What Is Time-Series Forecasting?

Time-Series Forecasting Explainer

Time-series forecasting estimates future values from observations ordered through time. Useful models may represent trend, recurring seasonal patterns, lagged relationships, external drivers, changing variance, and irregular shocks.

Unlike ordinary random splitting, evaluation normally preserves time order. Rolling or expanding backtests train on the past and evaluate on a later window, more closely matching how forecasts will be produced in operation.

Statistical models, tree-based methods, neural networks, and hybrids can all forecast time series. The strongest method depends on data frequency, horizon, stability, available history, known future inputs, and the cost of different errors.

Forecasts should include uncertainty and a clear horizon. Structural breaks, revisions, missing periods, leakage from future information, and feedback from actions taken on earlier forecasts can undermine performance even when historical fit looks strong.

Time-series forecasting uses ordered historical observations to estimate future values while respecting temporal structure.

Key terms include time series, forecast horizon, trend, seasonality, lag, rolling backtest, prediction interval, structural break and leakage.

Backtesting trains on earlier periods and evaluates on later periods without allowing future observations into the fit.

Multiple forecast origins reveal whether performance is stable across changing conditions.

Point estimates alone hide the range of outcomes consistent with data and model assumptions.

Prediction intervals, scenario analysis, and monitoring help users understand how confidence changes with the horizon.

Explore more "Explainers"

Discover additional explainers across politics, science, business, technology, and other fields. Each explainer breaks down a complex idea into clear, everyday language—helping you better understand how major concepts, systems, and debates shape the world around us.