We start the empirical results with descriptive statistics. Then, we focus on the correlation matrix. Finally, we present the results of regression.
4.1. Descriptive Statistics and Correlation
provides an indication about the performance of our sample, board structure, and firm intrinsic characteristics. First, the ROA is around 3.76% with a wide gap between the two extremities. It seems that the firms composing our sample are performing modestly. Furthermore, we find that the average Tobin’s Q in our sample is 1.23. This value seems to be comparable to that found by Boubaker in the French context which is 1.99 [48
As regards to the corporate governance characteristics, we found that, on average, board of directors is composed of 11 members with a maximum of 24 and a minimum of 4, which reveals the diversity and heterogeneity of the selected firms. However, this board of directors is considered with middling size compared to that in other countries. In this regard, Chinese board size is, on average, around nine members. The same result was found by Firth et al. [49
] and Conyon and He [50
]. This finding leads us to conclude that board size depends on the actual firm size, its specificities, its industry, and its policy. In addition, we found that on average, 58.78% of our sample is composed of firms that have at least three committees in their boards. This finding is not unexpected because the firms in our sample have a large size, requiring implementation of certain committees to avoid conflicts of interests and to ensure transparency and integrity in the market.
However, the average of independent directors is 6 with a minimum of 0 and a maximum of 13 which shows that the board of firms of our sample is dominated by inside members, which is consistent with the recommendations of the best practices. Finally, our sample has an average debt ratio of 58.21% and an average age of 56 years.
Confirming to Kervin, the presence of multicollinearity can produce a large predicting error and make it difficult to assess the relative importance of individual variables in the model [51
]. Taking into account the multicollinearity which can lead to imprecise regression estimation, this study performs the pairwise correlations among regressors in the models which can be found in Table 3
We consider 0.8 as the limit value of the correlation coefficient, which corresponds to the limit set by Kennedy, to confirm the null hypothesis [52
]. Hence, if correlation between two variables exceeds 0.8, we have to reject the null hypothesis and we start having serious problems of multicollinearity. Overall, all the pairwise correlation coefficients between the explanatory variables are low. The largest pairwise correlation is 0.5217, indicating no problem with multicollinearity among independent variables.
4.2. Model Estimation Results
To validate the model specification, it is often useful to identify the effects associated with each individual. This effect can be fixed or random. The critical difference between fixed effects and random effects is that the fixed effect allows a correlation between the unobserved effect and the explanatory variables. However, the random effect requires the absence of correlation. If the unobserved effect is uncorrelated with the explanatory variables, then the random effects estimator is more operational than the fixed effects estimator. The Hausman test selects the best estimator between the random effects estimator and the fixed effects estimator.
In addition, the question of correlation and heteroscedasticity in the context of panel data is raised.
The first step is to check the existence of individual effects in our data. We seek, therefore, to test the null hypothesis: there is no individual effect. The result is an F-statistic. If we accept the null hypothesis of homogeneity, we obtain a completely homogeneous model pool. If we reject the null hypothesis, then we must include the individual effects in the model and move to the second step.
Indeed, in the two dependent variables, F-statistics are significant at the 1% level. The result allows us to accept the panel, and we conclude that there are individual effects. The Hausman test identifies the nature of these effects (fixed or random). It is a specification test which determines whether two estimations’ coefficients are statistically different. The idea of this test is that, under the null hypothesis of independence between errors and explanatory variables, both estimators are unbiased, so the estimated coefficients should differ slightly. Results reveal that random effects models fit better with our specifications when using Tobin’s Q, and reveal that the fixed effects model is accepted when using ROA.
Heteroscedasticity refers to data that does not have a constant variance. It does not bias the estimation of the coefficients, but the usual inference is no longer valid since the standard deviations found are not accurate. Heteroscedasticity is a situation frequently encountered in the panel data, so it is important to detect and correct it. In statistics, the Breusch-Pagan test is used to test for heteroscedasticity in a linear regression model. It tests whether the estimated variance of the residuals from a regression are dependent on the values of the independent variables.
In the context of a heteroscedasticity test, the null hypothesis is that all coefficients of the regression of squared residuals are equal to zero. In other words, the variance of each individual error is constant, so there is homoscedasticity. The alternative hypothesis is the assumption of heteroscedasticity. Thus, if we reject the null hypothesis, we can conclude the presence of heteroscedasticity. In our study, each probability is less than 1%. So, distributions are heteroscedastic.
It is also important to check the presence of correlation between errors and individuals. The Wooldridge test for autocorrelation in panel data checks that the sum of the squares of correlation coefficients between errors is approximately zero. The null hypothesis of this test is the independence of residues between individuals. The results have led us to reject the null hypothesis of no autocorrelation at the level of 1%.
provides the regression results. The addition of the interaction terms explains significantly more variance (∆Wald
-statistic = 7.18 for ROA and 6.5 for Tobin’s Q).
The results found from our model first reveal that the board size negatively affects market performance. Specifically, any increase in board size reduces the profitability. Indeed, in the French context, when the number of directors increases, coordination between them will be hampered and the board cannot make appropriate decisions which negatively impacts firm performance. This observation contradicts Godard’s statements that there is no difference in performance between firms with small boards and those with large boards [53
]. The explanation for this phenomenon is that board size alone cannot explain the differences in performance between French companies. Lipton and Lorsch suggested that large boards represent a disadvantage because they are associated with weaker monitoring and slow decision making [39
]. Nguyen et al. studied the effect of board size on firm value in Australia [7
]. Using a large sample of Australian firms over the period 2001–2011, they find strong evidence of a negative relationship.
Our results also show that independent directors have a negative influence on market performance only. There has been varied discussion for several decades as to whether board independence adds any value to the firm, with no definitive conclusion reached thus far. Although board independence is considered to be a very crucial corporate governance mechanism with regard to the ability to monitor shareholders’ interests effectively, our findings confirm that board independence in isolation does not necessarily enable good monitoring roles by the board; it also requires a skilled and knowledgeable board.
The coefficients on boards holding at least three committees are statistically insignificant, which suggest that the implementation of committees for a board did not influence firm performance. This latter finding contradicts results found by [34
] confirming that the implementation of committees is an appropriate control mechanism which reduces agency costs of French firms. A probable clarification is that the board of directors in France has collegial rights and responsibilities. This indicates that administrators, either individually or in small groups, somehow do not have authority over the board. Consequently, committees may only have a consultative power. We are far from the American conception of the operative of the committees according to which the latter engage their responsibility vis-a-vis the shareholders. Moreover, insofar as they can only be custodians of a delegation of authority by the board of directors, committees do not meet the independence requirement conveyed by the Anglo-Saxon concept in French law. Under these conditions, some authors have suggested that the establishment of specialized committees in France was of a purely formal nature, designed to meet the requirements of Anglo-Saxon investors, without bringing a reality of operation that meets the expectations of the market, especially in terms of independence vis-à-vis the CEO, [22
It is noteworthy that when we have included an interactive variable, board size with the existence of at least three committees, the impact of the interaction becomes positive on the accounting performance. Thus, a large board should have several committees that have the primary function resolving problems of coordination and asymmetric information between its different members, in order to properly fulfill its mission. Mandzila et al. argued that committees represent a mechanism for companies to arrange their boards such that they make effective use of directors’ time [36
]. Authors also indicate that, in order to enable the French corporate board to effectively play its role, the establishment of three committees—audit, remuneration, and nomination—was recommended.
Unlike our predictions, we have also found that the interactive variable between board independence and the existence of at least three committees negatively, but not significantly, affect accounting performance. Such a finding may only be clarified by the high costs incurred by the firm in the case of several committees adding to the independent directors’ costs. These costs negatively impact the accounting performance. In addition, the reinforcement of the control inside any company is able to limit opportunistic behavior. Thus, the accounting performance is reduced.
A number of associated control variables, according to previous work, were included in the analysis models. The firm size does not affect firm performance. This result counters the assumption that large firm size is a sign of growth and expansion. Additionally, analysis models also control for firm age. The coefficient on firm age is negatively correlated with the accounting performance in this study. One possible explanation is that firm age may increase managerial discretion because older firms have more knowledge in their business, which may help decision-makers to generate more strategic options, and thus increase managerial discretion. Finally, we also detected that debt does not affect firm performance, which is inconsistent with our predictions.
On the whole, the results reinforce in part the interpretation of the establishment of committees in terms of safety device to reduce agency costs in large corporate boards. When the size of the board increases, the establishment of committees is both more feasible and useful to the extent that it increases the effective functioning of this body.
The lack of clear influence of the interaction between committees and the proportion of independent directors, for its part, is consistent with the interpretation of the creation of committees in the French context as the establishment of a purely cosmetic device to formally satisfy the requirements of good governance practices. This result is consistent with Menon and Williams that have set up an audit committee [54
]. Suspecting that the mere creation of such committee does not mean that it actually operates efficiently, the authors studied the activity of these committees by trying to connect to the intensity of control needs in the company. They show an almost total absence of connection between the two categories of variables.