*3.2. Statistics*

We use both accounting measures like Return on Assets (ROA) and Return on Equity (ROE) and market measures that is Tobin's Q (TQ) of profitability. We classify the firms producing a single product as specialized with a value of 0 and firms producing more than one product as diversified firms with a value of 1. Afza et al. (2008) apply the same criteria to measure product diversification. The proxy for geographic diversification is the proportion of foreign sales to total sales. Schmid and Walter (2012) also use this proxy in their study. We measure capital structure as a ratio of total debts to total assets of the firm. Bhaduri (2002) uses the same measure of capital structure. We calculate dividend per share as the proxy of dividend policy. Oloidi and Adeyeye (2014) use the same proxy in their study. We use the proxy for investment policy as a change in investment in fixed assets. Aivazian et al. (2005) apply this measure of investment in their study.

We measure board size as a number of board members. Bhagat and Bolton (2008) use this proxy in their research. For CEO duality, we use value 1 if the CEO is also a director of the firm, otherwise, it is 0 (Bhagat and Bolton 2008). Audit quality characteristics include audit quality, audit committee size and audit committee activity. We take value 1 if the firm is audited by big four audit firms, otherwise, 0 followed the approach of Francis and Yu (2009). We measure audit committee size as a total number of members in audit committee (Azim 2012). We calculate the audit committee activity as the frequency of audit committee meetings in a financial year (Xie et al. 2003). We calculate the firms' size as the natural log of total assets. In our study, we measure growth as the percentage change in sales. Further, we measure the firms' age as the difference between the year in which a firm starts and the year in which it exists in the sample. Hunjra et al. (2014); Muritala (2012) apply similar calculations of size, growth and age for analysis. The description of variables is given in Table 2.



We use the following equations to analyze the results:

$$\begin{aligned} \text{ROA}\_{\text{i,t}} &= \alpha\_{\text{i,t}} + \beta\_{\text{1}} \text{PD}\_{\text{i,t}} + \beta\_{\text{2}} \text{GD}\_{\text{i,t}} + \beta\_{\text{3}} \text{CS}\_{\text{i,t}} + \beta\_{\text{4}} \text{DP}\_{\text{i,t}} + \beta\_{\text{5}} \text{IP}\_{\text{i,t}} + \beta\_{\text{6}} \text{BSIZE}\_{\text{i,t}} + \beta\_{\text{7}} \text{CEOD}\_{\text{i,t}} + \varepsilon \\ &\quad \beta\_{\text{8}} \text{AQ}\_{\text{i,t}} + \beta\_{\text{9}} \text{ACSIZE}\_{\text{i,t}} + \beta\_{\text{10}} \text{ACA}\_{\text{i,t}} + \beta\_{\text{11}} \text{AGE}\_{\text{i,t}} + \beta\_{\text{12}} \text{CRTH}\_{\text{i,t}} + \beta\_{\text{13}} \text{SIZE}\_{\text{i,t}} + \varepsilon\_{\text{i,t}} \end{aligned} \tag{1}$$

$$\begin{array}{l} \text{ROE}\_{\text{i,t}} = \alpha\_{\text{i,t}} + \beta\_{\text{1}} \text{PD}\_{\text{i,t}} + \beta\_{\text{2}} \text{GD}\_{\text{i,t}} + \beta\_{\text{3}} \text{CS}\_{\text{i,t}} + \beta\_{\text{4}} \text{DP}\_{\text{i,t}} + \beta\_{\text{5}} \text{IP}\_{\text{i,t}} + \beta\_{\text{6}} \text{BSIZE}\_{\text{i,t}} + \beta\_{\text{7}} \text{CEOD}\_{\text{i,t}} + \varepsilon\\ \beta\_{\text{8}} \text{AQ}\_{\text{i,t}} + \beta\_{\text{9}} \text{ACSIZE}\_{\text{i,t}} + \beta\_{\text{10}} \text{ACA}\_{\text{i,t}} + \beta\_{\text{11}} \text{AGE}\_{\text{i,t}} + \beta\_{\text{12}} \text{GRTH}\_{\text{i,t}} + \beta\_{\text{13}} \text{SIZE}\_{\text{i,t}} + \varepsilon\_{\text{i,t}} \end{array} \tag{2}$$

$$\begin{aligned} \text{TQ}\_{\text{i},\text{t}} &= \alpha\_{\text{i},\text{t}} + \beta\_{\text{1}} \text{PD}\_{\text{i},\text{t}} + \beta\_{\text{2}} \text{GD}\_{\text{i},\text{t}} + \beta\_{\text{3}} \text{CS}\_{\text{i},\text{t}} + \beta\_{\text{4}} \text{DP}\_{\text{i},\text{t}} + \beta\_{\text{5}} \text{IP}\_{\text{i},\text{t}} + \beta\_{\text{6}} \text{OSIZE}\_{\text{i},\text{t}} + \beta\_{\text{7}} \text{CEOD}\_{\text{i},\text{t}} + \varepsilon \\ &\quad \beta\_{\text{8}} \text{AQ}\_{\text{i},\text{t}} + \beta\_{\text{9}} \text{ACSIZ}\_{\text{i},\text{t}} + \beta\_{\text{10}} \text{ACA}\_{\text{i},\text{t}} + \beta\_{\text{11}} \text{AGE}\_{\text{i},\text{t}} + \beta\_{\text{12}} \text{GRTI}\_{\text{i},\text{t}} + \beta\_{\text{13}} \text{SIZE}\_{\text{i},\text{t}} + \varepsilon\_{\text{i},\text{t}}. \end{aligned} \tag{3}$$

We use descriptive statistics to check the normality of data and correlation is used to check the multicollinearity. We apply the Generalized Method of Moments (GMM), as this method performs consistent parameter estimation for the small time period and for a large cross-section. The GMM estimators enable asymptotically efficient inferences employing a relatively minimal set of assumptions (Arellano and Bond 1991; Blundell and Bond 1998). We deal with the unobserved heterogeneity by applying a fixed effect or by taking the first or second difference. The ability of first or second difference to remove the unobserved heterogeneity is developed for two-step dynamic panel data models. Furthermore, these models contain one or more lagged dependent variables and allow modeling of a partial adjustment mechanism.
