Conditional distributions for the analysis of convergence are usually estimated using a standard kernel smoother but this is known to be biased. Hyndman et al. (1996) thus suggest a conditional density estimator with a mean function specified by a local polynomial smoother, i.e. one with better...