the proportion of its value held by non-corporate shareholders. Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 92 Table 5.12 SUR estimates of regressions of cash (DIVC) and non-cash (DIVS) dividend yields on percent ownership of voting rights by large corporate shareholders (IOWN), managers (MOWN), IOWN-sqrd, MOWN-sqrd and IOWNxMOWN, t-values are in parentheses.1 Constant Independent variables IOWN MOWN IOWN-sqrd MOWN-sqrd IOWNxMOWN DIVC .266E-01 .181E-03 -.240E-08 .481E-06 .180E-16 .164E-10 (5.535) (.976) (-.049) (.008) (.050) (.023) DIVS .503E-01 .474E-03 .107E-07 -.676E-05 -.143E-15 .144E-02 (.395) (.962) (2.951) (-1.508) (-5.376) (.278) 1 1188 observations are available on upto 320 firms for the years 1978-82. These firms are simultane-ously found in the Laval data base, Intercorporate Ownership data base, Moody's International Manual, and Report on Business data base. 2 System ii2=.0513. 3 f-statistic (with degrees of freedom 10 and 2366)=6.699, p-value=0.000. Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 93 Table 5.13 SUR estimates of regressions of cash (DIVC) and non-cash (DIVS) dividend yields on voting powers of the large corporate shareholders (ISHAP) and managers (MSHAP). t-values are in parentheses.1 Indep. variables Constant ISHAP M S H A P DIVC .02451 .01757 .00353 (6.3353) (3.9417) (.37213) DIVS .0256401 -.01889 .03973 (2.4971) (-1.5961) (1.5791) 1 1188 observations are available on upto 320 firms for the years 1978-82. These firms are simultane-ously found in the Laval data base, Intercorporate Ownership data base, Moody's International Manual, and Report on Business data base. 2 System i?2=.0241. 3 ^-statistic (with degrees of freedom 4 and 2372)=7.250, p-value=0.000. Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 94 Table 5.14 Distribution of the 1188 sample observations by zero- and non-zero dividend yields and by absolute- and non-absolute voting power.1 DIVC DIVS =0 ^0 =0 ^0 Total ISHAP 0 6 22 26 2 28 0 < - < 1 160 290 431 19 450 1 131 579 693 17 710 Total 1188 M S H A P 0 220 653 846 27 873 0 < - < 1 73 225 288 10 298 1 4 13 16 1 17 Total 1188 1 1188 observations are available on upto 320 firms for the years 1978-82. These firms are simultane-ously found in the Laval data base, Intercorporate Ownership data base, Moody's International Manual, and Report on Business data base. Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 95 Table 5.15 Descriptive statistics for the two voting power variables ISHAP and MSHAP, the two dividend yield variables DIVC and DIVS, eleven financial variables NE, SE, TA, CA, C L , LD, P E , STDR and M V E , and their correlations.1 Variable N Mean Std Dev M I N I M U M M A X I M U M DIVC 1188 3.827E-02 4.339E-02 0.000E+00 5.527E-01 DIVS 1188 1.252E-02 1.148E-01 0.000E+00 1.972E+00 N E 1478 2.091E+07 6.779E+07 -3.693E+08 1.531E+09 SE 1469 7.711E+06 3.775E+07 -4.600E+04 8.680E+08 TA 1478 6.844E+08 2.741E+09 1.047E+05 5.363E+10 CA 1307 1.460E+08 3.087E+08 1.000E+03 3.155E+09 CL 1307 9.203E+07 2.070E+08 1.901E+03 3.783E+09 LD 1307 1.680E+08 4.847E+08 0.000E+00 7.630E+09 P E 671 2.018E+01 2.232E+01 -1.966E+01 2.988E+02 ISHAP 1357 7.734E-01 3.335E-01 0.000E+00 1.000E+00 M S H A P 1357 3.589E-02 1.508E-01 0.000E+00 1.000E+00 STDR 1329 1.101E-01 6.023E-02 0.000E+00 5.703E-01 M V E 1331 2.079E+08 6.882E+08 0.000E+00 1.313E+10 1 These are based on upto 1188 observations for upto 320 firms covering the period 1978-82. For each year and firm in the sample: NE=net earnings, SE=proceeds from sale of equity, TA=total assets, CA=current assets, CL=current liabilities, LD=long-term debt, PE=average stock price-to-earnings ratio, STDR=standard deviation of monthly stock returns in a given year, MVE=market value of equity. Data source for the financial variables is Report on Business data base. Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 96 Table 5.16 Pearson correlation coefficients.1 D I V C D I V S N E L D I V C 2 S E T A C A C L L D P E I S H A P M S H A P S T D R M V E D I V C 1.00 1 1 8 8 D I V S -.044 1 1 8 8 L 0 0 1 1 8 8 N E . 0 4 3 1 1 8 8 .001 1 1 8 8 1.00 1 4 7 8 L D I V C .344 9 0 7 ..033 8 0 7 J 0 2 1 9 3 9 1.00 9 4 0 S E -.022 1 1 8 8 .098 1 1 8 8 .332 1 4 6 9 -.023 9 3 9 1.00 1 4 6 9 T A -.014 1 1 8 8 - . 0 0 3 1 1 8 8 .338 1 4 7 8 -.024 9 3 9 J 4 0 1 4 6 9 L O O 1 4 7 8 C A - . 0 1 1 1 0 5 7 -.021 1 0 5 7 .565 1 3 0 7 -.019 8 3 1 .371 1 3 0 7 .848 1 3 0 7 1.00 1 3 0 7 C L -.031 -.013 .379 -.041 .297 .871 . 8 3 7 1.00 1 0 5 7 1 0 5 7 1 3 0 7 8 3 1 1 3 0 7 1 3 0 7 1 3 0 7 1 3 0 7 L D - . 0 3 3 .007 .370 - . 0 4 3 .342 . 9 2 3 .620 .723 L O O 1 0 5 7 1 0 5 7 1 3 0 7 8 3 1 1 3 0 7 1 3 0 7 1 3 0 7 1 3 0 7 1 3 0 7 P E -.234 .051 -.013 -.269 .068 J 0 4 T .044 .014 a 1 6 1.00 6 0 3 6 0 3 6 7 1 3 9 4 « 7 1 6 7 1 6 0 8 6 0 8 6 0 S 6 7 1 I S H A P .128 -.082 J 0 7 6 . 1 1 3 .022 . 0 4 3 . 0 9 7 .066 . 0 4 7 - . 1 7 7 1.00 1 1 8 8 1 1 8 8 1 3 3 7 9 3 1 1 3 5 7 1 3 5 7 1 1 9 9 1 1 9 9 1 1 9 9 6 3 9 1 3 5 7 M S H A P -.057 .081 - . 0 3 3 -.035 -.018 - . 0 4 3 - . 0 3 3 -.020 -.024 .241 -.494 1.00 1 1 8 8 1 1 8 8 1 3 3 7 9 3 1 1 3 3 7 1 3 3 7 1 1 9 9 1 1 9 9 1 1 9 9 6 3 9 1 3 5 7 1 3 5 7 S T D R . . 2 1 2 .177 - . 0 9 3 - . 1 8 8 -.001 -.092 - . 0 8 6 -.066 -.019 .198 -.126 4 0 7 L O O 1 1 8 8 1 1 8 8 1 2 3 5 9 3 5 1 2 5 1 1 2 5 5 1 1 1 3 1 1 1 3 1 1 1 3 6 2 5 1 2 3 2 1 2 3 2 1 3 2 9 M V E - . 0 4 9 -.013 .501 -.054 .112 .478 . 5 9 9 .510 4 3 9 .039 .067 -.035 J 0 1 7 1 1 8 8 1 1 8 8 1 2 3 7 9 3 5 1 2 5 3 1 2 3 7 1 1 1 5 1 1 1 5 1 1 1 5 6 2 5 1 2 3 4 1 2 3 4 1 3 2 9 1 Numbers below the correlation coefficients are the number of observations on which the calculations are based. 2 The variable LDIVC=one period-lagged cash dividend. Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 97 Table 5.17 LISREL measurement model (design matrix A) — Total sample and Cases with non-absolute voting power.1 N E 1 Ai 0 LDIVC A 2 A 3 0 SE A 4 0 A 5 T A A 6 A 7 A 8 P E 0 1 0 ISHAP A 9 Aio An M S H A P 0 Al2 Al3 STDR Al4 Al5 Al6 M V E 0 0 1 1 Number of observations: Total sample — 523, Cases with non-absolute voting power — 170. Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 98 Table 5.18 LISREL measurement model (design matrix A) using ISHAP — Cases with absolute voting power.1 NE 1 Ai 0 LDIVC A2 A3 0 SE A4 0 A5 TA A6 A7 A8 PE 0 1 0 ISHAP A9 Aio An STDR Al2 Al3 Al4 MVE 0 0 1 1 Number of observations = 353. Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 99 Table 5.19 LISREL measurement model (design matrix A) using M S H A P — Cases with absolute voting power.1 NE 1 Ai 0 LDIVC A 2 A 3 0 SE A 4 0 A 5 TA A 6 A 7 A 8 P E 0 1 0 M S H A P 0 A 9 Aio STDR An Al2 Al3 M V E 0 0 1 Number of observations = 353. Table 5.20 SUR estimates of regressions of cash (DIVC) and non-cash (DIVS) dividend yields on voting powers of the large corporate shareholders (ISHAP) and of the managers (MSHAP) and the financial variables NE, LDIVC, SE, TA, PE, STDR and MVE. t-values are in parentheses.1 Independent variables Constant NE LDIVC SE TA PE ISHAP MSHAP STDR M V E DIVC .3161E-01 •6760E-10 .3962 -.23O9E-10 ..2430E.12 -.9007E-04 .3907E-02 -.6101E-02 -.9944E-01 -.4054E-11 (5.6942) (2.9449) (11.881) (-.9006) (..1786) (-1.6250) (.8184) (-.6037) (.4.323) (.2.219) DIVS ..5691E-01 ..3848E-11 .4402E-01 .3732E-09 ..3383E-U ..6797E-04 -.8907E-02 .6207E-01 .7294 -.5148E-11 (-1.989) (-.0325) (.2563) (2.826) (-.483) (-.238) (-.362) (1.192) (6.156) (-.547) 1 523 observations are available on upto 320 firms for the years 1978-82. These firms are simultane-ously found in the Laval data base, Intercorporate Ownership data base, Moody's International Manual, and Report on Business data base. 2 System fl2=.3946 3 F-statistic (with degrees of freedom 18 and 1028) = 14.581, p-value = 0.000. Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 101 Table 5.21 OLS regression estimates of parameters for calculating Dmax and D m t n , based on 353 absolute voting power cases i.e. cases for which either ISHAP=1 or MSHAP=1. t-values are in parentheses. Independent variables Constant NE LDIVC SE TA PE STDR R2 F-Statistic Dmax .0257 2.87E-11 .3615 -3.63E-12 -1.77E-12 -.93E-01 .1988 23.021 (.0024) (2.19E-11) (.0388) (2.91E-11) (1.19E-12) -3.323 .0401 -.347 .3335 - -1.03E-12 -.0001 -.1091 .2253 26.9741 8.796 -.028 8.457 -.984 -1.824 -3.056 1 each with degrees of freedom 5 and 348, p-value = 0.000. Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 102 Table 5.22 Coefficients of OLS regressions of DIVC on estimated D m a x , estimated D m i n , and shareholder voting powers (ISHAP and MSHAP) for the 170 cases of non-absolute voting power (neither ISHAP nor M S H A P has value 1). t-values are in parentheses. Independent variables Constant Dmax ISHAP MSHAP R2 F-statistic DIVC -.0686 2.5940 .3773 -.0025 -.0152 .6827 88.7542 (-8.713) (16.615) (2.287) (-.450) (-1.686) 2 with degrees of freedom 4 and 166, p-value =0.000. Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 103 Table 5.23 Estimates of the LISREL measurement model (design matrix A) for the total sample, t-values are in parenthese.1 Dmax Dmin OWN NE 1.000 -.283 .000 - (-2.326) -LDIVC .110 -1.805 .000 (.832) (-6.938) -SE 20.579 .000 -14.435 (.196) - (.187) TA 1.022 -.084 -.062 (5.406) (-.714) (-.399) PE .000 1.000 .000 ISHAP .161 -.640 .043 (1.044) (-4.190) (.367) MSHAP .000 .452 -.102 - (3.165) (-1.606) STDR .037 .826 -.122 (.232) (5.014) (-1.034) MVE .000 .000 1.000 Number of observations = 523. Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 104 Table 5.24 Estimates of the LISREL structural equations (design matrix T) for the total sample, t-values are in parenthese.1 Ana, Dmin OWN DIVC .131 -1.842 -.002 (.719) (-6.894) (-.014) DIVS .304 .315 -.262 (1.860) (2.069) (-2.074) Number of observations = 523. Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 105 Table 5.25 Estimate of the LISREL variance-covariance matrix of the unobservable attributes (design matrix <&) for the total sample, t-values are in parenthese.1 Dmax Dmin O W N D 379 (5.966) D m i n .047 .157 (1.582) (3.964) OWN .517 .064 .729 (10.082) (1.640) (4.392) 1 Number of observations = 523. apter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS Table 5.26 Estimates of the normalized residuals S — S for LISREL for the total sample. DIVC DIVS NE LDIVC SE TA PE ISHAP MSHAP STDR M V E DIVC -.002 DIVS -.003 .000 NE .027 .463 .000 LDIVC .759 1.176 -.321 .001 SE -.016 .241 -.080 .095 -.026 T A -.067 .111 -.139 -.163 -.082 .003 PE 1.059 .110 -.357 .198 .734 .371 .002 ISHAP -1.041 -1.158 -.060 -1.244 -.332 -.331 -1.987 -.002 MSHAP 1.486 1.354 .314 1.464 -.536 -.250 4.028 -10.180 -.003 STDR .352 3.142 -1.108 .833 -.707 -1.663 1.697 -1.018 -1.148 M V E .071 -.036 .071 .122 .034 .020 -.113 -.129 .223 Number of observations = 523. Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 107 Table 5.27 Estimates of the LISREL measurement model (design matrix A) using ISHAP for cases with absolute voting power, t-values axe in parenthese.1 Dmax OWN NE 1.000 -.529 .000 - (-3.010) -LDIVC .075 -1.794 .000 (.561) (-5.466) -SE -4.249 .000 4.402 (-.682) - (.738) TA .607 -.099 .303 (.469) (-.654) (.244) PE .000 1.000 .000 ISHAP -1.491 -.380 1.466 (-.558) (-1.979) (.574) STDR -7.634 1.181 7.224 (-.733) (2.988) (.725) MVE .000 .000 1.000 Number of observations = 353. Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 108 Table 5.28 Estimates of the LISREL structural equations (design matrix T) using ISHAP. for cases with absolute voting power, t-values are in parenthese.1 Dmax Dmin OWN DIVC 18.222 -2.724 -17.277 (.273) (-.953) (-.271) DIVS -81.384 3.834 77.590 (-.273) (.309) (.272) Number of observations = 353. Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 109 Table 5.29 Estimate of the LISREL variance-covariance matrix of the unobservable attributes (design matrix 3?) using ISHAP for cases with absolute voting power, t-values are in parenthese.1 OWN D m a x .642 (6.597) Dmin -079 .153 (1.514) (3.185) OWN .669 .075 .698 (9.461) (1.511) (8.472) Number of observations = 353. Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 110 Table 5.30 Estimates of the normalized residuals S — S for LISREL using ISHAP for cases with absolute voting power.1 DIVC DIVS NE LDIVC SE T A P E ISHAP STDR M V E DIVC .012 DIVS .002 -.062 N E .108 .656 -.041 LDIVC .319 .106 -.448 .004 SE -.040 -.141 2.438 -.571 .016 TA -.067 .114 -.746 .108 1.957 019 583 .026 223 .477 .028 P E .708 ISHAP -.785 STDR .097 M V E -.039 .615 -.220 -.413 -.435 .088 -.383 -.633 .178 -.094 1.407 -.863 .152 .526 .127 .302 -2.835 .198 .038 .211 -5.383 -.378 1.043 .331 -.082 1 Number of observations is 523. Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS Table 5.31 Estimates of the LISREL measurement model (design matrix A) using M S H A P cases with absolute voting power, t-values are in parentheses.1 Dmax Dmin OWN NE 1.000 -.429 .000 - (-3.198) -LDIVC .106 -1.702 .000 (.848) (-5.611) -SE 3.599 .000 -2.552 (1.329) - (-1.119) TA .826 -.040 .083 (5.231) (-.328) (.650) PE .000 1.000 .000 MSHAP .000 .397 -.040 - (2.266) (-.643) STDR -.085 1.036 .015 (-.468) (4.614) (.106) MVE .000 .000 1.000 Number of observations = 353. Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS Table 5.32 Estimates of the LISREL structural equations (design matrix T) using M S H A P cases with absolute voting power, t-values are in parentheses.1 OWN DIVC .180 -2.018 -.033 (.863) (-5.460) (-.239) DIVS .361 .476 -.346 (1.976) (2.299) (-2.328) Number of observations = 353. Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 113 Table 5.33 Estimate of the L I S R E L variance-covariance matrix of the unobservable attributes (design matrix $ ) using M S H A P for cases with absolute voting power, t-values are in parentheses.1 Dmax Dmin O W N D m a x .580 (6.799) D m i n .068 .150 (1.565) (3.202) O W N .675 .082 .922 (9.814) (1.654) (6.516) Number of observations = 353. Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 114 Table 5.34 Estimates of the normalized residuals S — £ for LISREL using MSHAP for cases with absolute voting power.1 DIVC DIVS NE LDIVC SE TA P E MSHAP STDR M V E DIVC .000 DIVS .000 .000 NE .009 .010 .000 LDIVC .030 .072 -.028 .000 SE .001 -.001 -.001 .010 .000 TA -.004 -.018 .001 -.010 .000 .000 P E .048 -.009 -.017 -.026 .038 .009 .000 MSHAP .042 -.042 .019 .026 -.032 -.010 .296 .000 STDR .017 .200 -.027 .061 -.001 -.033 .029 .023 M V E -.001 .001 .001 -.001 .000 .000 -.011 -.005 Number of observations = 353. Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 115 Table 5.35 Estimates of the LISREL measurement model (design matrix A) for cases with non-absolute voting power, t-values are in parentheses.1 Dmax Dmin OWN NE 1.000 .061 .000 - (.023) -LDIVC .277 -1.159 .000 (1.590) (-1.450) -SE -.100 .000 .424. (-.252) - (1.117) TA .268 .023 .646 (.844) (.031) (2.150) PE .000 1.000 .000 ISHAP -6.918 -7.546 7.846 (-.875) (-.387) (.908) MSHAP .000 1.614 -.556 - (3.549) (-2.714) STDR -1.491 .303 1.271 (-2.414) (.078) (2.170) MVE .000 .000 1.000 Number of observations = 170. Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 116 Table 5.36 Estimates of the LISREL structural equations (design matrix T) for cases with non-absolute voting power, t-values are in parentheses.1 D m a X Dmin OWN DIVC DIVS 4.291 (2.151) -1.593 (-1.618) -1.005 (-.091) .633 (.153) -3.625 (-2.054) 1.245 (1.442) Number of observations = 170. Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 117 Table 5.37 Estimate of the LISREL variance-covariance matrix of the unobservable attributes (design matrix $ ) for cases with non-absolute voting power, t-values are in parentheses.1 Dmax Dmin O W N Dmax .556 (.966) Dmin -108 .094 (.424) (1.884) OWN .600 .157 .693 (1.423) (2.235) (5.849) Number of observations = 170. Chapter 5. DATA, METHODOLOGY AND EMPIRICAL RESULTS 118 Table 5.38 Estimates of the normalized residuals S — S for cases with non-absolute voting power, t-values are in parentheses.1 DIVC DIVS NE LDIVC SE DIVC .001 DIVS .001 .000 NE -.105 .696 -.012 L D I V C 3.294 .952 .021 .001 SE .603 1.469 1.918 .004 .001 TA -.061 -.062 -.425 -.593 1.309 PE -.559 -.322 -1.649 -2.524 .213 ISHAP .052 .187 .400 -2.618 -.458 MSHAP 2.162 -.176 .666 .887 -.114 STDR .617 1.687 -.424 -1.355 -.177 M V E .244 -.564 .345 -.693 -1.994 TA P E ISHAP MSHAP STDR M V E -.004 -.848 -.014 -.032 -.759 -.024 .430 -.397 -1.257 .000 -.322 1.461 .218 -2.595 -.001 .209 -1.108 -.229 -.070 .206 Number of observations = 170. Chapter 6 CONCLUSIONS Explorations of the effects of how votes are distributed among a firm's shareholders on major firm policies have begun only recently. In particular, theories to explain why firms continue to pay dividends have so far tended to ignore the possibility that the observed dividend policies may be the "consensus" outcomes arrived at through a vote among shareholders. Such a vote would become critical to arriving at some consensus when dis-tinct tax-clienteles with different policy preferences coexist among a firm's shareholders, as they would when opportunities for tax arbitrage are absent. The present study has investigated this link between dividend policies of firms and the distribution of voting power among its shareholders. A model of the consensus dividend policy is proposed in which two major voting blocks play a voting game. The two voting blocks constitute two identifiable tax-clienteles among the firm's shareholders. That is, the two voting blocks have opposite preferences between cash dividends and capital gains. The incentive to play the voting game arises when it is not known which of the two blocks will attract the remaining votes, these being held by a large number of anonymous small shareholders. The outcome of the voting game is then a dividend level somewhere in between the two extremes preferred by the major blocks. Some empirically testable implications of the model are immediate. The greater the voting power of the large corporate shareholders, ceterus paribus the higher the cash dividend payouts and the lower the non-cash payouts. By contrast, the higher the voting power of the firm managers, ceterus paribus the lower the cash dividends and the higher 119 Chapter 6. CONCLUSIONS 120 the non-cash payouts . O f course, a test of any i m p l i c a t i o n of the m o d e l is rea l ly a j o in t test of the i m p l i c a t i o n and the basic m o d e l a s s u m p t i o n that d i s t inc t tax-cl ieneteles coexis t a m o n g firm shareholders to beg in w i t h . Tests of the m o d e l are ca r r i ed out us ing d a t a on firms w h i c h have t r aded on the T o r o n t o S t o c k E x c h a n g e d u r i n g 1976-88. For the e m p i r i c a l i nves t iga t ion , the two major v o t i n g b locks are ident i f ied as the large corpora te shareholders of these T S E firms and the firm managers . T h e C a n a d i a n t a x laws endow the former w i t h a preference for cash d iv idends ( D I V C ) and the la t te r w i t h a preference for non-cash d iv idends ( D I V S ) w h i c h w o u l d be t a x e d at the i r c a p i t a l gains t a x rate. Fu r the r , specific firm charac ter i s t ics are ident i f ied for use as observable proxies for the m o d e l var iables D m a x , Dmi„ and the ownersh ip s t ruc ture . T h e i r i m p a c t on cash and non-cash d iv idends is t hen e m p i r i c a l l y inves t iga ted for the sample firms. Inc luded among the proxies for the ownersh ip s t ruc ture axe the re la t ive v o t i n g powers of managers and the corpora te shareholders . A s a measure of the v o t i n g power of these two b locks , thei r Shap ley values ( M S H A P a n d I S H A P , respec t ive ly) are employed . Resu l t s f rom the e m p i r i c a l analysis show that for the t o t a l sample , managers ' and corpora te shareholders ' v o t i n g powers influence D I V C and D I V S i n a manne r p red ic t ed by the p roposed m o d e l . T h i s is evident f r o m the fact that b o t h I S H A P a n d M S H A P appear to be s ignif icant de te rminants of the var iab le Dmin, a n d that Dmin i n t u r n has s ignif icant effect on D I V C and D I V S . T h e same does not h o l d however , for the sub-sample of those cases i n w h i c h nei ther the managers nor the corpora te shareholders have absolute v o t i n g power . T h e result for the t o t a l sample thus appears to be d r iven b y the presence of those cases i n w h i c h one or the other of the two large v o t i n g b locks has absolute v o t i n g power . It mus t be, however , tha t i n the absolute power cases the d i v i d e n d p o l i c y is d r i ven b y factors other t h a n shareholders ' re la t ive v o t i n g powers. T h u s , the evidence f r o m the da t a fails to suppor t the major i m p l i c a t i o n s of the m o d e l . Chapter 6. CONCLUSIONS 121 Three potential explanations for the failure to accept the model axe offered. The first one is the possibility that the lack of evidence supporting the model is really a rejection of the simultaneous existence of distinct tax-clienteles among firm shareholders. If true, this would amount to irrelevance of dividend policy for different shareholder voting blocks in which case, search for a model of corpotate dividend policy has to start afresh. Assuming tax-clienteles do exist, a second explanation is an acknowledgement of the possibility of misspecification. Misspecification in the model may have arisen from any or all of the following of its features. Recall that the consensus dividends have no information content in the proposed model, the vehicle chosen for this purpose being the minimum payable dividend instead. Although a rather brave assumption in view of the evidence to the contrary, it reflects the view that information content of dividends is critical only at the time when policies axe being changed significantly rather than on an ongoing basis. Also reflected in it is the assumption that the distribution of voting rights is given and is independent of the firm's dividend policy. If information content of dividends is indeed relevant on an ongoing basis and if indeed dividend policy itself is a determinant of the ownership structure, contrary to the assumptions then the data reflect this, giving rise to results which fail to support the model. Finally, even if the model were to be truly reflective of the reality, it has been tested using proxies for the various attributes and the variables involved. If the proxies axe inappiopxiate or if some important proxies have been missed, then the misspecification error would bias the results and render the hypothesis tests invalid because of the unduly high variances in the relevant variables. Potential problems such as above axe expected to be xesolved in future extensions of the present study. At the theoretical level, the potential interdependence of dividend policy and ownership structure will be incorporated into the voting game. One way this may be done is by somehow specifying the value functions Yj and YM- This exercise Chapter 6. CONCLUSIONS 122 would also yield an explicit expression for the equilibrium dividend D* as a bonus. At the empirical level, some refinement of the major voting blocks among firm share-holders will be attempted. In the present study, voting blocks such as some pension funds which have total tax-exempt status are either included in the voting block I or in the ocean OC depending on their reporting status with Statistics Canada under CALURA. By implication, they are either endowed with a preference for dividends or are treated as being too small to be a viable voting block by themselves although their tax status would really make them indifferent between dividends and capital gains. Thus the role such blocks might play in the kind of voting game proposed here is not clear and needs to be investigated further. Appendix A TAXATION OF INVESTMENT INCOME A . l The Canadian Tax System As in most tax systems, all investment income is taxed in Canada — at progressive indi-vidual tax rates and with interest income, capital gains and dividends taxed at different rates. The essential details of investment income taxation in the Canadian system are provided below for corporate and individual investors resident in Canada as well as for non-resident investors. Also noted are the major changes in the tax code which have been legislated over the years.1 Taxation of corporate investors: Interest income received by corporations resident in Canada is taxed at the same rate as their income from operations. Capital gains are taxed only when realised i.e. when the asset is sold. Prior to 1972, capital gains were not taxed at all. Since 1972, only half of the capital gains are added to taxable income, effectively reducing the capital gains tax rate to half that of the rate for income from operations. Proposed amendments will increase the taxable portion of capital gains to two-thirds for 1988 and 1989, and to three-quarters after 1989. Cash dividends received by a Canadian public corporations are not taxed at all. The tax treatment of stock repurchases and stock dividends on the other hand, is a little more involved. Tax treatment of the former depends on the procedure used for repurchase. If the stock issuing firm executes the repuchase on the open market, then the seller receives 1Sources: (i) Canadian Income Tax Act with Regulations, different years covering 1970's and 1980's, C.C.H. Publishing, (ii) Tax Principles to Remember: An Introductory Course in Canadian Income Tax, Geoffrey M.Colley, published by The Canadian Institute of Chartered Accountants. 123 Appendix A. TAXATION OF INVESTMENT INCOME 124 a capital gains treatment since it cannot be established whether the buyer is general public or the firm. If instead the shares are tendered directly to the issuer then those sellers dealing at arm's length with the issuer get a capital gains treatment. For the rest the sale is treated as a combination of a dividend, a capital gain and a return of non-taxable paid-up capital. The overall effect of these rules is to make non-cash distributions to the corporate investor, taxable at a rate somewhat lower than that for capital gains but not tax-free as cash dividends are. Prior to 1977, stock dividends were treated as ordinary dividends. During the 1977-85 period, stock dividends were treated as shares acquired at zero cost. In effect, stock dividends were taxed as capital gains in the event of being sold. Since 1985, stock dividends are once again treated as ordinary cash dividends. Taxation of foreign investors is in general governed by the Canadian tax code ex-panded to impose levies on the investment income such as withholding taxes. The extent to which these additional levies prove to be onerous for the foreign investors depends however, on whether Canada has a tax treaty with the investor's country or not and how generous the terms of the treaty are. For instance, foreign investors are levied a 10-15% withholding tax on interest income2 and a 25% withholding tax on cash divi-dends received from Canadian sources. Under the US-Canada tax treaty however, the US investors receiving the interest and dividends can use the withholding tax to generate a tax credit against their US taxes. Also, till 1985 US investors were tax exempt from Canadian taxes on capital gains realised in Canada and were thus not subject to double taxation. The provision has been dropped from the revised treaty since 1985. Taxation of individual investors: Individual Canadian investors receive the first $1000 of interest and dividend income tax-free. Interest income beyond this amount is 2 Certain federal and provincial government and government-guaranteed bonds are exempt from the withholding taxes. Appendix A. TAXATION OF LNVESTMENT INCOME 125 taxed at the same rate as ordinary wage and salary income. For individuals the tax treatment of capital gains has undergone two major changes in Canada, once in 1972 and then in 1985. While not taxed at all prior to 1972, capital gains were taxed at half the ordinary income tax rate during 1972-85. Since 1985, individuals are allowed a life-time exemption of taxes on the first $100,000 of capital gains; the gains also include those accumulated between 1972 and 1985 as long as the asset is sold in 1985 or later. The tax-deductibility of capital gains together with the progressive tax rate structure for individual investors has the effect of generating preferences for dividends over capital gains among those in the low federal tax bracket. For those in the high federal tax bracket on the other hand, this has the effect of making capital gains more attractive compared to cash dividends. Tax treatment of stock repurchases and stock dividends is is the same for individuals as for corporations. In particular, the 1985 change in the tax code has eliminated the capital gains treatment of stock dividends for individuals. Thus stock dividends received by individuals are taxed like cash dividends since 1985. An additional feature of individual taxation in Canada is the dividend tax credit the objective of which is to reduce the effect of double taxation of shareholder income which occurs since dividends are paid out of after-tax corporate profits. While the actual amount of the tax credit has varied over the years, the underlying principle allows div-idends to be effectively taxed at the personal rather than at the corporate level. When first introduced in 1947, the dividend tax credit was calculated as a flat 20% of the div-idend income. In the current tax code, for dividend D paid by a firm, the shareholder is allowed a dividend tax credit amount equal to Tc(^^r) where Tc is the corporate tax rate. This tax credit applies to the tax payable on the amount C = JHJT which is referred to as the "grossed up" dividend and is the corporate cash flow required to support the dividend payment D. For the period 1972-77, dividends were grossed up by a factor of Appendix A. TAXATION OF INVESTMENT INCOME 126 1/3 and the tax credit was 20% of the grossed up amount. Since 1978, the gross up factor has been 50% and the tax credit amount 221% thereof. In practice, the calculation of individual taxes on investment income is complicated however, by the fact that in Canada the personal taxes are assessed at the federal and the provincial levels, that the provincial tax is a percentage of the federal tax, and that the corporate tax itself depends on whether the corporation is taxed as an ordinary business or as a small business. For instance, the 50% gross-up and the 221% of gross-up as tax credit imply a corporate tax rate of 331% and a provincial personal tax rate of 47% of the federal rate. In actual fact, small businesses pay taxes at 25% and ordinary businesses are taxed at 46%. Further, the provincial personal taxes vary between 38% to 60% of the federal taxes. In effect then, the individuals receiving dividends form small businesses are over-compensated for double taxation while investors with ordinary businesses are only partially compensated. In 1983, a dividend distribution tax was instituted for small businesses to eliminate this over compensation of the double taxation, but the tax has been dropped starting with the 1987 tax year. Appendix B SHAPLEY VALUE IN OCEANIC GAMES B. l Distribution of Voting Power Among Firm Shareholders Le t v denote the charac te r i s t ic func t ion for an n—person coopera t ive game i n w h i c h players a t t empt to a r r ive at a decis ion t h r o u g h a s imple yes—no vote , and i n w h i c h each person has one and o n l y one vote. Fo r a coa l i t i on S of s players a n d the c o a l i t i o n S of the r e m a i n i n g n — s p layers , v(S) denotes the un ique value to S of p a r t i c i p a t i n g i n the two—person zero—sum game of coa l i t i on S versus c o a l i t i o n S. Fo r p layer i, the expec ted u t i l i t y f rom p a r t i c i p a t i n g i n the game — his Shap ley value, is g iven by *(»)= £ ' K r e " f " 1 ) ! HS U {i} - v(S)} SCN-{i} n -where N is the g r a n d c o a l i t i o n of a l l the n p layers . T h e s u m m a t i o n covers a l l possible coa l i t ions S of s players e x c l u d i n g the p layer i. Fo r this f o r m u l a t i o n to be v a l i d , the charac te r i s t i c func t ion v is no rma l i s ed and is assumed to be supe radd i t ive . A l s o , i t has to satisfy the efficiency, s y m m e t r y and aggregat ion a x i o m s . 1 Shap ley value has a s imple i n t e rp re t a t i on : i t is the s u m of the con t r ibu t ions player si's a d d i t i o n is expec ted to make , 1 N o r m a l i s a t i o n : v(Q) = 0. S u p e r a d d i t i v i t y : v(S U T ) > v(S) + v(T)VS,T : S n T = 0. E f f i c i e n c y : J2ieN Mv) = V(N)-S y m m e t r y : 4>w(i) = 4>i{v) f° r permutation T  of the players. A g g r e g a t i o n : <f>i(v + u) = 4>i(v) + <fii(u) for any other characteristic function u. 127 Appendix B. SHAPLEY VALUE IN OCEANIC GAMES 128 towards transforming any possible coalition S into a winning one. It must equal the sum of the amounts of expected utilities i derives from joining all possible coalitions S. The quantity "H"-*- 1)' ^ the probability of coalition S forming and [v(S U {i} — v(S)} is what player i adds to the value of S. The vector [<j)i(v), (j>2(v), , <j>n{v)} of Shapley values for the n players represents the value to each, of participating in the game and the total value of the game is Y^ieN&iv)-The characteristic function v simplifies considerably when all value from participating in a game derives from winning while losing yields no value. Thus, in a weighted game represented by [c;w1,w2,..., wn], there are n players with the ith player carrying weight to,-and c specifies the majority quota. Any coalition with weights adding up to or exceeding c becomes the winning coalition. The characteristic function then is { 1 if Lies™. > c; v(S) = [ 0 if EiesWi < c. The Shapley value for the ith. player in a weighted majority game then becomes = ^ ^ SCN-{i} n-In the context of voting games, this representation of the ith player's Shapley value suggests the interpretation of <f>i(v) as the power of the ith player's vote. Each of the terms in the above summation stands for the probability that a coalition S will be formed and that by joining it, the ith player will transform it from a losing coalition into a winning majority. The summation is over all those coalitions for which the ith player is pivotal in achieving the majority quota c. In the above representation of the weighted majority games [c;wi,W2,-..,wn], the n players carry different weights. Equivalently, one might look at these games as being played by — say, m players who axe organised into n blocks and where the ith block consists of fraction of the m players. The blocks are like coalitions consisting of Appendix B. SHAPLEY VALUE IN OCEANIC GAMES 129 "identical" individuals who have committed themselves to a common strategy in the game and will participate in other coalitions only as a unit. There are many reasons why blocks such as these may arise. For instance, players within a block may make/receive "payments" to each other on the side to keep the block from breaking up. In the context of the voting games, a large fraction of the votes may actually be owned by one individual, An oceanic game, represented by (c;wi,W2,...,wn;a), is a large (number of players —> oo), weighted majority game. In an oceanic voting game, there are infinitely many players and a finite few (n) of them control a sizeable fraction (Y^,iwi) °f the votes. The remaining votes — fraction a, are held by a continuum of infinitesimal, minor players. Treating an oceanic game as a limiting case of a weighted majority game, Milnor and Shapley (1978) and Shapley and Shapiro (1978) employ the "pivotal player" approach and obtain the power index of the vote held by a major player as his Shapley value. The expression for a major player's Shapley value in an oceanic game is provided by the following result THEOREM (Shapiro and Shapley) For each major player, the value of the game converges to the limit £ Pps{±-p)n-a-ldp (B.9) SCN-{i} J t 2 where N = set of major players, {i} = set of major player i, s = no. of major players in coalition S, w(S) = Y^,ieswi = s u m °f weights of major players in S, p = a continuous variable on [0,1], h =< c-=^ >, t2 =< >, and • o if x < 0 < x >= < x i f 0 < x < l .1 if x > 1 The value of the game for the ocean as a collective is defined by the equation: * = 1 - X > (B.10) ieN Appendix B. SHAPLEY VALUE IN OCEANIC GAMES 130 Expressions B.9 and B.10 allow the voting powers of the major players and that of the oceanic collective of the minor players, to be calculated for a firm when the fractions of voting rights held by large corporate shareholders are known. In (12), fa is a sum of several integrals, each of which refers to a possible coalition of the major players without the player i, calculation of which can be programmed. For instance when there are three major players then the value of fa i = 1,2,3, is obtained by setting the weight of a fourth player to be zero. Probabilities that the various possible coalitions will formed can then be calculated and summed up. In a manner similar to the three major player game, voting powers in the more numerous or fewer major player games can be obtained. The richness of the notion of oceanic games is best illustrated by looking at a corpo-ration with two large stockholders (Appendix to Milnor and Shapley (1978)) and a great number of very small interests. Suppose the vote outcome at a shareholders' meeting is to be determined by a simple majority quota. We are thus looking at an oceanic game specified by [1/2;w\,w2;a] where a = 1 — W\ — w2. The power index of the first large stockholder as per the Shapiro and Shapley Theorem above, is given by: oc>\ Region I; < i,w2 < § , « < § Region II; 1 ^ i > § Region III; 0 w2 > \ Region IV. The second player's power index fa consists of similar expressions. The power of the ocean is given by $ = 1 — fa — fa. In region I, the ocean is large. As the two major players accumulate votes the ocean becomes smaller. However, in Region II the ocean holds the balance of power between the two large players. In regions III and IV, one or the other of the two large players becomes dominant and has total control. The powers of the players can be converted into their powers per vote held or their fa Appendix B. SHAPLEY VALUE IN OCEANIC GAMES 131 ownership structure power index power ratio (\;w1,w2,w3;oc) <h. (j)2 $oc Ri R2 Roc (.5;5;95) .053 - - .947 1.053 - - .997 (.5;10,30;60) .083 .417 - .500 .833 1.389 - .833 (.5;40,40;20) .250 .250 - .500 .625 .625 - 2.500 (.5;5,40,40;15) .160 .272 .272 .296 3.210 .679 .679 1.975 Table B.39: Examples of Shapley Values for Given Ownership Structures power ratios. Power ratios are then comparable among players: -n-i — — , xt2 — — , -fioc — — Wi W2 OL THEOREM (Milnor and Shapley) If a major player's weight in the general oceanic game tends to zero, his power ratio approaches the power ratio * of the oceanic player. The following examples of some ownership structures and their corresponding power structures demonstrate the non—trivial, non—proportional nature of this correspondence: B.2 Assumptions Underlying Shapley Value Most of the assumptions allowing Shapley value to be used as a proxy for shareholder voting power are the usual ones encountered in the cooperative game theory. The share-holders maximise von Neumann-Morgenstern utility functions. Side payments among shareholders are permitted. This allows some shareholders to compensate others for vot-ing against their preferences. While this assumption seems to imply that votes can be "purchased", in reality the side payments may manifest in some quid pro quo type ar-rangements struck among shareholders. In any case, this requires that utility be linearly transferable among players, so that the utility functions for participating in the game are effectively assumed to be risk-neutral. Appendix B. SHAPLEY VALUE IN OCEANIC GAMES 132 There axe no direct costs involved in casting a vote and in forming coalitions. Also, all shareholders axe assumed to cast theix votes at the meeting. This is not very realistic since in practice, there axe time and material costs to casting votes and small shareholders rarely cast their votes. In the proposed model, the undesirability of this assumption is compensated by allowing the oceanic shareholders to sell out of the firm if they disagree with the vote outcome. We have already noted the superadditivity, efficiency, symmetry and aggregation conditions on the characteristic function underlying the Shapley values. Superadditivity simply implies that individuals will not join a coalition if it makes them any worse off than if they act alone. The efficiency requirement forces individual players' valuations of the game to be consistent with each other. Superadditivity and efficiency are both reasonable realistic assumptions for assessment of a game by players before it is actually played. The same cannot however be said of the other two assumptions, namely symmetry and aggregation. The former makes Shapley value of a player into solely a function of the characteristic function and not of his identity. In a game with two players for instance, who can cooperatively produce one unit of profit and none on their own, the symmetric Shapley value splits the unit profit equally between the two players. The players thus supposedly represent identical constituencies or have equal bargaining power or put in equal effort in generating the unit profit. Clearly in the context of our shareholder voting game with one large and numerous oceanic players, symmetry is a fairly strong assumption. By way of a remedy, Kalai and Samet (1987) propose the use of weighted Shapley values in games with asymmetric situations. While they do not suggest any particular weighting schemes, they offer factors such as the players' relative bargaining abilities, patience rates, past experience, relative sizes, etc as some of the determinants of these weights. Nevertheless, while the proposed model employs ordinary Shapley values to index shareholder voting power, the distortion arising Appendix B. SHAPLEY VALUE IN OCEANIC GAMES 133 from the lack of symmetry is felt to be at least partially mitigated by weighting the game outcomes by shareholder voting powers. The aggregation condition requires that there be no interaction effects between any two games in the sense that a player is indifferent between playing the two games sepa-rately or playing them as one aggregated game. In practice, this condition may be easily violated since the structure and equilibrium outcome of the aggregated game may be very different from those of the individual games. If the universe of games to be played is restricted to those involving dividend policies only, the aggregation condition may not be very restricitve since the cash flows accrueing to different players are Hnear in the dividend policy outcomes of different games. B.3 Shapley value and the dividend voting game On first glance, a shareholder voting game to settle dividend policy would differ from a general oceanic game in one aspect, hi a general oceanic game, all the value from participating in the game is assumed to be derived from winning whereas losing entails zero value. Therefore, all votes are cast with the sole objective of being on the side of the winning majority. In particular, the major player with whom an oceanic player votes is determined ex ante by which of the major players is expected to experience the greatest marginal increase in the probability of win by getting this additional vote. By contrast, in a dividend voting game the behaviour of an oceanic shareholder can not be assumed to be guided purely by whether he will end up on the winning side, even if the winning side is the one with the "wrong" dividend preferences. Given this player's tax rates, the large shareholder he elects to vote with will be the one whose preference between dividends and capital gains is the same as his own rather than with the large shareholder most likely to win. Appendix B. SHAPLEY VALUE IN OCEANIC GAMES 134 The above difference between the general oceanic game and the dividend voting game notwithstanding, the framework is employed here based on the following argument: It is the possibility of side payments from one member to another within a voting block which keeps the coalitions in the general oceanic game from falling apart. Thus, by assumption members within a voting block either have identical motives for control of the firm to start with, or their motives are "made" identical through side payments. The assumption allows the oceanic voting game to proceed without reference to the original issues which may have brought on the need for a vote to be held in the first place. In the dividend voting game, these two steps in the resolution process merge into one. Recent work by Hart and Moore (1987), in which they extend the notion of Shapley value to non-cooperative games scenario, allows this to be accomplished even though there may no longer be any provision for side payments among members of a voting block. Bibliography [1] Ambrish, R., K . John and J. Williams, 1987, Efficient Signalling with Dividends and Investments, Journal of Finance, vol 42, no 2. [2] Asquith, P. and D. Mullins, 1983, The impact of initiating dividend payments on shareholder wealth, Journal of Business 56, 77-96. [3] Barclay, M . J. 1987, Dividends, Taxes, and Common Stock Prices: The Ex-Dividend Day Behaviour of Common Stock Prices Before the Income Tax, Journal of Financial Economics 19. [4] Baron, D., 1979, Investment Policy, Optimality, and the Mean-Variance Model, Journal of Finance, vol 34, no 1. [5] Baron, D., 1983, Tender Offers and Management Resistence, Journal of Finance vol 38, no 2. [6] Barone-Adesi, G. and R. E. Whaley, 1986, The Valuation of American Call Option and the Expected Ex-Dividend Stock Price Decline, Journal of Financial Economics, 17, September. [7] Bebchuk, L .A . , 1984, Litigation and Settlement Under Imperfect Information, Rand Journal of Economics, vol 15. [8] Berle, A . A . , 1962, Modern Functions of the Corporate System, Columbia Law Re-view, 433. [9] Berle, A . A . and G. C. Means, 1967, The Modern Corporation and Private Property, MacMillan, New York, Rev. Ed. 135 Bibliography 136 [10] Bhattacharya, S., 1979, Imperfect Information Dividend Policy, and the Bird-In-Hand Fallacy, Bell Journal of Economics, Spring. [11] Black, F. and M . S. Scholes, 1974, The Effects of Dividend Yields and Dividend Policy on Common Stock Prices and Returns, Journal of Financial Economics 1, 1-22. [12] Booth, L .D. , and D. J. Johnston, 1984, The Ex-Dividend Day Behavior of Canadian Stock Prices: Tax Changes and Clientele Effects, Journal of Finance 39, June, 457-76. [13] Brennan, M . J . , 1970, Taxes, Market Valuation, and Corporate Financial Policy, National Tax Journal 23, December, 417-27. [14] Brickley, J .A., 1986, Interpreting Common Stock Returns Around Proxy Statement Disclosuresand Annual Shareholder Meetings, Journal of Financial and Quantitative Analysis 21, no 3. [15] Brickley, J .A., 1983, Shareholder Wealth, Information Signalling and the Specially Designated Dividend: An Empirical Study, Journal of Financial Economics 12. [16] Brickley, J.A., R. C. Lease and C. W. Smith, 1988, Ownership Structure and Voting on Anti-Takeover Amendments, Journal of Financial Economics 20. [17] Charest, G., 1978, Dividend information, stock returns, and market efficiency - II, Journal of Financial Economics 6, 256-296. [18] Constantinides, G . M . and J. Ingersoll, 1984, Optimal bond trading with personal taxes, Journal of Financial Economics xx, 299-335. [19] Crete, R., 1986, The Proxy System in Canadian Companies: ACritical Analysis, Editions Wilson, Lafleur Martel Ltee, Montreal. Bibliography 137 [20] Dann, L., 1981, Common Stock Repurchases: An Analysis of Returns to Bond-holders and Stock-holders, Journal of Financial Economics June, 113-38. [21] DeAngelo, H. and L. DeAngelo, 1985, Managerial Ownership of Voting Rights: A Study of Public Corporations with Dual Classes of Common Stock, Journal of F i -nancial Economics 14. [22] Demsetz, H. and K. Lehn, 1986, The Structure of Corporate Ownership: Causes and Consequences, Journal of Political Economy. [23] Diamond, P., 1967, The Role of Stock Market in a General Equilibrium Model with Technological Uncertainty, American Economic Review 57. [24] Dillon, William R. and Mathew Goldstein,1986, Multivariate analysis, methods and Applications, J. Wiley. [25] Dybvig, Philip H. and Stephen A. Ross, 1986, Tax clienteles and asset prices, Journal of Finance 41, 751-762. [26] Eades, K . M . , 1982, Empirical Evidence on Dividends as a Signal of Firm Value, Journal of Financial and Quantitative Analysis 17. [27] Eades, K . M . , P.J. Hess and E. Han kim, 1984, On interpretting security returns during the ex-dividend period, Journal of Financial Economics 13, 1-153. [28] Easterbrook, F .H. , 1984, Two Agency-Cost Explanations of Dividends, American Economic Review 74. [29] Easterbrook, F .H. , 1983, and D. R. Fischel, Voting in Corporate Law, Journal of Law and Economics 26. [30] Eckbo, B. Espen, 1988, The Market for Corporate Control: Policy Issues and Capital Market Evidence, in Mergers, Corporate Concentration and Power in Canada, R. Bibliography 138 S. Khemani, D. Shapiro and W. T. Stanbury (eds.), The Institute for Research and Public Policy. [31] Eckbo, B. Espen and Ronald W. Masulis, 1989, Rights vs. firm commitment offere-ings of common stock: An empirical Analysis, Working paper, University of British Columbia. [32] Elton, J. and M . Gruber, 1970, Marginal Stockholder Tax Rates and the Clientele Effect, Review of Economics and Statistics 52. [33] Fama, E.F. and M . C. Jensen, 1983, Separation of Ownership and Control, Journal of Law and Economics 26. [34] Feenberg, D., 1981, Does the investment interest limitation explain the existence of dividends?, Journal of Financial Economics 9, 265-269. [35] Fama, Fisher, Jensen and Roll (1969), The Adjustment of Stock Prices to New Information, International Economic Review, February 1969, 1-21. [36] Gordon, M . J . , 1959, Dividends, Earnings and Stock Prices, Review of Economics and Statistics 41. [37] Grinblatt, M.S., R. W. Masulis and S. Titman, 1984 The Valuation Effects of Stock Splits and Stock Dividends, Journal of Financial Economics 13. [38] Grossman, S.J. and 0. D. Hart, 1980, Takeover Bids, the Free-Rider Problem and the Theory of the Corporation, Bell Journal of Economics 11. [39] Grossman, S.J. and O. D. Hart, 1988, One Share/One Vote and the Market for Corporate Control, Journal of Financial Economics 20. [40] Grossman, S.J. and J. Stiglitz, 1980, Stockholder Unanimity in Making Production and Financial Decisions, Quarterly Journal of Economics May. Bibliography 139 [41] Hakansson, N . , 1982, To Pay or Not to Pay Dividends, Journal of Finance May. [42] Hanjinicolaou, G. and A. Kalay, 1984, Wealth redistributions or changes in firm value: An analysis of returns to the bondholders and to the stockholders around dividend announcements, Journal of Financial Economics 13, 35-63. [43] Hart, 0 . and J. Moore, 1988, Property rights and the nature of the firm, Working Paper No. 495, Dept. of Economics, MIT. [44] Hart, 0., 1979, On Shareholder Unanimity in Large Stock Market Economies, Econo-metrica 47. [45] Hess, P.J., 1982, The ex-dividend day behaviour of stock returns: Further evidence on tax effects, Journal of Finance 37, 445-456. [46] Hess, P.J., 1983, Tests for tax effects in the pricing of financial assets, Journal of Business 56, 537-554. [47] Harris and A. Raviv, 1988, Corporate Governance: Voting Rights and Majority Rules, Journal of Financial Economics 20. [48] Holderness, C.J . and D. P. Sheehan, 1988, An Exploration of the Role of Large Block Shareholder in Publicly Held Corporations With Majority Shareholders, Journal of Financial Economics 20. [49] Jarrell, G.A. and A. B. Poulsen, 1988, Dual-Class Recapitalizations as Antitakeover Mechanisms, Journal of Financial Economics 20. [50] Jensen, M . C . and W. H. Meckling, 1976, Theory of the Firm: Managerial Behaviour, Agency Costs, and Ownership Structure, Jornal of Financial Economics 3. [51] Jensen, M . C . and J. B. Warner, 1988, The Distribution of Power Among Corporate Managers, Shareholders, and Directors, Journal of Financial Economics 20. Bibliography 140 [52] Johnston, D.L., 1977, Canadian Securities Regulation, Butterworths, Toronto. [53] Kalay, A., 1980, Signalling, Information Content, and the Reluctance to Cut Divi-dends, Journal of Financial and Quantitative Analysis 15. [54] Kalay, A., 1982a, Stockholder-Bondholder Conflict and Dividend Constraints, Jour-nal of Financial Economics 10. [55] Kalay, A., 1982b, The Ex-Dividend Day Behaviour of Stock Prices: A Reexamina-tion of the Clientele Effect, Journal of Finance 37. [56] Kalay, A. and U. Loewenstein, 1986, The Information Content and The Timing of Dividend Announcements, Journal of Financial Economics 16. [57] Kalai, E. and D. Samet, 1987, On Weighted Shapley Values, International Journal of Game Theory 16, no 4. [58] Kalay, A. and U. Lowenstein, 1985, Predictable Events and Excess Returns, The Case of Dividend Announcements, Journal of Financial Economics 14. [59] Khemani, R.S., 1988, The Dimensions of Corporate Concentration in Canada, in Mergers, Corporate Concentration and Power in Canada, R. S. Khemani, D. Shapiro and W. T. Stanbury (eds.), The Institute for Research and Public Policy. [60] Khoury, N . , and K. Smith, 1977, Dividend Policy and the Capital Gains Tax in Canada, Journal of Business Administration 1977. [61] Lakonishok, J. and B. Lev, 1987, Stock Splits and Stock Dividends: Why, Who and When?, Journal Of Finance 42. [62] Lakonishok, J. and T. Vermaelen, 1986, Tax induced trading around ex-dividend days, Journal of Financial Economics 16, 287-319. Bibliography 141 [63] Lakonishok, J. and T. Vermaelen, 1983, Tax Reform and Ex-Dividend Day Be-haviour, Journal of Finance 38. [64] Lease, R., J. McConnell and W. Mikkelson, 1984, The Market Value of Differential Voting Rights in Closely Held Corporations, Journal of Business 57. [65] Lintner, J., 1956, Distribution of Incomes of Corporations Among Dividends, Re-tained Earnings and Taxes, American Economic Review 46. [66] Litzenberger, R. and K . Ramaswamy, 1979, The Effects of Personal Taxes and Divi-dends on Capital Asset Prices: Theory and Empirical Evidence, Journal of Financial Economics June. [67] Litzenberger, R. and K. Ramaswamy, 1980, Dividends and shrt-selling restrictions, tax-induced investor clienteles and market equilibrium, Journal of Finance 35, 469-482. [68] Litzenberger, R. and K . Ramaswamy, 1982, The effect of dividends on common stock prices: Tax effects or information effects?, Journal of Finance 37, 429-443. [69] Long, J., 1977, Efficient Portfolio Choice with Differential Taxation of Dividends and Capital Gains, Journal of Financial Economics August. [70] Masulis, R., 1980, Stock Repurchase by Tender Offer: An Analysis of the Causes of Common Stock Price Change, Journal of Finance May. [71] McDonald, R. and N. Luderstrom, 1986, Dividend and Share Changes: Is There a Financing Hierarchy?, Working Paper No. 10, Kellog School, Northwestern Univer-sity. [72] Miller, M . and F. Modigliani, 1961, Dividend Policy, Growth and the Valuation of Shares, Journal of Business 34. Bibliography 142 [73] Miller, M . and K. Rock, 1985, Dividend Policy Under Asymmetric Information, Journal of Finance 40. [74] Miller, M . and M . Scholes, 1978, Dividends and Taxes, Journal of Financial Eco-nomics December. [75] Miller, M . and M . Scholes, 1982, Dividends and Taxes: Some Empirical Evidence, Journal of Political Economy March. [76] Milnor, J. and L. Shapley, 1978, Values of Large Games II: Oceanic Games, Math-ematics of Operations Research vol 3, no 4. [77] Morck, R., A . Schleifer and R. Vishny, 1988, Management Ownership and Market Valuation: An Empirical Analysis, Journal of Financial Economics 20. [78] Morgan, I.G., 1980, Dividends and Stock Price Behaviour in Canada, Journal of Business Administration 12. [79] Partch, M . , 1987, The Creation of a Class of Limited Voting Common Stock and Shareholder Wealth, Journal of Financial economics 18. r [80] Poterba, James M . and Lawrence H. Summers, 1984, New evidence that taxes affect the valuation of dividends, Journal of Finance 39, 1397-1415. [81] Pound, J., 1988, Proxy contests and the efficiency of shareholder oversight, Journal of Financial Economics 20, 237-265. [82] Richardson, G., S. Sefcik and R. Thompson, 1986, A test of Dividend Irrelevance Using Volume Reactions to a Change in Dividend Policy, Journal of Financial Eco-nomics 17. [83] Ross, S., 1977, The Determination of Financial Structure: The Incentive-Signalling Approach, Bell Journal of Economics Spring. Bibliography 143 [84] Rozeff, M . , 1982, Growth, Beta and Agency Costs as Deteminants of Dividend Payout Ratios, Journal of Financial Research 5. [85] Rydqvist, K.,1987, The Pricing of Shares With Different Voting Power and the Theory of Oceanic Games, Doctoral Dissertation, Stockholm School of Economics. [86] Schleifer, A. and R. Vishny, 1986, Large Shareholders and Corporate Control, Jour-nal of Political economy 94. [87] Shaefer, S., 1982, Tax-induced clientele effects in the market for British government securities: Placing bounds on security values in incomplete markets, Journal of Financial Economics 10, 121-159. [88] Smith, C. Jr. and J.B. Warner, 1979, On financial contracting: An analysis of bond covenants, Journal of Financial Economics 6, 117-161. [89] Stiglitz, J., 1972, Some Aspects of the Pure Theory of Corporate Finance: Bankrupt-cies and Takeovers, Bell Journal of Economics 3. [90] Vermaelen, J., 1981, Common Stock Repurchase and Market Signalling, Journal of Financial Economics June. 139-83. [91] DATA SOURCES: • Insider Reports, Ontario Securities Commission, 1978-82. • Inter-Corp orate Ownership Reports, Statistics Canada, 1976-88. • Laval University Corporate Data Tapes. 1976-86. • Moody's International Manuals. 1981^-82. • Report on Business Database, 1988.