Stationary data , performing fractional diff on data

AI & Machine learning
1

Hello 



Just looking at the topic of fractional diff to create stationary data.  Does this process have to be performed on OHLC data prior to creating features? Or can you create features for ML and then apply it? Whats the best practise and why? cheers anyone

2

Hi Stephen,



If you apply feature engineering techniques on non-stationary series, the resulting series will have a higher value of 'd'. This means that if you apply the fractional differencing method before feature engineering techniques, the resulting series will have more memory. Therefore, it is advisable that you can prioritise this step of fractional differencing.



I hope this helps.



Thanks!

Satyapriya

3

Ok so your saying that if you apply the frac diff before creating features then after creating those features it will retain more memory than creating features and then applying frac diff? Is that correct  . Therefore we should apply it to the OHLC candle data and then use this data to create features. 



How would this affect indicators if at all? 

4

Yes, this would have an impact on the indicators as the indicator would be calculated on a fractionally differentiated series. However, they would be stationary in nature.