Chapter 10: Uncertainty Intervals
Forecasting is essentially predicting the future, and with any prediction, there will necessarily be a particular amount of uncertainty. Quantifying this uncertainty provides the analyst with an understanding of how reliable their forecasts are and it provides the manager with the confidence to stake a lot of capital on a decision.
Prophet was designed from the ground up with uncertainty modeling in mind. Although you interact with it in either Python or R, the underlying model is built in the Stan programming language, a probabilistic language that allows Prophet to perform Bayesian sampling in an efficient manner to provide a deeper understanding of the uncertainty in the model, and thus the business risk of the forecast.
There are three sources of uncertainty that contribute to the total uncertainty in your Prophet model:
- Uncertainty in the trend
- Uncertainty in the seasonality, holidays, and additional regressors
- Uncertainty...
