I am working on supervised machine learning models and I had a couple of doubts about features fractional differentiation. In particular, it is not clear what are the features to be fractional differentiate.
If I am working with OHLC financial time series, this is the procedure I'd follow:
- feature engineering using OHLC historical data (like moving averages, indicators, ....)
- fractional differentiate all the features
- feature scaling
Now, my question is: Does it make sense to fractional differentite all the features or it will be enough to make the OHLC series stationary using fractional differentation and then compute the indicators? Will the indicator be stationary in this case?
