Should a regime model output a hard state or a confidence score?

Question

Accepted Answer

Hard states are easier to operate and can work well when the edge truly switches on and off. Confidence scores are better when the environment changes by degree and the main use of the model is to scale risk or rank opportunity quality. Short answer: Choose a hard state when the downstream action is naturally binary and the economic story really is about tradable versus non-tradable conditions. Choose a confidence score when the regime mostly changes how aggressive the strategy should be rather than whether it should exist at all. When a hard state is cleaner: Binary states are attractive when the strategy has obvious failure zones. They are easier to explain, easier to monitor, and often easier to audit after something goes wrong. When a confidence score is better: Graded outputs make more sense when the regime shifts by degree. They preserve uncertainty and can feed downstream sizing or capital-allocation rules without forcing every small change into an on-off switch. Test the downstream policy, not just the classifier: A regime score with good classification metrics can still be a bad trading object if the final size rule is fragile. The real evaluation target is the end policy: how the output changes trades, turnover, drawdowns, and operating clarity.