Evaluating probabilistic:

Hierarchical time series forecasting has been shown to improve typical forecasts in the point forecasting literature. However, there is little information about these results in the probabilistic forecasting setup. Recognizing the importance of probabilistic forecasts as both a prediction of future values and a measure of uncertainty, we evaluate two state-of-the-art algorithms for hierarchical probabilistic forecasting. In a Gaussian framework, we find that the same results of point forecasting hold, showing improvements over base forecasts through reconciling methods, with the MinT (Shrink) reconciliation being the best one overall. Nonetheless, the non-parametric framework shows different behavior, producing worst forecasts overall and less room for improving through reconciliation, at least with our selected data generating process.