In summary, the results of the proposed models are illustrated on task 4. In Fig. 5, the curves fitted to the training and test data are shown. Note that the MLP2 model performs best for price spikes. The cumulative distribution function corresponding to the quantiles calculated for task 4 by the MLP2 model is shown in Fig. 6. We can then read the correction 1C for the point forecast from this figure, to get its quantile; e.g., the quantile for p = 0.8 is about C ˆ(t) + 3. In our simplified approach, the corrections 1C for each point forecast C ˆ (t) , t = T + 1, T + 2, . . . , T + 24 are the same.