Hi everyone (new here, great library!). If I de-seasonalize a series (e.g. with MSTL decomposition) and then want to convert forecasts back to seasonalized data, how should I get the components in the future? Forecast the components as well? Or should I stick to something like monthly dummies as features?

Hey. I believe you can use the mstl_decomposition function (example) but instead of using the decomposition as features you add them back after the forecast step

I thought each component related to a period in the past. I don't know how to get components in the future

That function uses the MSTL model to compute the decomposition for the training set and then forecasts it

A follow up on this section, how is mstl_decomposition actually forecasting the components? Does it using ARIMA? Thanks.

It uses the trend forecaster to forecast the trend and projects the last seasonality