Glossary term
Train/Test Split
Separating history used to build a model from history used to evaluate it.
A train/test split helps check whether a model generalizes. The training period is used to choose parameters or fit the model; the test period is kept separate and used only to evaluate performance. In financial data, the split usually respects time order because the future cannot be used to train the past.
Example: A researcher might fit a signal on 2010-2018 data, tune it on 2019-2021 data, and evaluate it on 2022-2026 data.