Prediction intervals based on multiplicative model combinations

The aim of this paper is to define prediction intervals based on multiplicative combination of elementary density functions as an useful surrogate of the true unknown predictive model for the interest random phenomenon. The specification of the weights associated to the individual density forecasts is performed by considering the continuous ranked probability score (CRPS) and its weighted extensions. A simple simulation study shows that, using a suitable weighted version of the CRPS, the estimated combined model provides prediction intervals having a coverage probability closed to the target nominal value.