Previous values of a time series used as inputs for forecasting. A lag-1 feature is the value from one period ago, lag-12 might be from twelve periods ago. In forex robots, lag features capture momentum or seasonality: e.g. including yesterday’s price or last month’s return as predictors. For instance, to predict Monday’s price, the model might include the price from the previous Friday (lag-1) and one week ago (lag-5). Lag features help models learn temporal dependencies; one chooses useful lags by examining correlations or domain knowledge.

Back to Robots Glossary Index