Using a variant of the methods of Fama and MacBeth [1973],we estimate annual cross sections of (2) with statistical significance assessed within each year (by cross-sectional standard errors) and across all years (with the time-series standard error of
the mean coefficient). This method of assessing statistical significance deserves some explanation. In particular, one logical alternative would be a pooled setup with firm fixed effects and time-varying coefficients. We rejected this alternative mainly be
cause there are few changes over time in the Governance Index,and the inclusion of fixed effects would force identification of the G coefficient from only these changes. In effect, our chosen method imposes a structure on the fixed effects: they must be a
linear function of G or its components