We are interested in testing the existence of

We are interested in testing the existence of multiple regime dynamics in industrial productivity growth across countries. Therefore, the whole point in this paper is to allow the convergence β to vary. We estimate the following equation: Note that we did not include country dummies. This way, our results are directly comparable to the findings in Rodrik (2011b). To reduce the computational cost especially in the semi-parametric specifications, from now on we give up using the interaction of industry and period dummies. Fig. 1 shows the histogram of the estimated \'s. In Panel A, we see the results of regression (3) with no dummies; in Panel B, period dummies are included; finally, in Panel C the equation has industry and period dummies. The \'s histograms suggest that the dispersion of the convergence coefficient distribution should not be neglected. For the specification with no dummies, the standard deviation/mean ratio of the \'s is 16.2%; in the case with only period dummies, this ratio is 15.3% and, with period and industries dummies, it is 12.3%. We performed Wald tests, in which the null Sulindac sulfide cost is that all countries have the same coefficient. In all three cases (models with no dummies, with only period dummies and with period and industry dummies) the null is rejected (F-statistics around 5000 for the first two cases, and of over 9000 for the last specification) . Note also that in all the three cases, the estimated convergence coefficients are larger (in absolute value) than the analog estimated coefficients in Eq. (2), shown in Table 1. Actually, they are much closer to the ones estimated in Eq. (2) where country dummies are included. In a cross-country regression, the fact that the estimated is typically negative derives from the empirically suggested fact that industry productivity countries with low industry productivity levels grow faster than the analog for countries with high industry productivity levels. This could be a sign unconditional convergence, i.e., that there is only one steady state level of industry i productivity across countries. But note that if β is different across countries, their steady-state levels of productivity are also different. To see that, consider a model with only one industry. Hence,where α=D+π. In the steady-state, period effects vanish (i.e., D=D and π=π), Δυ=π and y equals the steady-state level . We can then write Therefore, if we can find variables that help us to group countries with the same β, we will also be identifying convergence clubs. To capture more accurately the relationship between the relative productivity growth β and country-specific indicators, we allow the convergence coefficient to also vary across decades. This way, we gain one more source of variation. We now estimate the following equation: The exact way this equation is estimated is shown in Appendix A. Table 3 shows the results of the linear regression of \'s (estimated in the equation with industry and period dummies) on various indicators, measured as its decanal initial level. We estimate:where is a combination the following country indicators: latitude, longitude, trade openness, executive constraints and years of schooling. Eight overlapping different decades are covered (1990–2000 through 1997–2007) so that each country enters the data (a maximum of) eight times. Linear regressions indicate that a more educated population in the beginning of the period is associated with a large industry productivity growth. One standard deviation of years of schooling is related to an increase in the convergence coefficient of 0.002–0.003, depending on the covariates considered. Once the standard deviation of is 0.0069 for the model with industry and period dummies, the magnitude of the estimated relation between years of schooling and the convergence coefficient is relevant. Note also that the R2 of regressions that involve years of education are larger then the ones that do not involve this education indicator.