Forecasting Concepts and Horizons
You'll learn the fundamental vocabulary and structure of making predictions with time series models.
Now that you've built and diagnosed time series models, it's time to understand what you're actually producing when you forecast into the future.
Point Forecasts
A point forecast is your model's single best estimate for a future value. If you predict sales will be 1,000 units next month, that's a point forecast. Think of it as your model's "best guess"—but remember, the future is uncertain, so this number alone doesn't tell the whole story.
Prediction Intervals
Because forecasts are inherently uncertain, we pair point forecasts with prediction intervals—a range where we expect the true value to fall with a certain confidence level (often 80% or 95%). These intervals typically widen the further you forecast ahead, reflecting growing uncertainty.
One-Step vs Multi-Step Ahead
One-step ahead forecasting predicts just the next immediate time period (e.g., tomorrow's temperature). It's typically the most accurate because you're only projecting one period forward.
Multi-step ahead forecasting projects multiple periods into the future (e.g., the next 7 days or 12 months). You can either:
- Forecast iteratively (use each prediction to forecast the next)
- Forecast directly (train separate models for each horizon)
Forecast Horizon
The forecast horizon is simply how far into the future you're predicting. A 12-month horizon means forecasting 12 months ahead. Longer horizons generally mean lower accuracy and wider prediction intervals—like trying to predict weather: tomorrow is easier than next month.
Key Takeaway: Forecasting produces point estimates (best guesses) and prediction intervals (uncertainty ranges), with accuracy typically decreasing as you extend your forecast horizon from one-step to multi-step ahead predictions.