The neural network generates point forecasts. To get probabilistic forecasts in quantiles, the standard deviation on the error from the training set is calculated as σ = std(C ˆ (t) − C (t)), t = T − 311, T − 310, . . . , T. When we have point forecasts C ˆ (t) for the test data and standard deviations σ , we use the inverse cumulative distribution