Summary

Uncertainty intervals are a vital tool for understanding your forecast. No prediction of the future can have absolute confidence. By explicitly stating the confidence level in your model, you provide your audience with an understanding of the risk involved in the model's predictions, to better guide their decisions.

In this chapter, you learned that all models built in previous chapters used MAP estimations to create confidence levels. This method requires less time to compute than the alternative, MCMC sampling, but can only model uncertainty in the trend component. Often, this is enough. However, for those times when you also need uncertainty stated for seasonality, holidays, or extra regressors, you also learned how to apply MCMC sampling in Prophet to build a more comprehensive model of uncertainty.

Finally, you learned of an inherent weakness of MCMC sampling in terms of its ability to apply regularization to trend changepoints. You will usually see a larger...