Understanding additive versus multiplicative seasonality
In our Mauna Loa example in Chapter 2, Getting Started with Facebook Prophet, the yearly seasonality was constant at all values along the trend line. We added the values predicted by the seasonality curve to the values predicted by the trend curve to arrive at our forecast. There is an alternative mode of seasonality though, where we would multiply the trend curve by the seasonality. Take a look at this figure:
Figure 4.1 – Additive versus multiplicative seasonality
The upper curve demonstrates additive seasonality—the dashed lines that trace the bounds of the seasonality are parallel because the magnitude of seasonality does not change, only the trend does. In the lower curve though, these two dashed lines are not parallel. Where the trend is low, the spread caused by seasonality is low; but where the trend is high, the spread caused by seasonality is high. This can be modeled with multiplicative...
