Glossary
Out-of-sample
Data that was not used to choose a rule, its settings or its filters, and so can judge the rule fairly. It must come later in time than the data used to build the rule.
Out-of-sample data is kept apart from every step of research: choosing the rule, tuning it, and deciding which variants to drop. Results on it are the fair measure of the finished rule. In-sample results are expected to be better than out-of-sample ones.
Once out-of-sample data is used for a choice, it becomes in-sample. Short out-of-sample periods say little: the standard error of an annual Sharpe ratio is about 1 over the square root of the years of data.
Example: a rule has an in-sample Sharpe ratio of 1.5 and an out-of-sample ratio of 0.6, so 40% of the in-sample figure holds.
Sources
- Advances in Financial Machine Learning. Wiley (Marcos López de Prado), 2018. Accessed 9 Nov 2026