RDP 9211: Dividends and Taxation: A Preliminary Investigation 5. The Determinants of Dividend Policy

The simple model of dividend behaviour presented below is based on the assumption that companies partiallyadjust dividends each period towards their target level. This partial adjustment behaviour is a common assumption in dividend models and may reflect both signalling behaviour and constraints that some companies face in capital markets. This target level of real dividends per share is given by,

where D*i,t is the target real dividend level per share, Xi,t represents the set of explanatory variables that affect the target level of real dividends per share, δi is a firm specific effect that captures the possibility that some firms may systematically wish to pay higher dividends, i=1,...., 55 is the company index and t=1, ....,11 is the time index. The adjustment of actual dividends to their target level is assumed to be determined by,

where ζ is the adjustment parameter. Substituting (1) into (2) and rearranging,

where Inline Equation

Note that the intercept term varies across firms but the other parameters are constant across firms. The set of explanatory variables Xi,t includes the tax variable θt, calculated earlier, and a cashflow measure to capture the capacity of companies to distribute funds to shareholders. The cashflow measure, Ci,t, is real cashflow per share after adjustment of revenue and cost items associated with options, partly paid shares and convertible securities and after adjustment for intervening pro-rata issues to existing shareholders.[5] The dividend measure is real dividends per share adjusted for bonus and rights issues. All variables, with the exception of the tax discrimination variable, are in logarithms.

Other potential determinants of dividend policy were also investigated. These included capital gearing variables; the cost of debt and cost of equity; industry dummies to capture the effects of possible common features within industry groupings such as capital structure, asset profile, operating cashflow requirements, size and profitability; a size variable to proxy firms access to capital markets; and a depreciation variable. There was no evidence that these factors had a significant influence on dividend policy and they were not included in the estimating equation reported below.

5.2 Estimation

Typically, panel data models are estimated using either fixed or random effects techniques. The fixed effect technique is predicated on the assumption that the effects which are specific to each firm are correlated with the other explanatory variables. As a result the individual effects are captured by the inclusion of a firm-specific dummy variable in each equation. The random effects estimator is based on the assumption that the individual effects are uncorrelated with the explanatory variables. As a result they are included in the error term and Generalised Least Squares is used.

In this case, neither technique is appropriate as (3) includes a lagged dependent variable (see Hsiao (1986)). Consistent estimates can, however, be obtained by first differencing (3) and then using instrumental variables,

The instruments used are the logarithm of the level of real dividends per share in period t−2, the contemporaneous change in the logarithm of real cash flow per share and the contemporaneous change in the tax discrimination variable. In the estimation, corrections were also made to the covariance matrix to allow for conditional heteroskedasticity. We also perform a test for parameter stability. To do this we split the sample into two subperiods 1980/81 to 1984/85 and 1985/86 to 1990/91 and perform a Wald test of the null hypothesis that the parameters are constant across periods. This procedure is used because we use an instrumental variables estimation technique and because we apply a heteroskedasticity correction. A review of estimation of panel-data models is provided in Appendix 1.

5.3 Results

The results of estimating equation (3′) are reported in Table 5. The tax discrimination variable and the cashflow variable are correctly signed and significant. A one per cent rise in real cashflow per share causes a 0.39 per cent rise in real dividends per share in the short run and a 0.58 per cent rise over the long run. This is consistent with the payout ratios described in Section 4 and suggests that the capacity to paydividends is an important determinant of dividend payout behaviour.

Table 5: Estimation Results

Estimated Equation:

Di,t − Di,t−1 = β1(Di,t−1 − Di,t−2) + β2t − θt−1)+ β3(Ci,t − Ci,t−1) + (εi,t − εi,t−1)

Estimation Period: 1981–1991; 55 Non-Financial Companies.

Tax Variable 0.34
Real Cashflow/Share 0.39
Real Dividends/Share lagged 0.32
(t-statistics in brackets)
Inline Equation
Structual Stability Test:
χ2(3) = 0.065 (p = 0. 99)

In addition, the results support the idea that the tax system provides an incentive structure that influences whether dividends or capital gains are the preferred form of returns for investors. A 0.1 unit rise in the tax discrimination variable causes a 3.4 per cent rise in real dividends per share in the short run and a 5.0 per cent rise over the long run. A 0.1 unit rise in the tax discrimination variable is interpreted as an additional 10 per cent of a unit of after-tax dividends foregone for any given level of after-tax retained earnings.[6]

The point estimate of the coefficient on the lagged dependent variable suggests that dividends adjust fairly quickly to the desired level, with about two thirds of the adjustment occurring in the first year. The explanatory power of the model was poor. Although other factors, such as those mentioned above, may also influence dividend decisions, the influence of these factors cannot be identified in our sample. The null hypothesisof structural stability between the first and second half of the sample period could not be rejected.

The results of these regressions allow an estimate of the increase in dividend payments that resulted from the tax changes implemented since 1985 to be made.[7] Real dividends per share increased by about 38 per cent between 1985/86 and 1990/91. We estimate that the tax changes accounted for about a 20 per cent rise in real dividends per share over this period.


For a discussion of the construction of these ratios see ‘The STATEX Guide to Ratios’, Australian Stock Exchange. [5]

The effective marginal tax rates are bounded by 0 and 1 and these results apply to values within the normal ranges of these variables. [6]

The full effect on real dividends per share over the five-year period is calculated using an expanded version of equation (3') in which the lagged dependent variable is replaced using backwards substitution. The changes in the tax discrimination variable in each year are calculated and the estimated coefficients are used to calculate the current and lagged effects. [7]