Proponents of the nonparametric approach argued that it
provides better forecasts than parametric approaches in an array
of noised and chaotic case studies (Perretti et al., 2013a,b). They
further argued that their approach was easier to apply than
computationally demanding parametric approaches. The results
of this study do not contradict these two arguments. Nonparametric
forecasts were indeed found to outperform parametric ones
in a number of – but not all – cases with strongly reduced computational
costs (Fig. 4). However, no methods are yet available to
assess the reliability of nonparametric forecasts which is likely to
vary greatly (Fig. 4). Not knowing when a forecast is likely to
significantly fail is a major problem that should not be disregarded
(May et al., 2008).