Curriculum/pbo-probability-of-backtest-overfitting
PBO — Probability of Backtest Overfitting
research process·L2 · idiom·stub
Replacesthe belief that out-of-sample performance proves a strategy is real.
PBO (Bailey-Borwein-López de Prado-Zhu 2014) measures, across CPCV splits, the probability that a strategy ranked #1 in-sample finishes below median out-of-sample. Random rank-coherence under the null hypothesis (no real edge, just multi-test) gives PBO ≈ 0.5; a real edge gives PBO < 0.2. The single most damning number you can put on a backtest.
Prerequisites
Unlocks—
Bridges
- bailey-lopez-de-prado-evaluation-protocolmodel to implementationPBO + DSR + walk-forward live performance comparison is the canonical Bailey-Lopez de Prado evaluation protocol. Apply all three before any strategy goes live; refuse to take results seriously until they do.
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/pbo-probability-of-backtest-overfitting/card.ts