Good question, after adding low entropy features you can check if they actually add predictive power by comparing model performance or using permutation importance. In trading, even low entropy tends to stay relatively high because financial data is inherently noisy, so what really matters is whether it improves out-of-sample predictability.
PS: We usually drop high-entropy variables, keep medium-entropy ones as features, and use low-entropy ones as potential targets.
Thanks for sharing, very helpful! Just wondering how do I know if this feature actually works after adding some low entropy features?
Good question, after adding low entropy features you can check if they actually add predictive power by comparing model performance or using permutation importance. In trading, even low entropy tends to stay relatively high because financial data is inherently noisy, so what really matters is whether it improves out-of-sample predictability.
PS: We usually drop high-entropy variables, keep medium-entropy ones as features, and use low-entropy ones as potential targets.