How to prevent information leakage in quant ML?
Quant ResearchLevel: Intermediate
How to prevent information leakage in quant ML?
Hints
- When does each feature become available at prediction time?
- Check the split, fitting of preprocessing and reuse of the test set.
Answer
Attach availability timestamps to prices, corporate reports and revisions; join data as it was known at prediction time. Keep preprocessing, feature selection and tuning within training folds, and protect a final chronological test. Overlapping outcome windows require purging or appropriate temporal separation. A feature using only t−20 through t is not inherently leakage when predicting t through t+5; leakage depends on availability, alignment and overlapping train/test labels. Audit joins, execution timing and revision handling explicitly.
Source and licence: Quant Research Interview Notes — SoYuCry; English translation by photon2563 · MIT · answer: Quantoros editorial