Of course, if δ is allowed to go to zero as the sample size passes to infinity, the properties of the BWM-estimator based on (3) are very different. In this case, under suitable regularity conditions, it can be shown that the estimator is consistent for the conditional mode of yi given xi, even if the errors are skewed and heteroskedastic ( see Kemp and Santos Silva, 2010). However, it is important to note that, although of interest in itself, the conditional mode does not generally coincide with the conditional mean and has very different properties.