Print Reference This Reddit This

Literature Review On The Theories Of Dividend

CHAPTER 2

2.1 Introduction

Dividend policy has been the subject of investigation and debate for almost 50 years, most of it conducted in the United Kingdom. Before regression analysis was applied by John Lintner in 1956 to the behaviour of a small group of industrial companies, dividends were good and should be maximised by firms wherever possible. Lintner, who showed that firms adopted and tended to adhere to optimal long-term dividend pay-out ratios which were relatively stable, suggested that managers would only raise a firm’s dividend if they were confident that the firm’s future earnings could be maintained at a consistently higher level in the future. An implication of this was that the announcement of a dividend increase might convey useful information about future earnings. Lintner’s work (and the work of Darling (1957) who confirmed the relationship between dividends and past and current earnings) opened a Pandora’s Box of dividend-related phenomena, the validity of which, other researchers have spent years and decades debating closely1. In a series of researches at the beginning of the 1960s, Miller and Modigliani (in particular, Miller and Modigliani, 1961) provided a mathematically consistent theory of capital structure in which dividends were shown to be irrelevant to a firm’s value. But this did not appear to coincide with the observed behaviour of dividend policy-setters in a sufficiently watertight fashion. A competing theory, which stated that dividends directly contributed to the value of a firm, was produced by Gordon (1962).Gordon’s model2 for the valuation of a firm’s share price. Hence, the current dilemma concerning the role of dividends in the stock market was laid bare almost forty years ago. It is best summed up in the words of Black (1976):“The harder one looks at the dividend picture, the more it seems like a puzzle, with pieces that just don’t fit together“. But the debate was, however, broadened by Miller and Modigliani (1961), who added several adjuncts to their assertion of dividend irrelevance to the firm’s value. One of these was the existence of tax clienteles. Investors would choose the kind of firm they wanted to invest in with respect to the firm’s dividend policy and thereby sort themselves into clientele groups. Investors who wished to accumulate long-term wealth would choose firms with low or zero dividend pay-outs, while those who wished to have a steady dividend income to meet short-term consumption needs would invest in firms with a tradition of high dividend pay-out ratios.3The existence and effects of tax clienteles have been analysed by a large number of scholars. Brennan (1971), for instance, argued that the existence of a clientele effect would logically have no impact on the value of the firm, while Long (1978) and Litenberger and Ramaswamy (1982) presented evidence that it did. No clear, irrefutable resolution to this debate has yet emerged. A further kind of clientele has also been discussed by Black and Scholes (1974) and Pettit (1977. The other important adjunct Miller and Modigliani posited was that a firm’s choice of dividend might be seen as a signal to investors (actual and potential), which contained hitherto unavailable information concerning the firm’s future earnings prospects. This conclusion was to be the starting point of a forty-year record of research into the existence and nature of signals putatively broadcasted in dividend announcements. But dividend research did not develop in isolation from other major developments. The 1950s and 1960s were fertile times in the development of financial economics. It was in this period that Sharpe (1964) and Lintner (1965) developed the Capital Asset Pricing Model (CAPM). This development in particular, provided stimulus for investigation of dividend policy behaviour associated with volatility.

1 Lintner’s findings have been reconfirmed by a large number of studies over the decades. Early confirmations with respect to United Kingdom data were made by Brittain (1964), Brittain (1966), and Fama and Babiak (1968). McDonald, Jacquillat and Nussenbaum (1975) observed dividend policies in France; Chateau (1979) published results with respect to Canadian companies; Shevlin (1982) observed the stability of dividend policies in India. More recently Leithner and Zimmermann (1993) tested the dividend stability of West German (prior to Unification), British, French and Swiss companies; and

Dividend stability in the United Kingdom was further assessed by Lasfer (1996). Dividend policies in Japan were found to be stable by Kato and Loewenstein (1995), and also by Dewenter and Warther (1998), who compared the Japanese market with the market in India. Adaoglu himself however, observing firms on the Indian Stock Exchange, confirmed Glen, Karmokolias, Miller and Shah (1995), who found relatively unstable dividend policies in emerging markets.

2 This has been shown to be based on at least one flawed assumption (Brennan, 1971), but has not been disproved to the satisfaction of all scholars.

In fact, it was not until the tail end of the 1970s that a proper basis for a theory of dividend policy was formulated. During the last fifty years the several theoretical and empirical studies are done leading to the mainly three outcomes: the increase (decrease) in dividend pay-out affect the market value of the firm or the dividend policy of the firm does not affect the firm value. Furthermore there are numerous theories on why and when the firms pay dividends. Miller and Modigliani (1961) suggest that in perfect markets, dividend do not affect firms’ value. Shareholders are not concerned to receiving their cash flows as dividend or in shape of capital gain, as far as firm’s doesn’t change the investment policies. In this type of situation firm’s dividend pay-out share effect their residual free cash flows and the result is when the free cash flow is positive firms decide to pay dividend and if negative firm’s decide to issue shares. It is concluded that change in dividend may be conveying the information to the market about firm’s future earnings. Gordon and Walter (1963) present the bird in the hand theory which says that investors always prefer cash in hand rather then a future promise of capital gain due to minimising risk. The agency theory of Jensen and Meckling (1976) is based on the conflict between managers and shareholder and the percentage of equity controlled by insider ownership should influence the dividend policy. Easterbrook (1984) gives further explanation regarding agency cost problem and says that there are two forms of agency costs; one is the cost monitoring and other is cost of risk aversion on the part of directors or managers. The explanation regarding the signalling theory given by Bhattacharya (1980) and John Williams (1985) dividends allay information asymmetric between managers and shareholders by delivering inside information of firm future prospects. Miller and Scholes (1978) find that the effect of tax preferences on clientele and conclude different tax rates on dividends and capital gain lead to different clientele

3 The pay-out ratio measures the dividend paid out as a percentage of net profit after tax available for potential distribution to shareholders.

Life Cycle Theory explanation given by the Lease (2000) and Fama and French (2001) is that the firms should follow a life cycle and reflect management’s assessment of the importance of market imperfection and factors including taxes to equity holders, agency cost asymmetric information, floating cost and transaction costs Catering theory given by Baker and Wurgler (2004) suggest that the managers in order to give incentives to the investor according to their needs and wants and in this way cater the investors by paying smooth dividends when the investors put stock price premium on payers and by not paying when investors prefer non payers. As regards the empirical literature the roots of the literature on dividend policy is related to Lintner (1956) seminal work after this work the model is extended by the Fama and Babiak (1968). D’Souza (1999) finds negatively relationship between agency cost and market risk with dividends pay-out. However, the result does not support the negative relationship between dividend pay-out policies and investment opportunities. The empirical analysis by Adaoglu (2000) shows that the firms listed on Indian Stock Exchange follow unstable cash dividend policy and the main factor for determining the amount of dividend is earning of the firms. DeAngelo (2004) document highly significant association between the decision to pay dividends and the ratio of earned equity to total equity controlling for size of the firm, profitability, growth, leverage, cash balance and history of dividends. In addition, the dividend payments prevent significant agency problems since the retention of the earnings give the managers’ command over an additional access to better investment opportunities and without any monitoring. Eriotis (2005) reports that the western firms distribute dividend each year according to their target pay-out ratio, which is determined by distributed earnings and size of these firms. Stulz (2005) observe significant association between decision to pay dividends and contributed capital mix. In investigating the determinants of dividend policy of Indian stock Exchange Naceur (2006) find that the high profitable firms with more stable earnings can manage the larger cash flows and because of this they pay larger dividends. Moreover, the firms with fast growth distribute the larger dividends so as attract to investors. The ownership concentration does not have any impact on dividend payments. The liquidity of the firms has negatively impacted on dividend payments. In Indian case Reddy (2006) show that the dividends paying firms are more profitable, large in size, and growing. The corporate tax or tax preference theory doesn’t appear to hold true in Indian context. Amidu and Abor (2006) find dividend pay-out policy decision of listed firms in Indian Stock Exchange is influenced by profitability, cash flow position, and growth scenario and investment opportunities of the firms. Megginson and Eije (2006) observe that the dividend paying tendency of fifteen European firms decline dramatically over this period 1989 to 2003. The increase in the retained earnings to total equity doesn’t increase the pay-out ratio, but company age does. They also find that the effect of catering the dividend systematically which is nor conclusive evidence of continent and wide convergence in dividend policy. Baker (2007) reports that Indian dividend paying firms are significantly larger and more profitable, having greater cash flows, ownership structure and some growth opportunities. Daniel (2007) concludes that managers treat expected dividend levels as a vital earning threshold for Indian firms. Jeong (2008) identifies that the Indian firms make dividend payments on the basis of firm’s stock face value which is very close to the average interest rate of deposits. The change in dividends is less likely to reflect change in fundamentals of the firms. They find the determinants of dividend smoothing, firm risk, size and growth factors play very important role in explaining the cross section of smoothing the dividend behaviour. One branch of this literature has focused on an agency-related rationale for paying dividends. It is based on the idea that monitoring of the firm and its management is helpful in reducing agency conflicts and in convincing the market that the managers are not in a position to abuse their position. Some shareholders may be monitoring managers, but the problem of collective action results in too little monitoring taking place. Thus Easterbrook (1984) suggests that one way of solving this problem is by increasing the pay-out ratio. When the firm increases its dividend payment, assuming it wishes to proceed with planned investment, it is forced to go to the capital market to raise additional finance. This induces monitoring by potential investors of the firm and its management, thus reducing agency problems. Rozeff (1982) develops a model that underpins this theory, called the cost minimisation model. The model combines the transaction costs that may be controlled by limiting the pay-out ratio, with the agency costs that may be controlled by raising the pay-out ratio. The central idea on which the model rests is that the optimal pay-out ratio is at the level where the sum of these two types of costs is minimised. Thus Rozeff’s cost minimisation model is a regression of the firm target pay-out ratio on five variables that proxy for agency and transaction costs. Transaction costs in the model are represented by three variables that proxy for the firm’s historic and predicted growth rates and risk. High growth and high risk imply greater dependency on external finance due to investment needs, and in order to honour financial obligations, respectively. More support and further contribution to the agency theory of dividend debate, is provided by Moh’d, Perry and Rimbey (1995). These authors introduce a number of modifications to the cost minimisation model including industry dummies, institutional holdings and a lagged dependent variable to the RHS of the equation to address possible dynamics. The results of a Weighted Least Squares regression, employing panel data on 341 UK firms over 18 years from 1972 to 1989 support the view that the dividend process is of a dynamic nature. The estimated coefficient on the institutional ownership variable is positive and significant, which is in line with tax explanations but contradicts the idea about the monitoring function of institutions. Holder, Langrehr and Hexter (1998) extend the cost minimisation model further by considering conflicts between the firm and its non-equity stakeholders and by introducing free cash flow as an additional agency variable. The study utilises panel data on 477 UK firms each with 8 years of observations, from 1983 to 1990. The results show a positive relation between the dependent variable and the free cash flow variable, which is consistent with Jensen (1986). Likewise the estimated coefficient on the stakeholder theory variable is shown to be significant and negative as predicted. The estimated coefficients on all the other explanatory variables are also shown to be statistically significant and to bear the hypothesised signs. Hansen, Kumar and Shome (1994) also take a broader view of what constitutes agency costs, and apply a variant of the cost minimisation model to the regulated electric utility industry. The prediction is that the agency rationale for dividend should be particularly applicable in the case of regulated firms because agency costs in these firms extend to conflicts of interests between shareholders and regulators. Results of cross sectional OLSQ regression for a sample of 81 UK utilities and for the period ending 1985 support the cost minimisation model and the contribution of regulation to agency conflicts in the firm. Another innovative approach to Rozeff’s cost minimisation model is offered in Rao and White (1994) who applies it to 66 private UK firms. Using a limited dependent variable, Maximum Likelihood (ML) technique, the study shows that an agency rationale for dividends applies even to private firms that do not participate in the capital market. It will be noted that perhaps by paying dividends by private firms can still induce monitoring by bankers, accountants and tax authorities. To summarise, the agency theory of dividend in general, and the cost minimisation model in particular, appear to offer a good description of how dividend policies are determined. The variables in the original cost minimisation model remain significant with consistently signed estimated coefficients, across the other six models reviewed above. Specifically, the constant is, without exception, positively related to the dividend policy decision, while the agency costs variable, the fraction of insider ownership, is consistently negatively related to the firms’ dividend policy. The latter is with exception of the study by Schooley and Barney (1994) where the relationship is found to be of a parabolic nature. Similarly, the agency cost variable, ownership dispersion, is consistently positively related to the firm’s dividend policy, while the transaction cost variable, risk, is consistently negatively related to the firm’s dividend policy regardless of the precise proxy used. The other transaction cost proxies, the growth variables, are also mainly significant and negatively related to the firm’s dividend policy, although past growth appears to be a less stable measure than future growth. However, in spite of the apparent goodness of fit of the cost minimisation model to UK data, its applicability to the Indian case may be challenged. Indeed, Samuel (1996) hypothesises that agency problems are less severe in India compared with the UK. In contrast, it may be argued that some aspects of the Indian economy imply a particular suitability of the agency theory, and of the cost minimisation model, to this economy. Notably, as explained in Haque (1999) many developing countries, including India, established state-centred regimes following their independence. These regimes drew their ideology from socialist and Soviet ideas and were accompanied by highly centralised economic policies, which may increase agency costs in at least three ways as follows. First, such policies may increase managers’ agency behaviour. Indeed Joshi and Little (1997) note that when domestic firms enjoy subsidies or a policy of protectionism, the pressure on managers to become more efficient is relaxed. Second, high state intervention means an extension of agency problems to shareholder-administrator conflicts. Indeed, Hansen, Kumar and Shome (1994) show that the degree of industry regulation enters the dividend policy decision. Third, to the extent that management of the economy is based on social philosophies of protecting the weaker sectors such as employees or poorer customers, this may influence managers to consider the interests of non-equity stakeholders. This implies that stakeholder theory should be particularly relevant to the Indian case, and, as shown by Holder, Langrehr and Hexter (1998) this may lead to a downward pressure on dividend levels. However, the relevance of stakeholder theory to the Indian case also implies extension of agency problems to conflicts of interests between equity holders and other stakeholders, increasing the need for shareholders to monitor management behaviour. It is thus the case that on the one hand stands the prediction by Samuel (1996) that agency costs should be lower in the Indian business environment. This implies that the agency rationale for dividends should be less applicable in the case of India. To contrast this, the agency rationale for dividends is predicted to become particularly applicable to India, due to the extension of agency conflicts on at least three accounts as explained above.

2.2 The cost and agency theory of dividend

The literature on dividend policy is mainly concerned with explaining observations on the dividend practices of firms. For example Lintner (1956) observes that dividend policy is important to managers and that the market reacts positively to dividend increase announcements and negatively to decreases. Two important theories to explain these observations include the signalling and agency theories of dividend. The signalling theory of dividend emphasises the role of dividend in conveying information about the prospects of the firm. The agency theory of dividend emphasises the role of dividend in controlling agency behaviour. In both cases dividend reduce information or agency problems but the limitation of using dividend for these purposes is the firm dependency. In the signalling models of Bhattacharya (1979) and Miller and Rock (1985) it is assumed that there is preference for internal finance and that dependency on external finance partly explains firm’s dividend policies. What distinguishes between good and bad quality firms is that in the case of the former the gain from high dividend more than offset the associated cost. In Bhattacharya (1979) frictionless access to extra external financing is assumed to be unavailable, and the cost of paying high dividend is the issue cost of having to resort to outside financing to meet the dividend commitment which implies that firms that face lower issue costs are able to use more signalling, In Miller and Rock (1985) the cost of paying high dividend is the need to cut planned investment. And thus the firm’s dividend policy is partly determined by the need for funds for expansion. Moreover, dependency on external finance explicitly enters the dividend model in a number of studies. For example, in the cost minimisation model of Rozeff (1982), the optimal pay-out ratio is at the level that minimises the sum of agency costs and the cost of raising external finance. Similarly in Higgins (1972) the optimal pay-out ratio is at the level that minimises the sum of the cost of holding idle resources and the cost of issuing external finance. Hence as is implied in the signalling theories of Bhattacharya (1979) and Miller and Rock (1985), the optimal dividend policy in Rozeff (1982) and in Higgins (1972) is explicitly modelled as an inverse function of dependency on external finance. This inverse relationship between dependency on external finance and the firm’s dividend policy is referred to as the transaction cost theory of dividend. In Rozeff (1982), dependency on external finance is measured in terms of growth prospects and firm’s risk. Other possible proxies for dependency on external finance include issue costs, ease of access to capital markets and the availability of surplus cash. However, regardless of how dependency on external finance is measured, the transaction cost theory of dividend is partly based on pecking order theory, information asymmetry and other market imperfections. This is the reason that the transaction cost theory should explain particularly well the dividend policies of firms that rely on capital markets that are characterised by distortions and imperfections. Indeed, these are the characteristics of many capital markets in emerging economies. Capital markets in emerging economies are often differentiated from their Counter parts in developed economies partly in terms of their effectiveness in fulfilling their intended functions. Failure in the case of the former is often attributed to high risk due to political and social instability, high transaction costs, lack of liquidity, and asymmetric information and agency problems. These problems are typically caused by lack of adequate disclosure, inappropriate trading systems, weak and erratic regulations and under-developed financial intermediaries that in efficient markets provide monitoring and market for corporate control. Indeed, Kumar and Tsetsekos (1999) argue that the institutional infrastructure of emerging markets tend to be inferior to that in developed markets in terms of the legal, technological and regulatory framework. A comparative analysis finds the financial and corporate sectors in emerging markets to be substantially less developed compared with those in developed markets. It is suggested that this can be partly explained by their more recent origins. Similarly, Glen, Karmokolias, Miller and Shah (1995), note that the dividend levels in developing countries are substantially lower compared with developed countries. It is suggested that the lower dividend level could be a reflection of less efficient markets, leading to greater reliance on internal finance. The study also finds evidence in a group of developing countries of a positive relationship between pay-out rates and the fraction of total investment that is financed by retained earnings. This is taken as another indication of a relationship in developing countries between dividend policy and the gap between external and internal finance.

Consistent with the above discussion and particularly with Glen, Karmokolias, Miller and Shah (1995) the dividend policies of firms in emerging markets should be particularly sensitive to dependency on external finance. The implementation of cost model of dividend should have a good fit when applied to firms from an emerging market. A company’s performance is to a large extent influenced by the risks that result from its operating and financial activities.

2.3 The Garman and Klass Model

Another method that has been frequently used in the literature is the Garman and Klass model. Garman and Klass have suggested several different estimators in their article with different degrees of efficiency, but the preferred estimator with the highest efficiency score, if the information is available, uses the opening-high low-closing prices. This is said to be 8.4 times more efficient than the classical estimator. It not only incorporates the close to close information but also combines the Parkinson measure. One of the shortcomings of the Garman and Klass model is that it was developed assuming that the underlying asset follows a continuous Brownian motion process. However, stock prices are observable only at discrete time moments, creating a possible source of bias. In an empirical study, Beckers [1983] reinforces this idea that non-continuous prices will bias downward the extreme value estimators. Marsh and Rosenfeld [1986], Edwards [1988], Wiggins [1991] and [1992] also emphasise that non trading activity will affect the efficiency of these estimators. In fact, only by pure chance, the extreme observable high and low prices are also the highest and lowest continuous prices. It is common for the low price to be higher than the "true" lowest price and the high price to be lower than the "true" highest price. This criticism assumes that prices are continuously formed but only some are registered on the stock exchange. Closely related to this topic is the frequency of stock price observation. If there are few transactions during the trading day, then the high and low prices are likely to be less close to the "true" high and low prices than if there are many transactions. This issue was also raised by Garman and Klass, who suggested an adjustment for their estimator, dividing the figures found by some constants based on trading frequency. A lower number of bargains on a stock will result in a higher adjustment. Performance of volatility is attributable to the inherent uncertainty of fluctuations in revenue and operating costs. It also results from the financial costs of the interest on debt financing, Many studies show that performance volatility conveys information about a company’s level of risk to the market (Howatt, 2009) and that higher degrees of volatility have a negative effect on firm value (Allayannis and Weston, 2003; Barnes,2001). Other studies are concerned about the impact of performance volatility on forecasts of future performance (Minton, 1999; Dichev and Tang, 2009; Petrovic, 2009; Brennan and Hughes, 1991; Schipper, 1991). Financial analyst’s andinstitutional investors are generally reluctant to make predictions about the performance of enterprises with higher levels of volatility because doing so may increase their forecast error and result in negative surprises (Badrinath, 1989). Enterprises that exhibit extreme performance volatility may reverse faster (Freeman, Ohlson and Penman, 1982), while high volatility may be due to the inclusion of temporary items, the sustainability of which is unlikely. Performance volatility may also have an impact on a company’s future financing costs (Trueman and Titman, 1988), as it signals a higher likelihood of failure. Another line of research has examined the impact of cash flow volatility on firm performance. For example, Minton and Schrand (1999) reported that cash flow volatility is positively correlated with average levels of capital expenditure, research and advertising costs, and significantly and negatively correlated with the cost of external financing. Allayannis and Weston (2003) reported that cash flow volatility has a significantly negative correlation with firm value. Moreover, the negative impact on firm value from fluctuations in accounting profits is of greater statistical and economic significance. These findings are entrenched in the financial and accounting literature (Petrovic 2009). As performance volatility conveys information to the market about firm value, future performance and future financing costs, it will be interesting to determine whether management is aware of the inherent informational value of earnings volatility and quality and that it subsequently takes action to control risks. In the next chapter methodological considerations and the data collected for the calculations will be discussed.

Need help with your literature review?

Our qualified researchers are here to help. Click on the button below to find out more:

Literature Review Service

Related Content

In addition to the example literature review above we also have a range of free study materials to help you with your own dissertation: