3.1. Sample Design
For the purpose of the study, a set of strongly balanced panel data consisting 112 manufacturing firms of BSE 200 index of India for the period of 2011 to 2018 are collected from financial database software namely Capitaline Plus and ‘ACE Equity’ developed by Capital Market Publishers Pvt. Ltd and Accord Fintech Pvt. Ltd. respectively. Further, the study also uses annual reports of the sample firms for different financial years. The study sets a range of x̅ ± 2σ (where x̅ and σ stand for the sample mean and standard deviation of each variable respectively) to eliminate the outliers from the panel dataset to avoid distorted results. However, we confine our study to manufacturing firms mainly because, the financial statements as a whole, capital structure, assets structure, working capital requirement etc. of service sector firms, especially, financial institutions are substantially different from that of other firms. Hence, inclusion of such firms would reduce uniformity and comparability of financial data across firms and results obtained thereby can’t be logically generalised for all the firms. Secondly, unlike other emerging Asian economies, concentration of ownership is much more prominent in manufacturing sectors in case of India (Selarka, 2005; Altaf, 2016). Hence, studies on ownership concentration in particular reasonably prefer manufacturing firms as the study sample.
3.2. Description of Variables
Ownership of domestic promoters
Ownership of domestic promoters is measured by the percentage of ownership stake held by the Indian promoters. The domestic promoters, by virtue of their considerable ownership rights, experience and expertise, are supposed to exert positive influence on the financial performance of firms by actively monitoring the activities of management.
Ownership of foreign promoters
Quite similar to the domestic one, a foreign promoter is also supposed to be highly aware, knowledgeable and competent in monitoring the management of affairs of the firm which he/she invested in. By working as an active monitor of the management the foreign promoters are also supposed to influence the functioning and thereby financial performance of a corporation.
Ownership of institutional investors
The category of institutional shareholders consists of banks, non-banking financial institutions, mutual fund companies, insurance companies etc. These financial institutions keep professionally qualified and highly experienced investment experts who undertake great deal of efforts in terms of rigorous monitoring and active participation in the management of affairs of the invested company to ensure good return on their investments. Therefore, institutional shareholding is another component of firm ownership that influences it performance and valuation.
Ownership of large shareholders
It is measured by the sum of holdings of the shareholders with at least five percent stake. The present study follows Salerka (2005), Vintila et al. (2014), Brendea (2014) to set a threshold of five percent shareholding for considering a shareholder as large. Moreover, for the purpose of testing the non-linear relationship, the study, following Miguel et al. (2004), Vintila et al. (2014), considers the square term of total concentration denoted by Large_Own2.
Ownership of the largest shareholder
In India, where most of manufacturing firms are family controlled (Selarka, 2005; Altaf, 2016) the largest owner plays the most dominative role in the management of affairs of companies. Recognizing the distinct importance of the largest shareholder, the study introduces this variable which represents the percentage of shareholding of largest shareholder in a particular firm for a particular year. To establish the non-linear relationship, the study following Vintila et al. (2014) again introduces the square term of this variable denoted by Largest_Own2.
Age
Age of firms, both theoretically and empirically, is known to have very strong connection with their efficiency, level of profitability and valuation. Age of firms is correlated with operational efficiency and performance of firms in number of empirical studies like Katz (1982), Hannan and Freeman (1984), Loderer and Waelchli (2010) etc. Therefore, this variable should be considered while modelling the relationship between ownership structure and firm value.
Liquidity
The theory of corporate finance advocates important implications of firms’ liquidity on operating efficiency and financial performance. The relationship is also sufficiently endorsed by number of empirical investigations like Saleem and Rehman, 2011; Niresh, 2012; Lartey et al., 2013 etc. This study takes firms’ liquidity measured by quick ratio as a control variable.
Assets utilization efficiency of firms
The assets utilization efficiency is measured by asset turnover ratio which is derived from dividing annual sales by average total assets of firms. It represents how efficiently the management of a firm is utilizing its assets to generate sales (Ang et al., 2000) and thereby to enhance performance of firms. It also reflects the existence of agency problem between owners and managers and the monitoring efficiency of large owners towards easing out such problem. ATR has been used as a popular representative of agency problem in number of studies Li and Cui (2003), Matusin et al. (2014).
Size
Size of firms is an important control variable in modelling the relationship between ownership structure and firm performance. The firm size is used as a control variable in many important corporate governance studies like Farooque et al. (2007), Zeitun (2009), Maqbool and Zameern (2018).
Leverage
Firm financial leverage is also proved to be an important determinant of a firms’ profitability and agency costs (Grossman and Hart, 1986; Jensen, 1986; Stulz, 1990, Pandey and Sahu, 2019b). Almost all of the studies which linked corporate governance parameters and corporate performance have considered firms’ financial leverage (mostly in the form of debt to equity ratio) as control variable.
Firm value
This study represents firm value by Tobin’s Q. Tobin’s Q is the most frequently used measure of firm valuation in most of the past studies like Morck et al. (1988), Demsetz and Villalonga (2001), Vintila et al. (2014), Mishra and Kapil (2017) etc.
Table 1
Variable
|
Acronym
|
Measurement
|
Ownership of domestic promoters
|
ODP
|
Percentage of shares hold by the Indian promoters
|
Ownership of foreign promoters
|
OFP
|
Percentage of shares hold by the foreign promoters
|
Ownership of institutional investors
|
OIIN
|
Percentage of shares hold by the institutional investors like banks, non-banking financial institutions, mutual fund companies etc.
|
Ownership of large shareholders
|
Large_Own
|
The sum of holdings of the shareholders with at least 5% stake
|
Ownership of largest shareholder
|
Largest_Own
|
Percentage of share hold by the largest shareholder
|
Age of firms
|
AGE
|
Age of the firm since establishment
|
Firms’ liquidity
|
QR
|
Ratio of quick assets to current liabilities
|
Assets utilization efficiency of firms
|
ATR
|
Ratio of annual sales to average total assets
|
Size of firms
|
FS
|
Natural logarithm of total assets
|
Leverage ratio firms
|
LVR
|
Ratio of debt to equity capital
|
Firm value
|
Tobin’s Q
|
The ratio of market value of equity plus book value of debt to total assets
|
Lagged dependent variable
|
Tobin’s Qit−1
|
One-year lagged Tobin’s Q ratio
|
Source: Prepared by Authors |
3.3. Methodology
The study first introduces static panel data model to establish the relationship between various forms as well as concentration of ownership and firm value. The static panel data model includes three regression models namely, Ordinary Least Square Model (OLS), Fixed Effect Model (FEM) and Random Effect Model (REM). The analysis also includes the selection of best fit regression model among these three models, because in panel data analysis it largely influences conclusions on the individual coefficients. In panel data, when the number of cross-sectional units is very larger than the number of time-series units, as in the present case, the estimates obtained by the FEM and REM differ significantly. Besides, all these three regression models have different underlying assumptions. The OLS model assumes the intercept as well as the slope coefficients to be same for all the 112 sample firms taken in the study. The FEM allows the intercepts to vary across the firms to incorporate special characteristics of the cross-sectional units. Finally, the REM assumes the intercept of a particular firm to be a random drawing from a large population which varies non-systematically with a constant mean value. As all these three conditions can’t prevail simultaneously, so the study needs to select an appropriate model for regression. The study introduces restricted-F test, Breusch-Pagan Lagrange Multiplier test suggested by Breusch and Pagan's (1980) and Hausman test suggested by Hausman (1978) to have a selection among the three regression models. The estimated model would be in the following form:
Tobin’s Q it = α + γ1 (ODP) + γ2 (OFP) + γ3 (OIIN) + γ4 (Large_Own) + γ5 (Large_Own2) + γ6 (Largest_Own) + γ7 (Largest_Own2) + β1 (AGE) + β2 (FS) + β3 (QR) + β4 (ATR) + β5 (LVR) + εit
Here, Tobin’s Qit refers to market value of ith firm at time period t, α represents the constant term, γ1 to γ7 represent the coefficients of ownership composition and concentration variables respectively, β1 to β5 represents the coefficients of the control variables and εit represents the error term.
Besides, considering the dynamism of the relationship and bias caused by potential endogeneity of the explanatory variables the study introduces Arellano-Bond (1991) dynamic panel estimation technique that determines the joint effects of the explanatory variables on the explained variable while controlling for potential bias due to endogeneity of the explanatory variables including the lagged dependent variable. The dynamic panel data model is mostly preferred when the number of cross-section units is larger than that of time series units, as in the present case. This is because of the fact that, the estimation methods don’t require larger time periods to obtain consistent parameter estimates (Mishra, 2008). The dynamic panel data regression model includes lagged dependent variable as one of the independent variable with the supposition that, the lagged dependent variable is correlated with the random disturbance term of the model and inclusion of it in the model accounts for the dynamic effects (Wintoki et al., 2012; Altaf and Shah, 2018). Notably, in such a situation when the lagged dependent variable is likely to be correlated with the error term of the model, the static panel data models like OLS and FEM become biased and thereby produce inconsistent estimates as these models largely ignore unobserved time-variant effects and the endogeneity of dependent variable.
Therefore, following the previous literatures we also take one year lagged Tobin’s Q as one of the independent variable to uncover the dynamism of relationship and to take into account the effect of some unobservable historical factors on the current dependent variable (Wooldridge, 2009).
Although, some of the previous literatures have considered instrumental variable in estimating dynamic panel data model (Anderson and Hsiao, 1982; Bhargava and Sargan, 1983) but following Mishra (2008) we adopt Arellano-Bond (1991) dynamic panel model which is based on Generalised Method of Moments (GMM). Besides, according to Ahn and Schmidt (1995) GMM estimator is likely to convey more information on data during the course of estimation than the method of instrumental variables. In GMM method we control the potential bias due to endogeneity of independent variables by taking one year lagged value of the lagged dependent variable and other independent variables as instruments (Basant and Mishra, 2013). Additionally, the study introduces Arellano-Bond test for autocorrelation and Sargan test (1985) of over-identification to check the presence of autocorrelation and validity of the instruments used in the model respectively.
Notably, there are two versions of Arellano-Bond estimator namely one step and two step estimator. The asymptotic standard errors of one step estimator are unbiased and more reliable to draw inferences on the individual coefficients but at the same time, under this estimation the Sargan test over-rejects the null-hypothesis of over-identification restriction in the presence of heteroskedasticty. Moreover, the robust standard error under one-step estimation can largely control the problem of heteroskedasticity but it can’t produce the Sargan statistic. Therefore, we execute both the estimations wherein we consider the individual coefficients of one-step estimation with robust standard error to draw inferences and the statistics of two-step estimation like Sargan statistic, Wald–Chi2 statistic to check the over-identification restriction and overall significance of the model. In nutshell, recognising the dynamism of the relationship and the issue of endogeneity we extend our analysis from static approach to dynamic approach which ultimately leads us to most robust estimates and thereby strong inferences.