NSE Research Initiative, Project Report no. 229 / 2009 Determinants and the Stability of Dividends in India: Application of Dynamic Partial Adjustment Equation using Extended Instrumental Variable Approach Dr. Manoj Subhash Kamat Dr. Manasvi Manoj Kamat Summary This paper improves on earlier research on stability and determinants of dividend policies by using a more advanced estimation methodology, a larger and more representative sample of panel data (PD), and different proxies for a longer time window 1971-2007.

It is aimed to find whether the Indian private corporate sector follow stable cash dividend policies, whether dividends smoothen earnings in India, to estimate the implicit target payout ratio and speed of adjustment of dividends towards a long run target payout ratio. We further test applicability of dividend stability hypothesis and add to the relatively limited literature on aspects of dividend decision by examining the dynamics of relationship between dividend payouts and a host of other explanatory variables.

We estimate the basic static PD model, GMM-in-Levels {GMM (in-Lev)} model, and its other variations GMM-in-first-differences {GMM (in-Diff)] and GMM-in-Systems {GMM (in-Sys)}so to include other lag structures. This procedure shows us how much the size of the dividend determinants, the speed of adjustment coefficient and the one of the implicit target payout ratio varies across the different estimation techniques. In addition, it will also be useful to compare our results with those of Pooled OLS-estimation with alternate data definitions for checking the robustness of the results.

Keywords: Dividends, Determinants, Stability, Panel Data, Partial Adjustment Model, GMM, GMM (in-Diff), GMM (in-Sys), India. 1 Determinants and the Stability of Dividends in India: Application of Dynamic Partial Adjustment Equation using Extended Instrumental Variable Approach Introduction There is no consensus in the financial markets or in financial literature about the need, importance and factors affecting dividend policy behavior.

On one hand there is a view that dividends significantly affect the value of firm and shareholders’ wealth as per Jensen (1986); while there prevails a skeptical view about the ‘value added’ by dividends on the other hand according to Modigliani and Miller (1958) and Miller and Modigliani (1961). Though Damodaran (2000) points, dividend decisions like investments and financing decisions are crucial and involve tradeoffs, there seems to be little consensus on what should lead us in terms of a “right” dividend policy.

The theories in financial literature dealing with determinants and stability of dividend can be grouped into two categories. Those based on the implicit assumption of asymmetric information, and that based on the explicit assumption. The seminal work in that of Lintner (1956), Fama and Babiak (1968) and Marsh and Merton (1986) hypothesize asymmetric information, whereas the theories based on explicit assumption of dividends include the agency theory, pecking order theory and the dividend signaling theory.

Among the foremost papers on dividend policy, Lintner (1956) embodies dominant patterns of decision-making with respect to dividends. The decisive contribution to the theoretical modeling of dividends by Miller and Modigliani’s (1961) view dividend payment policies as a passive residual of retentions; prior to their work it was believed that the dividend payment by firms would increase firm value. Further the proponents of signaling theories like Aharony and Swary (1980) and Kwan (1981) present that the firms change their dividend policies to signal relatively better information to the market.

Since Lintner neither considers the factors like size, debt, investment, managerial aspects etc. nor considers regulatory constraints in determining dividends, of late this led other researchers to explore and investigate other plausible variables which might possibly be significant. The issue of dividend stability and determinants has been researched and proved for across countries, except for some very recent studies in emerging markets. 2 Objectives

This piece of research is planned in context of an emerging market, India and aims to set the stage for enquiry into relevance of dividend policy by emphasizing its importance to the firm. As such, this is a first attempt to take a holistic view of dividend using rich set of unexplored panel data pertaining to Indian companies for the period 1971 through 2007. In the backdrop of findings of prior researches we review herein, the objectives are to analyze issues relating determinants and stability of the corporate dividend structures in India.

It would be intriguing to find whether the Indian private corporate sector follow stable cash dividend policies, whether dividends smoothen earnings in India and to estimate the implicit target payout ratio and speed of adjustment of dividends towards a long run target payout ratio. We further test applicability of dividend stability hypothesis and add to the relatively limited literature on aspects of dividend decision by examining the dynamics of relationship between dividend payouts and a host of explanatory variables.

The factors as to how liberalization process affects these determinants and whether these factors have changed over time are also explored. Very particularly, we examine the role of industry type and select macro-economic factors in determining the Indian corporate payout policy behavior by interpreting the existence and importance of firm and time effects on data and if so, look whether the information in these effects is more parsimoniously captured by our variables, that vary only over firms or only over time. . Motivation

The proposed study attempts to unearth various factors that determine the dividend policy decisions in India. Although tax policy, depreciation policy, retention policy, interest rate, size of the firm, age of the firm and investment opportunities etc. are theoretically assumed to be major determinants of the corporate dividends, in the light of lower effective corporation tax rate than nominal rate and higher effective depreciation rate than its nominal or general rate, the meager dividend performance in India cannot be attributed to the taxation and depreciation systems.

It is contemplated to shed light on several unresolved issues on dividend policy from a developing country perspective. Detailed empirical evidence for a developing countries’ viewpoint is important, because the institutional frameworks can differ significantly from those in the developed countries. Given that the Indian capital market is developing and the economy is targeted to be one of the largest in world, our results could fill an important gap in 3 empirical literature. Dividend policies have implications on financing and investment behavior of firms.

Payment of dividends reduces free cash flows and alternatively the scope for investments in newer and efficient projects. Deciding what percentage of earnings to payout as dividends is a basic choice confronting managers because this decision determines not only how much funds flow to investors, but also what firms are retained for reinvestments. Thus, the decisions taken by managers relating dividend are interwoven with that of investments. Conflicting opinions exists regarding whether dividend is decided first and retained earnings are residual, or retained earnings is a active variable and dividends the result thereof.

This attempt could highlight the importance of dividends by enquiring its specific role and significance amongst other investment and financing decisions. The question we wish to address is whether corporate investments and financing patterns lead to payouts or it is the other way round. According to Stable Dividend hypothesis, a firm’s value is influenced by the regularity of its dividend payout. Firms with stable dividend policies enjoy better valuation in the capital markets than those with variable dividend policy.

It therefore follows that the investors of firms following stable dividend policy will enjoy better opportunity for wealth creation. Stable dividend policy results in more predictable cash flows in the hands of the shareholders; this reduces uncertainty and consequently the required rate of return whereas variable dividend policy makes the cash flow in the hands of shareholder more variable and hence increases their risk and subsequently, the required rate of return. Managers may then have to satisfy the share holder’s preference for increases in rate of return; else the value of the firm will be subsequently affected.

Likely Contribution to Knowledge The proposed study is different from rest in many ways. Unlike earlier studies we take a holistic view of dividend using Panel Data (PD) pertaining to Indian companies for the years 1971 through 2007. Second, earlier studies on dividend policy did not control for unobserved firm-specific effects which might be correlated with other explanatory variables causing Ordinary Least Squares (OLS) and Within-Groups estimators to be biased and inconsistent. We use the Generalized Method of Moments technique developed by Arellano and Bond (1991), Arellano and Bover (1995) and Blundell and Bond (1998). We hypothesize that dividend policy of the firms is chosen, and is not randomly distributed among companies. We also expect the strong influence of industry, financial and macroeconomic factors. We demonstrate specifically the firm, inter and intra-industry effects across varying periods and the significance or otherwise, of time and random effects by pooling time series and cross- sectional data. Few studies in the West demonstrate that dividend payments tend to follow aggregate economic activity in the economy. Some macro-economic indicators like interest rates, inflation, etc. re likely to affect dividends in some particular way. Thus dividends are roughly assumed be influenced by, or may interalia influence macro economic policies like that of general price levels and interest rate cycles based on aggregate demand activity in the economy. The analysis of behavior of corporate cash payouts therefore assumes significance from the point of macro-economic and microeconomic policies. An enquiry into a number of such variables and the analysis of plausible impact of structural reforms could make study of the Indian case more interesting.

Literature on dividend policies reviewed herein for purpose of present work reinforces the fact that number of studies on dividends in emerging market context is scanty. Dividend policy theories are exhaustively propounded; critically evaluated and empirically tested in the West, and mostly in the context of developed markets. Use of reliable databases, wide and deep sample frame and use of contemporary econometric techniques characterize research on the given subject. Given the imited published work in developing countries like India, a need is felt to attempt a comprehensive integration of both, qualitative support to the quantitative findings on dividend policy. Further, the limited numbers of studies in emerging markets most of them we are able to review, suffer from inadequateness due to scanty coverage of data. This is also true for India. No major private players were able to collect and disseminate widebased data, till some limited sources very recently. The rich data compiled by Reserve Bank of India (RBI) on Company Finances is extensive, but scantly used by researchers.

In fact the RBI has been regularly publishing studies on financial performance of Private Corporate Sector for over three decades. The usage of such a consistent, reliable and wider data canvas can improve reliability of tested models. Panel Data Analysis (PDA) on corporate dividend policies has emerged in dividend related literature over past decades in the developed economies, due to presence of strong and reliable long run databases at government and at a private level. Only a couple of studies on dividends in the context of India make use of PD, though for maximum period of ten-fifteen years.

Majority of studies either use time series 5 data confined to a particular industry or check for the cross-sectional trends and determinants over few industries. Major manufacturing industries like Jute, Textile, and Chemicals etc. are mainly considered. It is for this reason the usage of PD covering Private and Public limited companies spread over different industries for the longer time frame could be insightful. The tendency to pay dividends is under going a metamorphosis in developed and developing countries as well.

The earlier studies explore typical dividend determining variables, examine influence of traditional theories and fit basic regressions on time series or cross-sectional data. Recent developments in interdisciplinary research and advances in computational methods have led to use of different variables, test of emerging explanations, use of pooled data analysis, lag dependent variables, and qualitative variables to explain dividend behavior in developed markets. No systematic attempt to comprehensively apply these emerging techniques in iscovering the determinants of corporate dividend policies in Indian context is yet evident. We resort to the use of classical OLS based analysis, static panel analysis (time, firm and random effects) and also dynamic panel data analysis for our interpretations. We subject our PD estimates to a host of alternate model specifications across three different time series, over a longer time frame of 35 years. This study aims to extend understanding of the importance and determinants of dividend policy and may provide guidance on forecasting dividend yields of a company.

Moreover, complements the emerging body of literature on payout policies in emerging economies. One could rely on the methods and models empirically tested and those which have been proved to be most useful in explaining dividend behavior of firms in developed countries by attempting to exploit the theoretical advances and analytical advances in this area. Such an analysis will also indicate as to how the behavior of specific variables in Indian context differs from those in the developed markets 2. Review of Literature

Several studies {Smith (1963, 1971), Dhrymes and Kurz (1967), Plattner (1969), Hakansson (1969), Long (1978), Chateau (1979), Murray (1981), Penman (1983), Poterba (1986), West (1988), Han et. al. , (1989), Frankfurter and Lane (1992), Cochrane (1992), Isa (1993), Elston (1994), Christie and Nanda (1994), Lee (1995), Raaballe and Bechmann (2000), Desai et. al. , (2002), Scott et. al. , (2003), Elston et. al. , (2004), Faulkender et. al. , (2004), Omran and Pointon (2004) for Egypt and Luders et. al. , (2004)} depict the impact of various factors determining dividend policies.

Brittain (1966) elaborately captures the effects 6 of various financial and macro-economic variables on the dividend policies of the firms while Fama (1981 and 1984) study the impact of macro-economic factors on dividend adjusted stock returns while in his later paper examines the relation between dividends and investments. Campbell and Shiller (1988a and 1988b) study the effect of stock prices, discount factors and earnings on dividend policies of the firms and Mohd. Perry (1995) uses firm size and industry representation as control variables. The former, controls for both the transaction cost and agency cost proxies.

Industry representation is used as a control variable for it is an important factor in payout decisions. It is found that the dividend policy is positively related to the firm size, amount of institutional holding and number of shareholders and is negatively related to past and future growth, operating and financial leverage risk. Redding (1995) studies interrelationships between firm size and liquidity on dividend payments from a theoretical and empirical perspective and it is shown that the dividend decision is quite robustly positively correlated with company size and the liquidity of company’s shares.

The effect of the proxies of size and liquidity on the level of dividend payment is also examined wherein the dependent variable is the dividend yield and suggests that size and liquidity has its strongest contribution in explaining the dividend decision. Other informational factor such as monitoring and signaling remains strong determinants of the level of corporate dividend. Ang et. al. , (1995) examines the diversities in dividend policies for Indonesian firms whereas, Kester and Md. Isa (1996) compares the dividend policy behaviour of firms in Malaysia.

Sarig (1984, 2001) also demonstrates firm effects. In the later study, using Vector Auto Regression estimation for the data period 1950-1997 find that the corporate investment decisions determine payout policies and not the other way round. Booth (2002) in his study of the Importance of dividends reveals the firm effect. Carvalhal-da-Silva and Leal (2002) attempts to link corporate governance indicators, market valuation tools and dividend indicators in Brazil whereas in a more recent study Kowalewski et. al. (2007) constructs measures of the quality of the corporate governance for 110 non-financial companies listed on Warsaw Stock Exchange to find evidence that an increase in the transparency indices leads to an increase in the dividend-to-cash-flow ratio. They also find that more profitable companies have higher dividend payouts, while riskier and more indebted firms prefer to pay lower dividends. The studies like that of Mazumdar (1959), Rao and Puranandam (1965), Kumar and Manmohan (1966), Sharan (1980), Rao et. al. (1984), Khurana (1985), Dholakia and Bhatt (1986), Chawala and Srinivasan (1987), Kevin (1992), Panigrahi et. al. , (1991), Mahapatra 7 and Sahu (1993), and Roy and Mahajan (2003) depict the impact of financial variables in evaluation of dividend policies for India. Of late researchers resort to use Static and Dynamic PDA is in determining dividends with the use of limited dependent variable techniques like Tobit, Probit and Logit regressions, the Fixed and Random Effect Models (FEM and REM respectively), and also emerging techniques like that of the Generalized Methods of Moments (GMM).

Some prominent studies that use PDA are those by Frankfurter and Gomg (1993), Lasfer (1996), Benito and Young (2001, 2002), Kang (2001), Pandey (2001), Barclay et. al. , (2003), Baker and Smith (2003), Kumar (2003), Benzinho (2004), Stacescu (2004), John and Kayazenva (2006), Gopalan et. al. , (2006 and 2007) among others. Lee and Xiao (2003) investigate cash dividend paying behavior in China and find no correlation between FCF and the probability of paying dividends, that current profitability is a precondition for cash payments and that cash dividends may be used as a tool for expropriating minor shareholders.

Bebczuk (2003) analyses PD of 55 listed companies in Argentina for the period 1996-2002 using Tobit estimation instead of dynamic GMM based technique for he notes that the dependent variable is truncated at zero with many observations displaying such a value and that, endogeneity doesn’t seem to be particular in the subject understudy. Dummy variables are used for time, for ADR (American Depository Receipts) issues and for foreign owned firms and Industry. Study reports that the bigger and more profitable firms, firms with more good investment opportunities and more fluid access to debt pay more dividends.

Riskier and more dividends indebted firms prefer to pay lower dividends and the same applies to foreign owned firms and do not seem to care about maintaining stable payout ratios over time. The industry dummies tend to turn non-significant. Benito (2003) uses PD methodology to examine the dividend policies of firms in Spain. His results are consistent with a tax discrimination model in which cash flow is the marginal source of funds. High degrees of persistence are also found in binary PD models that control for unobservables and initial conditions.

Whilst companies in Spain use the dividend to adjust the balance sheet, the paper finds that such persistence occurs slowly. De Angello et. al. , (2004) uses PD for the period 1973-2002 to suggest that firms with relatively high amounts of earned equity (retained earnings) are especially likely to pay dividends. Using a broad variety of multivariate Logit specifications, they consistently observe a positive and highly significant relation with FamaMacBeth t-statistics in the double digits between the probability that a firm pays dividends and the relative importance of earned equity in its capital structure controlling for firm size, current and lagged profitability, growth, leverage, cash balances, and dividend history. In the regressions, earned equity has an economically more important impact than does profitability or growth. This evidence is consistent with the hypothesis that firms pay dividends to mitigate agency problems. Employing the PD methodology Omet (2004) examines the dividend policy behavior of companies listed in Amman Securities Market, Jordan. The study uses a balanced PD for 44 firms and employs Pooled OLS, the FEM and the REM.

The dummy variable measures the differential intercept and the differential earnings per share coefficient based on the time period 1985-1999 whereas, Goergen et. al. , (2004a, 2004b) use the GMM technique consistent with Arellano and Bond (1991), Arellano and Bover (1995) and Blundell and Bond (1998) procedures on firm level PD and find German companies are more willing to cut the dividend in the wake of a temporary decrease in profitability.

Chay and Suh (2005) by examining cross-sectional determinants of corporate dividend policy in twenty-four countries around the world including India, suggest that cash flow uncertainty has negative relation with corporate dividend policy around the world. Renneboog and Trojanowski (2007) using PDA estimations on the Lintner (1956) framework examine whether or not dividend policy is influenced by the firm’s corporate control structure whereas, Kowalewski et. al. (2007) using PD for Poland suggest that large and more profitable companies have a higher dividend payout ratio.

Furthermore, riskier and more indebted firms prefer to pay lower dividends. More recently, Hedensted and Raaballe (2008) based on a total sample largely uncontaminated by share repurchases in Denmark find that the characteristics of dividend payers are: Positive earnings, high ROE (net earnings to book equity), low volatility in ROE, high retained earnings, large firm size, and whether the firm paid out dividends in the previous year. MV/BV, leverage and owner structure play no role in whether a firm pays dividends or not. Andres et. l. , (2008) employ partial adjustment model on dynamic panel data find that German firms do not seem to base their dividend decisions on long term target dividend payout ratios based on public earnings. Regarding the speed of adjustment of dividends towards the long term target payout ratio, the authors find that UK and US companies slowly adjust their dividend policy whereas German companies tend to be more willing to cut the dividend in the wake of a consistent decrease in profitability. 3. Data Specifications and Methodology

For the purpose of empirical analysis, the period of study is taken from 1971-2007 and is sourced from the various annual studies based on the annual accounts of selected 9 companies from among the non-government non-financial Public and Private limited companies and non-government financial and investment companies. This is the largest possible span for which firm level data is currently available for Indian firms. The unpublished private corporate firm level data for the empirical study is requested from the RBI and sourced from the database maintained from its Annual Studies on Company Finances.

Banking, insurance and other financial companies as also companies limited by guarantee and associations, organizations functioning not-for-profit or in formative stage and those not operative for more than six months during the year are excluded in the dataset. The average number of public limited companies for which equity dividend data is available in the full period is 1815, and the numbers of equity dividend related firm level observations are 67,174 (see table 1, in Appendix).

For time series and static panel modeling, the entire time frame is divided into 1971-1992, 1993-2007 denoting the pre-reform and the post-reform periods respectively, and for the entire period 1971-2007. The subdivision of panels enables us to illustrate the effect of economic liberalization on the corporate dividend policy in India. Since the use of dynamic models involve variable in their own lagged form, the entire sample had to be revised. In such case the entire sample period ranges from 19752007, with the periods 1975-1992 and 1995-2007, classified as pre-liberalization and postliberalization periods respectively.

Using PDA, the models like Fixed Effects, Random Effects for Static PDA and GMM technique for Dynamic PDA have been used for the estimation of our dividend data. The primary motivation for analyzing Panel Data is to control for unobservable firm heterogeneity. Hsiao (1985) argue that pooling data, using appropriate estimation techniques, and grouping individuals according to certain a priori criteria can help overcome this heterogeneity problem.

However it is rather difficult to establish exogeneity between the regressors and error term especially in company financial data and therefore the direction of causality between variables might be ambiguous because of the potential endogeneity. Consequently, the contemporaneous data for both dependent variable and its determinants may cause spurious results. In financing literature the endogeneity problem is either largely ignored or corrected for only using fixed effects or control variables approach.

We control for this important problem by employing GMM technique to avoid significant bias in estimates. 3. 1 Static Panel Data Technique 10 We prefer PDA, as it is possible to include time effects as well as to control for the heterogeneity of firms by including firm-specific effects, which may be random or fixed. However, the Fixed Effects Model (FEM) is costly in degrees of freedom because it is equivalent to the use of a dummy variable for every firm, Greene (2003). The assumption involved in FEM is that the effects are fixed that means the error term is assumed to be random.

In this type of model the regressors may be correlated with the individual and time effects. For the error, which is generally denoted as µit having the properties E( µit )=0, and that µit is uncorrelated across i and t. This model is also called as Least Square Dummy Variable Model. If the coefficients are assumed to be fixed then the coefficients are estimated by dummy variable models. This estimation is called as Fixed Effect Approach which yields consistent estimates regardless of correlation between firm specific error component and regressors.

When we take the dummy variables for the firms only then that model is called as One Way FEM, while when we take a dummy both, for firm and time, that model is known as Two Way FEM. In the One Way FEM, the µi ’s are assumed to be fixed parameters to be estimated, and the remainder disturbances stochastic with ? it independently and identically distributed IID (0,? v ) . The xit are assumed independent of the vit for all i and t. The FEM is 2 an appropriate specification when we focus on a specific set of N observations and our inference is restricted to the behavior of these sets of firms or observations.

If the µi and ?t (unobservable time effect) are assumed to be fixed parameters to be estimated and the remaining disturbances stochastic with vit :IID (0,? v ) then uit = µi + ? t + vit represents a Two 2 Way FEM or the Error Component Model (ECM). Inference in this case is conditional on the particular N individuals and over the specific time periods observed. If there are two many parameters in the FEM and the loss of degrees of freedom is very high it can be avoided if the µi can be assumed to be random. This is the Random Effects Model (REM).

If an effect is assumed to be the realized value of a random variable, it is called a Random Effect. In this case the µit :IID (0,? 2 µ ),? it :IID (0,? 2 ? ) and the µi are independent of vit . The individual effect is characterized as random and inference pertains to the population from which the sample was randomly drawn. If the ui :IID (0,? 2 µ ), ? i :IID (0,? 2 ? ), and ? i :IID (0,? 2 ? ) are independent of each other then this is the Two way REM. Inference in this case pertains to the large population from 11 which the sample was randomly drawn.

REM assumes the independence between error terms and explanatory variables. In this set up it is assumed that the effects are random variables except for the additive constant, which is a fixed quantity. In FEM the effects of omitted variables are treated as fixed constants over time. But in the case of REM the individual or time effects are treated as random variables. 3. 2 Dynamic PDA using Extended Instrumental Variable (IV) Technique Dividend decisions are dynamic by nature and could be modeled as such. PDA allows us to study the dynamic nature of the dividend decisions at the firm level.

Dynamic panel-data models can be estimated by the Generalized Method of Moments developed by Hansen and Singleton (1982), Holtz-Eakin, Newey and Rosen (1988), Arellano and Bond (1991) and Arellano and Bover (1995) to estimate the structural model of dividend. GMM is used when the regression is dynamic and include lagged dependent variables. However the lagged dependant variables can create a bias of estimates obtained through classical regression analysis because the error term by definition is correlated with the lagged dependent variable.

Due to such a correlation the OLS assumptions will be biased as the assumptions of nonspherical error terms are violated. Similarly, if there is a target dividend ratio, then firms should take the appropriate steps to reach this objective. However, the fixed or random effects models may also give biased and inconsistent estimators due to the correlated error term with lagged variable. To deal with variables that may be correlated with the error term, Instrumental Variables (IV) can be used. Application of GMM to econometric models can be considered as an extension of IV estimation method.

IV estimation is widely used for models with random regressors (e. g. lagged dependent variable) which exhibit the correlation with model errors. Using IV has the additional advantage of solving problems encountered in static models, mainly the simultaneity bias between the dividend measure and the explanatory variables, and the measurement error issue. The prime advantage of GMM is that the model need not to be homoscedastic and serially independent. The covariance matrix of the averages of sample moments is taken into account for minimizing the GMM criterion function.

The advantage of GMM is that it finds the parameters of interest by maximizing an object function which includes the moment restriction that the above mentioned correlation between the error term and the lagged regressor is zero. GMM differs from other estimation principles such as least 12 squares, or maximum likelihood in the objective of the minimization problem as the GMM estimators are defined by choosing the parameters to minimize the criterion function. For notational convenience, let X be a combined data matrix of endogenous (dependent) and predetermined (independent or explanatory) variables in the model. is a Kelement vector of unknown parameters. Suppose there are L moment equations, m(X,? ) = (m1(X,? ), ... , mL(X,? )), where L? K …………………………………………(1) GMM sets the moment or orthagonality restrictions close to zero. The GMM estimator is the value of the parameters that satisfies the sample moment condition. Corresponding to the moment conditions E(m(X,? )) = 0, we write the sample moment equations as follows m(? ) = 1/N ? i=1,2,... ,N m(Xi,? )' = 0 …………………….. ………………... ……………(2) Assuming pth order auto-covariances, the well-known White-Newey-West estimator of covariance matrix of sample moments is Var (m( ? ) = S0 + ? j =1,2,... , p (1- j /( p + 1))( S j + S j ' ) .. …….. ……………………. ……….. (3) Where S0 = m( ? )m( ? ) ' = 1/ N 2 ? i =1,2,... , N m( X i , ? ) ' mi ( X , ? ) , ………………………... ….. (4) S j = m( ? )m- j ( ? ) ' = 1/ N 2 ? i = j +1,2,... , N m( X i , ? ) ' m( X i ? j , ? ) and j = 1,... , p < N . …. (5) Given a positive definite symmetric weighting matrix W, the goal is to minimize the quadratic function: Q(? ) = m(? )'W m(? ) …………………………………………………………………. …. (6) Optimally, W is chosen to be the inverse of the consistent estimator of asymptotic covariance matrix of m(? ). That is, W=W(? )= [Var(m(? ))]-1 ................................................................................................... (7) The GMM estimator ? * of ? is obtained from solving the zero gradient conditions: 13 ?Q(? *)/?? = 0. Let G(? *) = ? m(? *)/?? , which is L by K matrix of derivatives The estimated variance-covariance matrix of ? * is Var(? *) = [G(? *)'[Var(m(? *))]-1G(? *)]-1 .………(8) ....................................................................... (9) The asymptotic efficient estimator ? * is normally distributed with mean ? and covariance matrix Var(? *). ) The intuition behind GMM is to choose an estimator for ? that solves g ? ? ) =0. These GMM estimators allow controlling for unobserved individual effects which is present in the static model, endogeneity and simultaneity of explanatory variables and the use of lagged dependent variables, Hansen (1982). Firm and individual effects are taken care by first differencing the variables while use of time dummies for each year takes care of time-effects. Consider the following model yit = ? yit ? 1 + ? ?xit + ? ? f i + uit ? ………………………………. …………….. ……. ……(10) …………………….... …………….. (11) where uit = ? i + ? it and E (? it / xi 0 ,.... , xiT ,? i ) = 0 fi is an observed individual effect and ? is an unobserved individual effect. In this model, regardless of the existence of unobserved individual effects, unrestricted serial correlation in vit implies that yit ? 1 is an endogenous variable. In estimating our dividend model we want to allow for the possibility of simultaneous determination and reverse causality of the explanatory variables and the dependent variable. We therefore relax the assumption that all explanatory variables are strictly exogenous. In principle, the simultaneity bias in the estimated models can be tackled by the use of instrumental variables to obtain consistent estimates of the coefficients.

Consistent GMM estimation requires that the instruments used be uncorrelated with the unobservable effects to the function since these effects may be included in the error term. Examples of these effects include attributes of the mangers of firms such as ability and motivation, or their attitudes towards taking risk. They might also include time-invariant industry specific effects, which are specific to the industry in which the firm operates. These might involve those structural characteristics such as entry barriers, market conditions and industry wide business risk. While the time dummies take note of the macro 4 economic shocks common to all the firms, these effects are mainly macroeconomic effects such as prices and interest rates (inflation levels and yield curve in our model). Mostly these effects will be captured by the presence of firm specific and time specific dummies. Considering the following model: Yit = ? Yit ? 1 + ? 1Yit ? 2 + ? X it + ? 1 X it ? 1 + ? 2 X it ? 2 + uit ………………………………………... (12) Where uit = µi + ? t + ? it and E (? it ) =0. ….. ….. ……….. ……………………... …….. (13) In this model, regardless of the existence of unobserved effects, unrestricted serial correlation in ? t implies that Yit ? 1 is an endogenous variable. Relaxing the assumption that all explanatory variables are strictly exogenous i. e. explanatory variable is uncorrelated with the error term at all leads and lags, and assuming weak exogeneity i. e. explanatory variable is uncorrelated with future realizations of the error term (i. e. may be affected by past and current dividend payout ratios, but not by future ones) of the explanatory variables, the joint endogeneity of the explanatory variables requires an IV procedure to obtain consistent estimates of the coefficients of interest.

In case the unobserved effects are not present, we can employ GMM in Levels {GMM(in-Lev)} under the assumption that the error term ? it is serially uncorrelated or at least follows a moving average process of finite order and also assume that the future innovations of the dependent variable do not affect current values of explanatory variables, the observations viz. (Yit ? 2 , Yit ? 3 ..... , Yin ) and ( X it ? 2 , X it ? 3 ..... , X in ) can be used as valid instruments in the GMM estimations.

However in the presence of unobserved individual effects since the GMM (in-Lev) estimator produces inconsistent estimates, one can estimate the specific model in first differenced form, refered to as GMM in Differences or the ‘Difference estimator’{GMM (in-Diff)}. In this case: ? Yit = ?? it ? 1 + ?? it ? 2 + ?? X it + ? 1? X it ? 1 + ? 2 ? X it ? 2 + ?? it .. ………………. …... …. ……. …(14) Using first differences eliminates the specific firm effect, thus avoiding any correlation problem between unobservable firm specific characteristics and explanatory variables.

First differencing equation removes the firm-effect and produces an equation that can be estimated 15 using instrumental variables. This has the additional advantage that it solves the problem of possible endogeneity in the regressors. The use of instrumental variable is thus again required because ?? it is correlated with ? Yit ? 1 by construction and joint endogeneity of the explanatory variables. Under the assumption that the error term ?? it is not serially correlated and the explanatory variables are weakly exogenous the following moment condition apply to the lagged dependent variable and the set of explanatory variables E (Yit ? s ?? t ) = 0 E (X it ? s ?? it ) = 0 ?s ? 2; t=3,----- T ……….. …………………. …………... (15) ?s ? 2; t=3,----- T ..... ………….. ……………………....... (16) so that (Yit ? 2 , Yit ? 3 ,..... , Yin ) and ( X it ? 2 , X it ? 3 ,..... , X in ) are valid instruments. Arellano and Bond (1991) have shown that under the assumptions that the error term ? it in equation 10 is not serially correlated and the explanatory variables are weakly exogenous, i. e. GMM (in-Diff) is an efficient GMM estimator for the above model. Although GMM (in-Diff) solves the problem of the potential presence of unobserved individual effects, the estimator has some statistical shortcomings.

Blundell and Bond (1997) show that when the dependent variable and the explanatory variables are persistent over time, lagged levels of these variables are weak instruments for the regression equation in differences. Blundell and Bond (1997) suggest the use of Arellano and Bover’s (1995) ‘System estimator’ {GMM (in-Sys)} to overcome the statistical problems associated with GMM (in-Diff) estimator. Arellano and Bover (1995) show that, when there are instruments available that are uncorrelated with the individual effects ? i , these variables can be used as instruments for the equations in levels.

They develop an efficient GMM estimator for the combined set of moment restrictions relating to the equations in first differences and to the equations in levels. The GMM (in-Sys) estimator makes additional assumption that differences of the right-hand side variables are not correlated with the unobserved individual effects, however there may be correlation between the levels of the right-hand side variables and the unobserved individual effects. E (Yit? i ) = E (Yis? i ) E ( X it? i ) = E ( X is? i ) ? t , s, ……………... ……………………………. ……….. ……….. (17) ?t , s, ………….... ………….. ………………………………........ 18) 16 These assumptions may be justified on the grounds of stationarity. Arellano and Bover (1995) show that combining equations 15-16 and 17-18 gives the following additional moment restrictions E (uit ? Yit ? 1 ) = 0 E (uit ? X it ? 1 ) = 0 …………………....... ………….. ……………………………........... (19) …………………..... ………….. ……………………………........... (20) Thus, valid instruments for the regression in levels are the lagged differences of the corresponding variables. The instruments for the regression in differences are the same as before, that are, the lagged levels of the corresponding variables.

Hence, we use (Yit ? 2 , Yit ? 3 ,.... , Yi1 ) and ( X it ? 2 , X it ? 3 ,.... , X i1 ) as instruments for the equations in first differences, and ? Yit ? 1 with ? X it ? 1 as instruments for the equations in levels. Again, these are appropriate instruments only under the above assumption of no correlation between the right-hand side variables and the unobserved individual effect. To assess the validity of the assumptions on which the three different estimators are based we consider four specification tests.

The test statistic m2 for the null hypothesis of no second order serial correlation is reported along with the result of two Wald tests; Wald Test1 for the joint significance of the time dummies variables and Wald Test2 for the joint significance for all variables respectively. The m2 test of second-order serial correlation of the error term checks whether the error term in the differenced model follows a first-order moving average process where the use of endogenous variables dated t–2 as instruments is valid only if n is serially uncorrelated, implying a first-order moving average error term in the differenced model.

However, following the recommendation by Arellano and Bond (1991), their two-step GMM estimator is applied for inference on model specification. Specifically, with respect to the validity of the instruments on which these estimators are based, we conduct the Sargan Test of over-identified restrictions, which tests validity of instruments for the null hypothesis that the over identifying restrictions are valid. This is based on hetroskedasticity consistent two-step GMM estimator that tests for the validity of extra instruments in the equation.

The statistics is asymptotically distributed as a chi-square with as many degrees of freedom as over identifying restrictions under the hypothesis of the validity of the instruments. The Hausman specification test checks the validity of the additional instruments used in the levels equations of the system estimator. 17 4. Explanatory Variables and Hypothesis The result of intensive modeling and theoretical examination of dividends brings out a broad understanding on the various sets of variables affecting dividend policies.

Several studies use different combinations of variables for explaining the dividend behavior. These factors vary from country to country and affect in different magnitudes due to variations in socio-economic and legal environment of each country. To motivate the expected signs on these determinants of dividends, we draw upon our review of the literature and select a list of plausible variables that are priori expected to influence cash dividend distribution, and subject them to procedures to identify their relative dominance over time, 1971-2007.

The definition of the underlying determinants and their nature of relationship expected with the Dividend Payout Ratio (DPR) are classified into those that vary both across firms and time, and those that vary only over time, and are briefly indicated below. 4. 1 Variables those Vary both Across Firms and Time (Xit) i. Earnings (ERNG) Return on Assets defined as net earnings after taxes by total assets of the firm surrogates ERNG variable. Earnings of the firm undoubtedly expected to have the largest influence on dividend payment decision.

It is hypothesized that the net income or the profit after tax of the firm would be positively related to dividends as it is negatively with the debt levels. Loss making and low profit margin firms are more likely to omit dividends whereas poor quality firms cannot afford to match dividend payments because they face high transaction costs when the cash flows don’t materialize. Large firms are mature, have sufficient internal funds to finance profitable investment opportunities and can obtain funds for investments through the internal sources without issuing additional equity.

Owing to their magnitude of size and profits large firms are in a better position to distribute residual funds as dividends even if tax system discriminates against dividends, Siddharthan et. al. (1991), and Aurebach and Hasset (2002). It is found that earning profits is not the essential criterion which influences payers to pay. Firms reporting losses also demonstrate their liking for paying dividends, however the tendency to pay is more pronounced in profit making firms. 18 ii. Firm Size (SIZE) Theoretically, the relation between size and dividends is not clear.

The variable firm size can serve as an inverse proxy for unobservable credit risk, a proxy for diversification, external cost of financing, information asymmetry and also for agency cost. The relationship with dividends depends on what size proxies for. This variable has been the subject of attention in determining dividends especially by Fama and French (1999) and also by Aivazian et. al. , (2001), accordingly depicts contradictory signs with dividends in numerous studies. Many studies argue that larger firms tend to be more diversified and hence are less likely to go bankrupt and hence smaller dividend distributions.

On the other hand, Warner (1977) and Titman and Wessels (1988) document that bankruptcy costs are relatively higher for smaller firms and hence larger firms tend to be more diversified and fail less often. Accordingly, the trade-off theory predicts an inverse relationship between size and the probability of bankruptcy. If diversification goes along with more stable cash flows, this prediction is also consistent with the FCF theory by Jensen (1986) and Easterbrook (1986). Fama and Jensen (1983) argue that larger firms tend to provide more information to lenders than smaller firms.

Therefore, the monitoring cost should be smaller for larger firms and hence these arguments predict a negative relationship with dividends, which is used as a prominent signaling device. Also, it may be expected that smaller firms grow faster through retentions and so there would be a negative relationship between the retention ratio and firm size, and hence a positive relationship between the DPR and firm size is expected. Reeding (1997) show, that firm size and liquidity explain the decision of whether to pay dividends well, whereas existing informational explanations (such as monitoring and signaling) explain the level of dividends well.

On the other hand, size may be inversely related to the level of information asymmetries between insiders and outside investors, Rajan and Zingales (1995). Equity holders of larger firms put less pressure on the firm’s managers for issuing excess dividend. On the contrary, smaller firms will pay out higher excess dividends to mitigate the agency problem resulting from asymmetric information. Moreover Smith and Watts (1992) point out, the theoretical basis for an impact of size on dividend policy is not strong, and indeed some negative relationships have been observed, Keim (1985) and Allen and Michaely (1995).

The inclusion of size may be best regarded as a simple control variable, without a particular sign expectation. Our measure of size is natural logarithm of net sales following 19 Titman and Wessels (1988) as logarithmic transformation accounts for the conjecture that small firms are particularly affected by size effect. Alternatively, one could use the natural logarithm of total assets. However we think that net sales is a better proxy for size, because many firms attempt to keep their reported size of asset as small as possible, e. g. , by using lease contracts. iii. Investment Ratio (INVR)

In confirmation with the Pecking Order Theory large investment opportunities imply higher growth opportunities for the firm and interalia, low payout. The trade-off model predicts that firms with more investment opportunities have less leverage because they have stronger incentives to avoid underinvestment and asset substitution that can arise from stockholder-bondholder agency conflicts. This prediction is strengthened by Jensen’s (1986) FCF theory, which predicts that firms with more investment opportunities have less need for the disciplining effect of debt payments to control FCF.

A rapidly growing concern will have constant needs of long-term funds to seize favorable opportunities and for that purpose it may need to finance greater part of its funds for expansion, Pogue (1971), Pruitt and Gitman (1971) and Smirlock and Marshal (1983). Such a decision will mean that dividend must be kept at a minimum. Mason and Merton (1985) point out the firms with growth options are those that have relatively more capacity expansion projects, new product lines, acquisitions of other firms and maintenance and replacement of existing assets.

Tax based theory, signaling theory, and agency theory explain the association between growth opportunities and financing decisions. The tax argument relies on the progressivity in taxes which implies that expected tax liabilities are higher when there is greater volatility in taxable income. Thus, firms with high growth options and high cash flow volatility have incentives to reduce debt in their financing mix over the range of progressivity, Smith and Watts (1992). According to agency theory firms with more growth opportunities are less likely to issue debt for two reasons.

First, the underinvestment problem suggests that firms generally issue only risky debt that can be supported by assets-in-place. If not, managers acting on behalf of shareholders may decide not to undertake positive NPV investments to avoid the possibility of the payoffs going to debt holders. Second, given that debt has been issued, the asset substitution problem occurs when managers acting on behalf of shareholders opportunistically substitute higher variance assets for lower variance assets. In this way, wealth is being transferred to the shareholders provided the debt was issued and priced on the basis of low variance assets.

Asset substitution 20 is less likely when there are more assets-in-place since it is relatively easy for outsiders such as auditors to monitors the existence and value of these assets such as land, building, and plant. However, when a firm has more intangible growth opportunities, asset substitution in more likely since outside monitoring of these assets is more difficult. Thus, firms with more growth opportunities are likely to pay lower dividends, other things being equal which is also consistent with the Residual theory. The amount of retained profits of the firm can be expected to be positively related to the growth rate of the firm.

It is argued that a high growth rate of the firm reflects greater investment opportunities, higher profits, and greater need for finance. All such factors would make the firm to earn higher proportion of its net profits and in turn would distribute smaller dividends. A negative relationship between dividends and investment ratio of the firm is expected. The ratio of fixed and inventory investment along with R&D spending to total capitalization is taken as a measure the investment ratio of the firm. iv. Tangibility of Assets (TNGA)

Our choice for the inclusion of the tangibility variable amongst the independent variables emerges from the theoretical support of the agency model, asset substitution, and the trade-off theory model. Consistent with Aivazian et. al. , (2003) we hypothesize that the firms most likely to pay a dividend are also likely to access the public debt markets if they are larger in size and have more tangible assets. In this case, they are also more likely to follow a dividend smoothing policy. The positive relationship between a firm’s liquidation value and the level of debt is predicted by both tax model and the agency model.

In contrast, firms that are unlikely to pay dividend are more likely to seek out the lower rescheduling risks attached to informed bank debt, if they are also smaller with few tangible assets. However, if these firms do pay dividends, they are more likely to follow a genuine residual policy, since there is little need for them to smooth their dividends. A positive relationship between collateralisable assets and dividend payout is expected. We use the ratio of net fixed assets to total assets as a proxy for tangibility of collateralisable assets in our empirical tests. v.

Financial Slack (FSLK) Financial Slack surrogate Business Risk and is proxied by long term borrowing to total assets. The theory of finance suggests that risky firms or firms that have high possibility to default should not be highly levered. High fixed operating costs or business risk may affect 21 the firm's dividend payout, all else constant, to the extent that these will increase the frequency of costly additional external financing. This is due to the greater variability in earnings and funding needs that high operating leverage or business risk may induce in a firm.

The same reasoning applies to interest charges, which are characterized by Rozeff (1982) as "quasi-fixed costs". Both these operating and financial risks translate into a high total risk of the firm’s stock returns. In addition, as observed by Holder et. al. , (1998), transaction costs of new issues in the form of underwriting fees is usually larger for riskier firms. According to the Pecking Order theory, firms should prefer to finance investment by retentions rather than by debt. A higher retention ratio implies a lower DPR, so a lower payout ratio should be associated with lower gearing rather than higher gearing.

Conversely, a higher payout ratio should be associated with higher gearing. Thus, if the sign of the regression coefficient attached to the gearing variable is positive, this would be consistent with both the pecking order theory and the greater financial risk proposition. Higher leverage ratios face the greater pressure of paying back the principal as well as the interest. The debt covenant may also prohibit the firms from paying higher dividend. Therefore the management tends to pay lower dividends for highly leveraged firms and thus a negative sign is expected. The expected sign of the coefficient of financial slack is negative. i. Cost of Borrowings (COBW) This cost is measured as the total interest payments adjusted to corporate tax rate to percentage of total borrowings of the firm. This variable would force the firms to distribute smaller dividends. When the cost of borrowing increases, the dependence on borrowed funds is likely to decline as a result the retention ratio is expected to have a positive relationship with the cost of borrowing. This would force the firms to exert more reliance on internal funds. Firms may get into financial distress if they fail to adjust themselves to adverse shocks.

Using interest coverage ratio consistent with James (1996) as the proxy of the severity of financial distress, Chemmanur and Fulghieri (1994) theorize that firms with lower financial distress probably opt for public debt against bank debt since the lower interest cost of public debt outweighs the benefits of flexible renegotiations in bank debt. A negative relationship between cost of borrowings and dividend is expected. vii. Operating Risk (ORSK) Operating Risk is a proxy for observable credit risk and increases the probability of our independent variable, Operating Risk (distress), Johnson (1997).

This variable is also 22 hypothesized to measure Earnings Volatility alternating Information Asymmetry consistent with Ikenberry and Vermaelen (1996). Many authors include a measure of operating risk as an explanatory variable, Titman and Wessels, (1988), Kremp et. al. , (1999), and Booth et. al. , (2001) implying riskiness of cash flows consistent with the Signaling theories that less volatile cash flow results higher future dividends. To the extent that the high figures of variability are correlated with firm’s FCF in the Jensen (1986) sense and associated agency costs, expected dividend payouts will be lower.

Two issues are particularly noteworthy. First, DeAngelo and Masulis (1980) argue that for firms which have variability in their earnings, investors will have little ability to accurately forecast future earnings based on publicly available information and the market will demand a premium to provide debt. This drives up the cost of debt. Second, to lower the chance of issuing new risky equity or being unable to realize profitable investments when cash flows are low, firms with more volatile cash flows tend to keep low dividends. A negative relation between operating risk and dividend is expected.

This relation also props up from a tradeoff theory and the pecking order perspectives; firms with high volatility of results try to accumulate cash during good years to avoid under investment issues in the future. The variable operating risk (earnings volatility) measured as the standard deviation in the ratio of operating income to total assets of the firm lagged three years. viii. Corporate Tax Rate (CTAX) It is suggested by a number of authors that the taxation policy of the government may negatively affect the dividends distributed by the company.

High corporate tax rates increase the total tax payments of the firm, reduces its net income which in turn, reduces its retained profit, Panda and Lal (1993), Damodaran, (2000). The impact of taxation on financing is twofold. On the one hand, companies have an incentive to take debt because they can benefit from the tax shield. On the other hand, since revenues from debt are taxed more heavily than revenues from equity, firms also have an incentive to use equity rather than debt. As suggested by Miller (1977), the financial structure decisions are irrelevant given that bankruptcy costs can be neglected in equilibrium.

DeAngelo and Masulis (1980) show that the firms with large non-debt tax shields have a lower incentive to use debt from a tax shield point of view, and thus may use less debt. Empirically, this substitution effect is difficult to measure as finding an accurate proxy for tax reduction that excludes the effect of economic depreciation and expenses is tedious, Titman and Wessels (1988). This variable is measured 23 as the ratio of the total tax payments to total profits before tax (with negative values truncated to zero).

CTAX is regarded as a simple control variable, with no particular sign expectation. 4. 2 Variables that Vary only over Time (Zt) Studies like that of Modigliani and Cohn (1979), Roppaport (1981), Lawson and Stark (1981), Lee (1992), Rao and Radjeswari (2000) for India, and Valckx (2003) document the influence the impact of macro-economic variables like inflation and interest rates on dividend policies. Such variables are assumed to vary only over time and have a uniform effect on dividend behaviour of firms.

The variables CPID and YLCR capture the effects of inflation and interest rate differential on dividend distribution decisions. i. Consumer Price Index Deflator (CPID) Consumer price index (inflation) would have a negative relationship with dividend and have a positive relationship with debt if higher inflation increases the wealth transfer to debtors, generated by the tax deductibility of nominal interest payments. We anticipate that increases in real general prices will generate upward pressure on firms' demands for funds thus, raise leverage and constrain dividends.

The higher is inflation the greater is the tax deduction gained by the borrower, not only on that component which reflects the real cost of funds but also on that part which represents compensation for reduction in the real value of t principal. However, the tax advantages of debt disappear under if borrowing rates increase more than one for one with inflation to keep after tax real returns unchanged, the increased tax deduction that inflation creates may be completely offset by higher borrowing costs.

It is also likely that an aggregate measure of the real cost of debt and an aggregate measure of the real cost of equity influence firms' gearing decisions. In equilibrium, the cost of debt, plus some risk premium, should be equal to the cost of equity. However, equilibrium conditions may not hold continuously. If this is the case, and if deviations in relative real cost of debt are not just firm-specific, then this factor may influence managers' gearing decisions. When the real cost of debt rises relative to the real cost of equity, firms can be expected to increase their gearing.

Such higher levels of debt are consistent with a greater likelihood of dividend omission and reductions as it increases the probability of financial distress in future years. This tendency is associated with the fear of assets seizure in case of default posted as collateral, psychological costs associated with bankruptcy and loss of control over the firm. A highly leveraged firm caused due to higher inflations would tend to lower its DPR because of high fixed financial commitments. A negative relation between inflation and dividend is expected. 4 ii. Yield Curve of Interest Rates (YLCR) Bolton and Freixas (2000) highlight the effect of monetary policies on corporate sector and adds that the effect may not be similar across the sample countries. Kashyap et. al. , (1993) argue that tight monetary policies increase the cost of banks’ capital, which in turn discourages firms from bank borrowings. In the same spirit, Oliner and Rudebusch (1996) contend that lenders would not be funding low-quality firms under such conditions.

Pandey and Bhat (2004) in their study of dividend behavior under monetary policy restrictions find that the restricted monetary policies have significant influence on the dividend payout behavior of Indian firms. The YLCR variable is measured as the difference between the call/notice money rates and the long term rates for term greater then 5 years for the fiscal year-end. A negative sign on the yield curve differential variable is expected. This is because, as the term structure of interest rate increases, relative cost of debt rises and in accordance with the Managerial Model have a negative charge on dividend distribution. . 3 Variables that Vary only across Firms (Wi) i. Industry Uniqueness Dummies These are categorization variables and used to pick up commonalities across industries. If a firm offers unique products or services, its consumers may find it difficult to find alternatives in case of liquidation, and hence, the cost of bankruptcy increases. Firms in the same industry also follow some different characteristics or procedures. Also many characteristics of the firms may be reasonable similar within the industry groups, but cannot be captured easily.

In a theoretical model, Titman (1984) shows that a firm’s financing depend on the uniqueness of its product. For these reasons the industry classifications of firms are included in our specification. Four broad industry classifications used here are Textiles (TEXL), Trading (TRDG), Chemicals, Cement and Metal Industry (CHCM), and Food Manufacturing (FDMG). Related to this prediction is the observation reported in Bhole (1980) and in Pandey (2001), that a firm’s industrial classification is an important determinant of dividends.

Their previous empirical results are in broad agreement and show that the in