*3.3. Model Specifications*

Firstly, we analyzed whether the dividend policy for the NSE-listed companies varies across sectors. We used a one-way analysis of variance test to analyze whether there are any statistically significant variations between the means of three or more independent (unrelated) groups. This technique tests the null hypothesis that the mean values of dividend policy are the same for all sectors. The null hypothesis is rejected if the calculated F statistics is greater than F critical value, or if the *p*-value is less than ∝, where ∝ denotes the level of significance.

The study then uses the panel data techniques to develop a model establishing a relation between the dividend policy and the explanatory factors. Hsiao (2003) has highlighted that using panel data sets for research provides researchers with more advantages compared with using conventional cross-sectional or time-series data sets, for example (Hsiao 1985, 1995; Hsiao and Sun 2000).


**Table 2.** Description of variables with expected signs. ROE—return on equity; P/B—the price-to-book value.

The relation between dividend policy and the factors affecting it for selected NIFTY 500 companies and for each of the sectors is expressed in the form of an empirical model, as follows:

$$\frac{Div\_{\prime}t}{TA\_{\prime}t} = \infty \,\, i + \sum\_{j=1}^{n} \beta i\_{\prime} \, j \,\, \text{Xi}\_{\prime} \, j\_{\prime} \, t + \epsilon i\_{\prime} \, t \tag{1}$$

where *i* = 1, ... , *N*, where *N* is the number of cross-sectional units = 16 sectors, *t* = 1, ... , *T*, where *T* is the time period = 12 years and *j* = 1, ... , *n*, and where *n* is the number of explanatory variables = 12 (Hill et al. 2007). *Xi,j,t* is the explanatory variable *j* for firm *i* at time *t*, β*i,j* is the slope for explanatory variable *j* for sector *i*, ε*i,t* is the random error term for sector *i* at time *t*, *Divi,t*/*TAi,t* is dividend-to-total asset ratio subscripted for sector *i* at time *t,* and α*<sup>i</sup>* is the intercept.

Table 2 summarizes the dependent and explanatory variables, which form part of the empirical model expressed in Equation (1). For some companies, the data for LgSales, OpProfit, the price-to-book value (P/B), and CFPS are not available, hence the regression equation is estimated with the remaining variables, as the absence of this data limits the sample size of these sectors.

The major benefit of panel data is that it improves the efficiency of the econometric estimation by giving researchers more data points, which increases the degrees of freedom and reduces collinearity between explanatory variables (Gujarati 2009). Hence, this study uses the panel data technique, which will enable the maximum utilization of data by considering both cross-sectional and time-series data.

There are four basic models available in the panel data model, namely: pooled ordinary least squares (POLS), fixed effect model (FEM), random effect model (REM), and Seemingly Unrelated Regression Equations (SUR) model (Adkins 2010). For determining the most suitable panel data model, panel diagnostic tests are available in GRETL (Baiocchi and Distaso 2003). The first-panel diagnostic test being F-statistics. It suggests that if the *p*-value is less than alpha, then the null hypothesis is rejected, and FEM is more suitable than the POLS model. Secondly, the Breusch–Pagan test helps to determine whether POLS or REM is applicable. It suggests that if the *p*-value is less than alpha, then the null hypothesis is rejected, and REM is more suitable than the POLS model. If both the F-statistics and Breusch–Pagan test are found to be positive, then finally, the (Hausman 1978) test is used for determining whether FEM or REM is suitable for estimation. The SUR model is used to estimate panel data models that have large time periods and a small number of cross-sectional units

(Adkins 2010). Using these test results, we determine the appropriate model to be used for estimating the regression equation.

## **4. Empirical Results and Discussion**

#### *4.1. Discussion on Consolidated Regression*

In this sub-section, we present our panel regression results, which identify the factors influencing the dividend policy for selected NSE companies. As discussed in the methodology section, the panel diagnostic test results are utilized to determine the appropriate panel data model. The results in Table 3 reveal that the *p*-value of F-statistics is 0.935 which is >∝ (i.e., 0.05), hence we cannot reject the null. This suggests that POLS technique is more suitable than FEM (Adkins 2010; Cottrell and Lucchetti 2017). Hence, the findings are reported based on the POLS regression estimate.


**Table 3.** Consolidated (NIFTY 500) panel regression.

Notes: Regression coefficients are estimated using pooled ordinary least squares. Dividends divided by total assets is the dependent variable. The independent variables are as follows. Tangibility is total assets minus current assets divided by total assets. BusRisk is the standard deviation of ROI (lagged three years). LgSales is the natural logarithm of sales in local currency. LgMCap is the natural logarithm of market capitalization in local currency. OpProfit is EBIT divided by total sales. DebtRatio is the long-term debt divided by total assets. IntCover is EBIT divided by interest. CurRatio is current assets divided by current liabilities. ROE is NPAT divided by net worth. ROI is EBIT divided by CE. P/B is the market price per share divided by book value per share at the end of the year. CFPS is net cash flow from operating activities divided by the number of equity shares outstanding. The *t*-statistics are given in parentheses. \*\*\* Significant at the 1% level; \*\* significant at the 5% level. POLS—pooled ordinary least squares.

The results in Table 3 reveal that scale of operations, interest coverage ratio, current ratio, ROE, ROI, and P/B ratio have a positive relation to the dividend policy. This suggests that companies with the larger scale of operations, higher interest coverage ratios, current ratio, profitability and growth opportunities are likely to pay higher dividends. However, there is evidence that business risk, debt ratio, and CFPS have a negative relation with dividend policy for NSE firms. This implies that firms with higher business risk, debt levels, and cash flows are likely to pay lower dividends.

The results for the scale of operations, interest coverage ratio, current ratio, ROE, ROI, business risk, debt ratio, and CFPS, except for the P/B ratio, are in line with our hypotheses. We had expected that companies having more growth opportunities would most likely pay lower dividends, but for our sample, the P/B ratio had a marginal positive coefficient. For the remaining variables—tangibility of assets and operating profit—we found no significant relationship with dividend policy.
