Understanding Forecast Uncertainty Using Prediction Intervals and Projection Bands

--- Updated on July 14, 2020

Unlike mathematical models, statistical models are able to incorporate measurement errors (uncertainty) of the observations. Our models not only make predictions but also provide a prediction interval to give you a sense of the uncertainty surrounding the forecast. As you can see from the above figure, the predicted number of cumulative cases for the state of New York on July 14 is 412,684, and the 95% prediction interval is (412,566, 412,802). The observed count for New York on the same day is 412,889, which is slightly higher than the upper bound of the 95% predcition interval. Prediction intervals provide an upper and lower expectation for the real observation. These can be useful for assessing the range of actual possible outcomes for a prediction and a better understanding of the skill of the models.

We also provided a projection band to quantify the uncertainty of the long-term forecast path. From the above plot, you can see that the projection band is getting wider and wider as we try to make longer projection.