Curriculum/triple-barrier-labelling
Triple-Barrier Labelling
research process·L1 · combinator·stub
Replacesthe belief that 'return after N days' is the obvious labelling.
Triple-barrier labelling (López de Prado, AFML ch. 3) assigns a discrete label (+1 / 0 / −1) based on whichever of three barriers is hit first: profit-take, stop-loss, or time-decay. The label is asymmetric and path-dependent — the same return endpoint with different intra-period paths gets different labels. The labelling that makes ML on financial data answer the right question.
Prerequisites
Unlocks—
Bridges
- meta-labellingmodel to implementationMeta-labelling (AFML ch. 3) layers a second classifier on top of triple-barrier labels: the first model decides direction, the second decides whether to take the trade. The split halves the data size each model sees but improves precision-recall trade-offs.
- label-quality-vs-feature-qualityshared measurementML-on-finance is bottlenecked by label quality (path-dependence, regime-conditioning) more often than feature quality. Triple-barrier is the canonical move that improves labels without touching features.
This concept is a node in the curriculum DAG. The full lab — page blocks, done state, references — has not been authored yet. The relations above describe where it sits in the graph.
Author at: content/concepts/triple-barrier-labelling/card.ts