This section is deemed to be the most important part of the study. It principally describes how data were gathered, calculated and constructed. Before proceeding further, it is worthwhile to formulate the basic hypotheses on which the whole analysis will be based on. Based on the theories and empirical evidences gone through, the formal hypotheses stated in null and alternate form are as follows:
H0: Change in CSP is positively related to current and future changes in financial performance after controlling for size, industry and prior year's financial performance.
H1: Change in CSP is negatively related to current and future changes in financial performance after controlling for size, industry and prior year's financial performance.
As can be induced from the hypotheses, this section tends to test the significance between the independent and dependent variables, which is based on a multiple regression model to explain variations between the relationship of corporate financial performance (CFP) and corporate social performance (CSP) with other determinants as well, such as, firm's size and industry.
4.1 Research Questions
Is there an optimal CSP which maximizes CFP?
Is there a positive or negative relationship between CFP and CSP (causality or correlation)?
Has there been any major change in the trend of CSR contribution in the chosen companies?
What can be the other determinants contributing to growing CFP to firms in Mauritius?
4.2 Data
4.2.1 Primary Data
Primary Data are data that will be collected throughout this research. These include questionnaires, sampling and among others. But, for the current research purpose, use of secondary data will mostly be made which saves time that would otherwise be spent in collecting data and, particularly in the case of quantitative data, provides larger and higher-quality databases. However, to construct the variable of interest, the use of questionnaires will be done to gather up data to construct the index of CSP.
4.2.2 Secondary Data
Secondary Data are data that have been gathered by other users in the past. For the purpose of this section, these will include the annual reports, handbooks of the 47 companies listed on the Stock Exchange of Mauritius, that is, which are on the official market. From the annual reports, data such as the accounting measures, CSR contribution and so on will be used to analyse the relationship between CSP and CFP of the different companies over a period of 4 years (2007-2010). The annual reports are regarded as important documents chiefly because of their high credibility in lending information reported within them to a number of stakeholders and also due to their widespread distribution (Unerman, 2000) by the quoted companies.
This research consists of a random sample of 47 companies which are listed on the stock of exchange of Mauritius (SEMDEX). The list below shows the breakdown of companies in the sample by industry segment.
Listed Companies of SEM:
BANKS & INSURANCE AND OTHER FINANCE
The Mauritius Commercial Bank Ltd
Mauritian Eagle Insurance Co. Ltd
Mauritius Leasing Co. Ltd
Mauritius Union Assurance Co. Ltd
State Bank of Mauritius Ltd
Swan Insurance Co. Ltd
COMMERCE
Compagnie des Magasins Populaires Ltée
Harel Mallac Ltd
Innodis Ltd
Ireland Blyth Ltd
Rogers & Co. Ltd
Shell Mauritius Ltd
INDUSTRY
Gamma Civic Ltd
Phoenix Beverages Ltd
Mauritius Chemical & Fertilizer Industry Ltd
Mauritius Oil Refineries Ltd
Mauritius Stationery Manufacturers Ltd
Plastic Industry (Mtius) Ltd
United Basalt Products Ltd
INVESTMENTS
Belle Mare Holding Ltd
Caudan Development Ltd
Fincorp Investment Ltd
ENL Commercial Limited
The Mauritius Development Investment Trust Co. Ltd
National Investment Trust Ltd
Promotion and Development Ltd
P. O. L. I. C. Y Ltd
United Docks Ltd
LEISURE & HOTELS
Automatic Systems Ltd
New Mauritius Hotels Ltd
Naiade Resorts Ltd
Sun Resorts Ltd
SUGAR
Harel Freres Ltd
Omnicane
ENL Land Ltd
TRANSPORT
Air Mauritius Ltd
FOREIGN
Dale Capital Group Limited
AUTHORISED MUTUAL FUND
Ipro Growth Fund Ltd
Boyer Allan India Fund Inc.
Global Investment Opportunities Fund
Kotak Investment Opportunities Fund
Copex Diversified Strategies PCC
Global Diversified Fund PCC
Africa Sustainability Fund
Cerulean ( Mauritius ) PCC
Imara Portfolio Selector PCC
Grand Towers Africa Fund PCC
As data are being taken for a period of 4 years; the regression model to be based on such data will be a panel data regression one.
4.2.3 Panel Data
Basically there are three types of data that are generally available for empirical analysis, namely, time series, cross section and panel. In time series data, values of one or more variables are observed over a period of time while in cross-section data, values of one or more variables are collected for multiples sample units at the same point in time.
Panel Data, also called longitudinal data or cross-sectional time series data, are data where multiple cases, for instance, people, firms, countries and so on are observed over time, that is, at two or more time periods. In short panel data have space as well as time dimensions. There are other names for panel data, namely, pooled data which combine time series and cross-sectional observations; micro-panel data, longitudinal data which study a variable or group of subjects over time; event history analysis, which impose successive states or conditions upon the subjects and study the movement over time; and cohort analysis. Despite the variations in the names, all together connote the movement of cross-sectional units over time.
Panel data are now being used widely in economic research due to its multiple benefits they offer. Through the list of advantages provided by Baltagi [1] , the case to advocate for panel data over cross-section or time series data is given below:
Since panel data relate to individuals, firms, states, countries and so on, there is bound to be heterogeneity, that is, individuality or uniqueness in the units used for the regression. The techniques of panel data estimation can take such heterogeneity explicitly into account by allowing for subject-specific variables.
By combining time series of cross-section observations, panel data gives "more informative data, more variability, less collinearity among variables, more degrees of freedom and more efficiency."
By using the repeated cross-section observations, panel data are better suited to examine the dynamics of change. As such, spells of change in CSP and CFP are better studied with panel data.
Panel data are best at detecting and measuring effects that cannot be simply observed in pure cross-section or pure time-series data. As such, the effects of successive financial performance can be better studied in a panel data regression model.-
By making data available for several thousand units, panel data can avoid the biasness which can occur if broad aggregates of individuals or firms were taken.
4.3 Models
After gathering the data and formulating the hypotheses, the next step involves the modelling of the data into variables which will dependent and independent ones. The dependent variable is known as the regressand which is explained by the explanatory variables, that is, the independent variables, also known as the regressors. The main independent variable is called the variable of interest, while the other determinants included are called the control variables. Before erecting the econometric model, it is worthwhile to first build up the economic model which is as follows.
4.3.1 The Economic Model
Change in Corporate Financial Performance = Change in Corporate Social Performance + Size + Change in prior year's financial performance + Industry
After building up the economic model based on the various theories and empirical works analysed in the chapter before, an empirical model, based on the earlier stated hypotheses can be restated as follows.
4.3.2 Econometric Model
The following equation is estimated to examine the effect of change in CSP on CFP:
∆FINi,t = α + β1∆CSPi + β2Sizei,t + β3Industryi,j + β4∆FINi,t-1 + µt
where,
∆FINi,t = Growth in sales, ∆Return on Equity, or ∆Return on sales for firm i from time
period t-1 to t.
α = Constant, that is, the intercept
β1 = The regression coefficient for Change in Corporate Social Performance
∆CSPi = Change in CSP for firm i from 2007 to 2010
β2 = The regression coefficient for Size which is Log of Sales of firm i at time t
K = The number of industry categories.
Ii,j = The industry group to which firm i belongs, represented as a dummy variable.
β4 = The regression coefficient for the prior year's financial performance.
µt = error term
The degree of relationship is to be measured by computing the coefficient of correlation. The relationship between CFP and CSP is to be measured in a similar manner. It is to be noted that the data gathered are in panel data set. A panel dataset should have data on n cases, over t time periods, for a total of n Ã- t observations. Data like this is said to be in long form. In some cases, data may come in what is called the wide form, with only one observation per case and variables for each different value at each different time period. Hence, to analyse such data, we can make use of the software called Stata.
4.3.3 Software
Stata is a statistical package that integrates statistics with graphics and data management, as is common to most software these days. It is available for all kinds of operating system, be it for Windows, Macintosh, or Unix computers. Even though the software has brought along the new point-and-click interface, Stata is, however, more efficiently used by those with more than a cursory knowledge of the field, as due to the depth and breadth of the techniques offered. As such, this package is geared to the professional statistician or scientific researchers with in-depth knowledge in statistics. This general-purpose system is also intended for use by medical researchers, biostatisticians, epidemiologists, economists, sociologists, political scientists, geographers, psychologists, social scientists, and other research professionals.
Besides its general-purpose capabilities such as summary statistics, ANOVA, linear, logistic, and probit regression, and the like, Stata offers survival analysis with Kaplan-Meier survivor function estimates, Cox proportional hazards models, survey analysis, time series, multivariate analysis, and panel data estimators including random-effects, fixed-effects, and multilevel mixed-effects for continuous, binary, and count outcomes.
For the purpose of analysis and deep evaluation of the data collected, version 10 of Stata has been used, namely the STATA SE 10. Stata 10 is available for Windows (2000, XP, Vista), Macintosh, and Unix. Released in June 2007, Stata 10 is a powerful, versatile, and flexible statistical package with a wide range of user-friendly and accurate time series analytical and forecasting commands. The professional versions for multiple core/multiple processor units are capable of working with up to 32,766 variables, and observations limited only by memory. This version also briefly describes associated graphics, diagnostics, documentation, help facilities, output, customization, and support. Altogether, these features characterise a very well-designed and well-crafted and powerful time series analysis and forecasting package (Baum 2004).
Stata is now far more user-friendly, containing a wider menu of tests and a depth far beyond the needs of the casual user. To provide more support, the help contents are there for the novice in the form of manuals, on-line help, the FAQ Web sites and formal courses to get well-indexed information easily.
All in all, being is a complete, integrated statistical package that provides everything you need for data analysis, data management, and graphics, Stata 10 adds many new features such as multilevel mixed models, exact logistic regression, multiple correspondence analysis, a graph editor, and time-and-date variables. Hence, Stata is really appreciated and held in high regard for its excellence in many of the more advanced methods and specialised applications for its routines.
Stata provides a number of tools for analysing panel data. The commands all begin with the prefix xt and include xtreg, xtprobit, xtsum and xttab - panel data versions of the familiar reg, probit, sum and tab commands.
To use these commands, Stata needs first to be told that the dataset is panel data. A variable is needed to identify the case element of the panel and a time variable is also required which is to be in Stata date format. After inputting these variables, the data need to be sorted by the panel variable and then by the date variable within the panel variable. It is only after this process that the 'tsset' command can be issued to identify the panel and date variables.
4.4 Variables
4.4.1 Measurement of Financial Performance (CFP)
Based on the works of B. M. Ruf et al. (2001), the accounting measures such as the growth in sales, Return on equity (ROE) and Return on Sales (ROS) are chosen as the basic data to be used in the analysis for financial. Growth in sales here is measured by the percentage change in sales from one year to the next. The ROE is defined as the earnings before taxes divided by total stockholders' equity. ROS, on the other hand, is the net income before taxes divided by sales. Data to calculate all these financial measures are taken from the annual reports published by the 47 different companies. Concurrent to the above econometric model presented earlier on to be determined on a four-year time period, to construct the ∆CFP, change in growth in sales, ROE and ROS are combined for the year 2006 to 2007 (year 0) and for three subsequent years: 2007-2008 (year 1), 2008-2009 (year 2) and 2009-2010 (year 3).
4.4.2 Measurement of CSP
(Survey)
To evaluate whether companies are meeting the demands of their stakeholders, a corporate social performance (CSP) measure was developed based on a methodology proposed by Ruf et al. (1998). This approach develops a composite measure of CSP by 1) identifying the dimensions of CSP; 2) evaluating the firms' performance on these dimensions; 3) determining the relative importance of each dimension; and 4) synthesizing the results of relative importance and social performance scores. For this study, five CSP issues were identified: community relations, employee relations, environment, product/liability, and concern for women/minorities. Although consensus on the totality of relevant CSP issues has not been reached, these five issues have been consistently identified in the literature for over twenty years. For examples, see the Research and Policy of the Committee for Economic Development (1971), American Institute of Certified Public Accountants (1977), Ernst and Ernst (1978), and studies by Rockness and Williams (1988), Harte et al. (1991 and Kurtz et al. (1992).
Evaluation of the five CSP dimensions, the second step, was obtained from the Kinder, Lydenberg, and Domini Corporation (KLD) social rating system. Since its inception in 1990, KLD has annually rated approximately 650 firms on the five dimensions of CSP cited above. Although other rating systems exist, the KLD ratings are considered "the best-researched and most comprehensive CSP measure currently available" (Wood and Jones, 1995: 239). Several advantages of KLD over other measures of CSP are: 1) KLD employs consistent criteria annually, 2) KLD uses the same research staff to evaluate companies within an industry, 3) Annually, KLD evaluates a large number of companies with respect to each dimension. Performance ratings for the five CSP dimensions are on a 5-point scale from -2 (major concern) to neutral to 2 (major strength).
To determine the relative importance of the five CSP issues, a survey was mailed to 400 financial executives. The financial executives were selected from the Financial Executive Institute mailing list. The response rate was thirty-two percent. Financial executives represent high level corporate management who are generally heavily involved with strategic planning and policy decision. Because strategic planning and policy decisions cover corporate social performance issues as well as merger and acquisition activities, this group of stakeholders was considered appropriate for the survey.
Relative importance of CSP dimensions is determined in the survey by using the principles of the Analytical Hierarchy Process, AHP (Saaty, 1980, 1986). The AHP provides a "fundamental scale of relative magnitudes expressed in dominance units in the form of paired comparisons" (Saaty, 1980). For each pair of social dimensions, the respondents were asked to indicate (a) their preference for a particular dimension, and (b) the strength of their preference using a scale of "equal importance" to "absolutely more important" using a scale of 1 to 9. (For further discussion on the procedures for deriving the weights see Ruf et al. (1998). The aggregation of the results of the questionnaire represents an overall measure of the relative importance of the dimensions for the entire group of respondents, with the aggregated weights assigned to each dimension of social performance. An independent evaluation of the firm's performance on each dimension is determined next, ranking the performance of a given company on each dimension of social performance. The product of the performance score on a given dimension and the weight of that dimension are then computed. The process is repeated for each dimension. Finally, the composite measure of CSP is computed as the sum of the products. Graves and Waddock (1994) also advocate and use this method of determining CSP.
(Computing Index)
4.5 Tests to be performed
Since the data concerned are of panel data category, the following suit of tests found on Stata menu can be carried out for dynamic panel data which are highlighted below.
4.5.1 Test of stationarity
The test of stationarity is compulsory. According to K. P. Jönsson (2005), there are some previously suggested moments that are to be used when standardising the panel data stationarity test. But, these cause size distortions when samples are small and serial correlation in the disturbance terms is allowed for. Since, the data-set is of finite sample, there is a need to standardise the moments that are to be used in a panel data stationarity test when samples are small and serial correlation in the disturbances may be an issue. A serious small-sample bias in the panel data stationarity test can be documented when a linear trend is present in the data.
4.5.2 Unit Root Tests
Using Stata's new xtunitroot command implements a variety of tests for unit roots or stationarity in panel datasets. The Levin-Lin-Chu (2002), Harris-Tzavalis (1999), Breitung (2000; Breitung and Das 2005), Im-Pesaran-Shin (2003), and Fisher-type (Choi 2001) tests have as the null hypothesis that all the panels contain a unit root. The Hadri (2000) Lagrange multiplier (LM) test has as the null hypothesis that all the panels are (trend) stationary. Such options permits fixed effects and time trends to be included in the model of the data-generating process. Fixed effects regression is the model to use when you want to control for omitted variables that differ between cases but are constant over time. It lets you use the changes in the variables over time to estimate the effects of the independent variables on your dependent variable, and is the main technique used for analysis of panel data.
The assorted tests make different asymptotic assumptions regarding the number of panels in the dataset and the number of time periods in each panel. xtunitroot has all the bases covered, including tests appropriate for datasets with a large number of panels and few time periods, datasets with few panels but many time periods, and datasets with many panels and many time periods. The majority of the tests assume a balanced panel dataset.
4.5.3 Granger Causality
The Granger Causality Test looks at the causality relationships among the variables. Granger causality will also be conducted to test for causality. These tests will examine both long run and short run relationships between CSP and CFP and the other variables used. The VECM analyses can also be included to provide some support for the argument that the lagged values of CFP have a significant influence on change of CFP. There have been a variety of proposed methods for implementing stationarity tests (for example, Dickey and Fuller, Sargan and Bhargava; Phillips and Perron, among the others). Granger causality can measure temporal ordering, high correlation and predictive ability, which are the necessary elements of causality (Malina, 2007).
A paper by C. Hurlin and B. Venet (2003) proposes a simple procedure of causality tests in panel data models with fixed coefficients, based on Granger (1969). Given the heterogeneity of the data generating process, four definitions of causality relationships were proposed. A procedure of tests was then defined and a particular attention was done to finite sample properties of these tests. Likewise, an application to the link between CSF and CFP can be proposed with a panel data of the 47 listed companies over the period 2007-2010.
4.5.4 Impulse Response Analysis
When further aspects of short run dynamics of the relationship cannot be addressed by Granger causality, the Impulse Response Analysis is used whereby it traces out the response of a variable when a shock is given to it.
4.5.5 Vector Error Correction (VEC)
A Vector Error Correction (VEC) model is a restricted VAR that has cointegration restrictions built into the specification, so that it is designed for use with nonstationary series that are known to be cointegrated. It restricts the long-run behaviour of the endogenous variables (variables with causal links) to converge to their cointegrating relationships while allowing a wide range of short-run dynamics. The error correction term is the cointegrating term since the deviation from long-run equilibrium is corrected gradually through a series of partial short-run adjustments.
4.5.6 Arellano-Bover/Blundell-Bond system Estimator
Using a dynamic panel-data analysis on Stata, improved command 'xtabond' implements the Arellano and Bond estimator, which uses moment conditions in which lags of the dependent variable and first differences of the exogenous variables are instruments for the first-differenced equation.
4.5.7 Ssmaller bias with persistent AR processes
'xtdpdsys' is an extension of 'xtabond' and produces estimates with smaller bias when the AR process is too persistent. An Autoregressive (AR) process is a statistical forecasting model in which future values are computed only on the basis of past values of a time series data.
4.5.8 Serially Correlated Disturbances
Because serial correlation in linear panel-data models biases the standard errors and causes the results to be less efficient, researchers need to identify serial correlation in the idiosyncratic error term in a panel-data model. A new test for serial correlation in random- or fixed-effects one-way models derived by Wooldridge (2002) is attractive because it can be applied under general conditions and is easy to implement. A paper by D. M. Drukker (2003) presents simulation evidence that the new Wooldridge test has good size and power properties in reasonably sized samples.
4.5.9 Test Over-identifying Restrictions
The overid routine has been extensively rewritten. It can be invoked after ivreg or ivreg2, and now produces a variety of test statistics for over-identifying restrictions (Sargan's statistic with and without small-sample correction, Basmann's statistic, and pseudo-F forms of each). The help file is much more informative, with references to the econometric literature.
The effect of varying the number of moment conditions used on the finite sample properties of the Sargan test of over-identifying restrictions is investigated in the context of dynamic panel data models. The use of too many moment conditions causes the test to be undersized and to have extremely low power. Interestingly, the Exponential Tilting Parameter test is found generally to possess worse size properties than the Sargan test in this context.
4.6 Expected Outcomes
The statistical tests will involve a null hypothesis that there is a relationship between the CSF and CSP. Thus, the expected outcomes would be to see a significant positive relationship between a change in CFP being affected by a change in CSP and other relative determinants over the period 2007-2010 as per the stated hypotheses. If this holds true, the same can be used for forecast in the coming years.
