I recently studied and implemented the intraday periodicity model of Andersen and Bollerslev from here.
https://www.sciencedirect.com/science/article/abs/pii/S0927539897000042
The seasonality captured by the model on an equity index looks fine on the 5 minute timescale.
Can anyone validate this result and explain the next step in the process for modelling returns and volatility.
Edit : So i fitted Arima models to both un-adjusted and adjusted returns (Rf = R/si) ,but couldn't see any significant improvements in r-squared.
Edit 2 : Have tried a bunch of things (prediction , backtesting ,..). The results do not improve for any model. Could someone point me to the right direction as to what i am doing wrong.
Edit 3: Just tried the weirdest idea i ever had : "Intraday Periodicity adjusted Bollinger Bands". The results weren't great else would have published a paper :).