Additive vs Multiplicative mode to choose

[algomaschine]

Gents, I was reading a few articles in the internet on how to choose the best seasonality mode, but the answer is still not too clear to me.

For example in this article the author is testing an error via a few metrics with additive & multiplicative and then selects multiplicative. https://exploratory.io/note/kanaugust/Time-Series-Forecasting-with-Prophet-Additive-vs-Multiplicative-aGB3gVk2gO But the difference is very small really, does it make any sense or tell us anything about the persistence of this setup in the future? I'm not sure.

Maybe anyone could enlighten me on the best test methodology to select the right seasonality mode when I have around 40k various data series, most of them financial? Thanks!

[ghost]

The question is should the seasonality factors be added or multiplied to the trend. If it grows with trend and the rate increases, use multiplicative. Typically, I have seen most seasonalities be additive in my work. The documentation has good details and examples on this.