Generating Forecasts with SARIMA
You'll learn: How to produce forecasts from a fitted SARIMA model, including both point predictions and uncertainty intervals that widen over time.
From Fitted Model to Future Values
Once you've fitted a SARIMA model and validated it through diagnostics, you're ready to generate forecasts. A SARIMA model doesn't just give you a single "best guess" for future values—it provides point forecasts (the most likely value) and prediction intervals (a range showing where the actual value will likely fall).
Think of it like predicting tomorrow's temperature: a meteorologist might say "72°F" (point forecast) but also "between 68°F and 76°F with 95% confidence" (prediction interval). The same principle applies here.
Understanding Forecast Uncertainty
The further you forecast into the future, the less certain you become. This is natural—short-term predictions rely heavily on recent observed patterns, while long-term forecasts depend more on the model's understanding of underlying trends and seasonality.
Prediction intervals grow wider as the forecast horizon extends. A 1-step-ahead forecast might have a narrow interval (high confidence), but a 12-step-ahead forecast will have a much wider interval. This reflects accumulating uncertainty: each step's prediction error can compound with the next.
The Forecast Output
When you generate SARIMA forecasts, you typically receive:
- Point forecasts: The expected values for each future time step
- Confidence intervals (often 95%): Upper and lower bounds around each point forecast
- These intervals account for both parameter estimation uncertainty and the random component in your time series
Key Takeaway: SARIMA forecasts provide both point estimates and prediction intervals that naturally widen over time, reflecting increasing uncertainty about the future—the model quantifies not just what might happen, but how confident it is about each prediction.