Turning now to the conditions for the BWM-estimator to be consistent for the parameters of the conditional mean, we investigate the skewness and heteroskedasticity of the conditional distribution of ln(zc). For this particular model and sample, the statistic for the non-normality-robust symmetry test of Godfrey and Orme (1991) has a value of 5.21, to which corresponds a p-value of 0.02, thus confirming the significance of the skewness revealed by Figure 1. As for heteroskedasticity, we use the non-normality-robust version of the Breusch and Pagan (1979) test proposed by Koenker (1981) to check whether the variance of εc depends on ln(rc), the regressor of interest whose coefficient is more sensitive to the choice of estimator. The test statistic in this case has a value of 22.26, to which corresponds a p-value virtually equal to zero. Therefore, in this particular application, there are signs of skewness and very strong heteroskedasticity, and therefore the OLS and the BWM-estimator are likely to identify measures of central tendency with different slope parameters.