Table 4 presents the mean statistics and t-test of important variables such as RIGHT, LAW, Q, and SIZE, by groups of legal system.
The mean of RIGHT is 7.432 for common-law countries group, and the difference of RIGHT between common-law and civil-law countries group is statistically significant with P value less than 0.0001. We also find that the difference of LAW between common-law and civil-law countries group is statistically significant with P value. This is similar with previous studies that show firms in common-law countries have better country-level governance than those in civil-law countries (La Porta et al., 1998).
Next, the difference of Q and SIZE between common-law and civil-law countries group is also statistically significant with P value. This indicates that firms in common-law countries have more investment opportunities to pursue the corporate value than those in civil-law countries because large firms with lower financial constraints can raise funds easily without
1
Collins, Rozeff, and Dhaliwal, 1981).
Overall, the univariate statistics show that common-law countries have stronger country-level governance than do civil-law countries; therefore, their stockholders have higher shareholder rights, and firms in common-law countries have lower financial constraints.
Hence, firms in common-law countries have more investment opportunities to pursue corporate value than those in civil-law countries.
Table 4 Whole sample firms t test of important variables by legal system
This table presents the mean statistics of important variables and tests the differences of important variables, such as RIGHT, LAW, Q, and SIZE, with t test between common-law and civil-law countries. The data consist of 9,030 firm-year observations (firms in common-law countries = 7,483; firms in civil-law countries = 1,547) for the period 2002–2008. The RIGHT is the index of shareholder rights. The LAW is the index of rule of law.
Both RIGHT and LAW are collected from the IMD World Competitiveness Online, and the index is from 0 to 10 (best). The Q here is (market value of equity＋book value of liability) divided by book value of total assets.
The SIZE is natural logarithm of firms’ total assets. P value represents P values from t tests for difference in means with unequal variances. *, **, and *** indicate significance at 10%, 5%, and 1%, respectively.
Variables Total Common-law
In the present paper, we use OLS regression and random effects regression to consider cross-sectional and time-series effects. The dependent variable in our models is CASH. The explanatory variables are RIGHT, and the common law dummy. The other variables, which have been shown to be statistically significant in explaining CASH in previous empirical studies, are control variables in our models.
Table 5 reports the OLS regression results. To avoid multicollinearity, we do not place these explanatory variables into the same model because the shareholder rights and the rule of law, as proxies for country-level governance, are highly correlated.
Model (1) in Table 5 contains only the level of the shareholder rights and industry dummies, defined at the two-digit SIC code level, as explanatory variables. Consistent with
our expectation, RIGHT is positively related to cash holdings. According to previous empirical studies, firms with weaker corporate governance structures have smaller cash reserves (Harford, Mansi, and Maxwell, 2008).
Model (2) in Table 5 also contains only the industry dummies and the common law dummy (Common), which is defined as 1 for common-law countries and 0 for civil-law countries, as explanatory variables. The coefficient of Common is significantly positive to cash holdings at 0.024, which indicates that the common-law countries hold more cash holdings than do civil-law countries. This is consistent with previous studies that firms with better country-level governance reduce the sensitivity between cash holdings and agency problem because the stockholders of firms with better country-level governance have more shareholder rights to monitor managers and to avoid agency problem between managers and shareholders (Hillier et al., 2011).
Next, Models (3) and (4) repeat the previous analysis; however, they include another three factors (i.e., investment opportunities, financial constraints and firm-specific characteristics) as control variables in models. The coefficient on RIGHT increases from 0.011 in model (1) to 0.013 in model (3). In addition, all our control variables are significant and have the expected sign. Thus, controlling for industry alone is not sufficient to capture the dispersion in the cash ratio.
Consistent with previous evidence, small firms with higher investment opportunities have higher information asymmetry than large firms; thus, they have more cash holdings than large firms (Bates, Kahle, and Stulz, 2009; Dittmar, Mahrt-Smith, and Servaes, 2003; Opler et al., 1999).
Moreover, because leverage (LEVERAGE) and non-cash liquid assets (NWC) are substitutes for holding high levels of cash, firms can use them when they have cash shortfalls.
Hence, LEVERAGE and NWC are significantly negatively related to cash holdings (Baskin,
idiosyncratic risk as proxy for firm specifics in our samples. Consistent with previous evidence, firms with higher idiosyncratic risk are expected to hold more cash to avoid passing up the valuable investment opportunities (Minton and Schrand, 1999; Ferreira and Vilela, 2004).
Table 5 Pooled cross-country regression
This table reports the OLS regression results of explanatory variables, such as RIGHT and the common law dummy, on CASH for the period 2002–2008. All regressions include industry dummy variables, defined at the two-digit SIC code level. Next, we divided our sample into two groups: common-law and civil-law countries. *,
**, and *** indicate significance at 10%, 5%, and 1%, respectively. The values of t statistics are in parentheses.
Variable Predicted sign (1) (2) (3) (4)
Because the OLS regression may have interdependencies of observations within an industry and within a country, we use random effects regression to make sure that our findings persist after controlling for these independencies (Wooldridge, 2002, p. 169).
The results of this analysis are reported in Table 6 using the same structure in Table 5.
After examination, we find that the coefficients of our explanatory variables, such as RIGHT and the common law dummy, remain highly significant in all models. In addition, the coefficients on the control variables are also similar in magnitude and significance to those reported in Table 5.
Table 6 Pooled cross-country regression with country and industry random effects
This table reports random effects regression results of explanatory variables, such as RIGHT and the common law dummy, on CASH for the period 2002–2008. All regressions include industry dummy variables, defined at the two-digit SIC code level. Next, we divide our sample into two groups: common-law and civil-law countries.
*, **, and *** indicate significance at 10%, 5%, and 1%, respectively. The values of t statistics are in
In many situations, data are observed in clusters, such that observations within a cluster are correlated, whereas observations between clusters are uncorrelated; these are so-called cluster-correlated data. A major statistical problem with cluster-correlated data arises from intracluster correlation or the potential for clustermates to respond similarly. This phenomenon is often referred to as overdispersion or extravariation in estimated statistics beyond what is expected under independence (Williams, 2000).
To avoid possible spurious relationship on panel regressions and to increase credibility on regression coefficients, we use clustered robust standard errors model proposed by Petersen (2009) as robustness to check it.
Table 7 Robustness tests of pooled cross-country regression
This table reports the original panel regressions and the panel results adjusted by clustered robust standard errors as robustness check. Models (1) and (2) are random effects models. Models (3) and (4) are robustness tests after adjustments. *, **, and *** indicate significance at 10%, 5%, and 1%, respectively. The values of t statistics are in parentheses.
Table 7 reports the original panel regressions and the panel results adjusted by clustered robust standard errors as robustness check. Models (1) and (2) are random effects models.
Models (3) and (4) are robustness tests after adjustments.
After adjustments, we find that most estimating errors on panel regression coefficient estimates in Table 7 are decreasing; however, the degree of decrease in the panel regression coefficient estimates is inconsistent. Empirical results are consistent with the original before adjustments.