Non-parametric estimation of forecast distributions in non-Gaussian state space models

Data in the economic realm often deviate from the continuous, symmetric bell shape of the Gaussian, or normal distribution, due to the inherent positivity, or discreteness, of economic and financial variables, and to the prevalence of extreme values. Such characteristics present challenges to the statistical analyst, not least of which is generating estimates of the future distribution of the variables over the different possible values. We propose a new approach to constructing accurate estimates of these forecast distributions. As well as yielding more reliable predictions, the proposal will produce significant methodological and theoretical advances in the application of non-parametric techniques to dynamic settings.