How to fit when there are level jumps in your data?

[blazespinnaker]

In the time series that I'm trying to fit on, there are a number of well known level jumps occurring. Eg, at the beginning of the new year, the underlying data is being changed to be based on new fundamentals so the december -> january isnt necessarily relevant.

Is there known approach for handling these with models such as ARIMA?

[AzulGarza]

hey @blazespinnaker! One approach would be to build an external regressor, let's say jump, equal to 1 when there is an expected or historical jump and zero otherwise. And then use AutoARIMA with that external regressor. Here's an example.