Comment on ``Nonparametric forecasting of low-dimensional dynamical systems''