Modeling shocks such as COVID-19 lockdowns

In mid-2020, forecasters the world over were at a loss for what to predict in the coming months and years. The COVID-19 pandemic utterly transformed life around the world and, with it, many time series. Online purchases skyrocketed beyond anything anyone had predicted at the beginning of 2020; consumption of media such as Netflix and YouTube dramatically increased, while in-person event attendance dramatically decreased.

As brilliant as Prophet can be when it comes to forecasting, it cannot simply predict the future. In the midst of the pandemic, Prophet would have struggled just as much as the forecasting experts at predicting when the pandemic would end and how time series would behave both during and after the lockdowns. However, we can model such shocks to the system after the fact in order to understand what effect they had. And just like the NatGeo promotion we modeled in the previous section, we can predict what would result from...