I’ve loaded up the AlgoSeek sample and run the Intraday Features notebook from Chapter 12 (machine-learning-for-trading/12_gradient_boosting_machines/10_intraday_features.ipynb at main · stefan-jansen/machine-learning-for-trading · GitHub). I get the same results for the IC as the notebook in GitHub does, which is that the only positive IC is +1% for RDOWN while all the rest are negative.
In the book (page 404) the chart starts off with nearly +17 for RET1MIN and the next 11 are all positive (most over 3).
There is obviously some issue with the scaling in both cases because IC is usually in the range +1 to -1, but even just looking at the relative values these seem totally opposite (from very promising to seemingly marginal at best).
Anyone else looked into this? I’d like to come up with useful signals for an intraday strategy in this vein.
Thanks,
Jim