Forecasting with Simple Exponential Smoothing
You'll learn how to generate forecasts using SES and understand their inherent limitations.
Making Predictions with SES
Once you've fitted a Simple Exponential Smoothing model with your chosen smoothing parameter α, you can generate forecasts. SES produces a flat forecast — meaning all future predictions are identical, equal to the most recent smoothed level.
One-Step-Ahead Forecasts
For the next time point (t+1), your forecast is simply the current level estimate. If your latest smoothed value is 250 units, your forecast for tomorrow is 250 units. This works well when data has no trend or seasonality — just random fluctuations around a changing level.
Multi-Step-Ahead Forecasts
Here's where SES shows its limitation: the forecast for t+2, t+3, and beyond are all the same as your one-step-ahead forecast. SES essentially says, "Based on recent history, my best guess for the future is that things will stay at the current level."
Think of it like estimating tomorrow's temperature based only on today's temperature, ignoring seasonal patterns or warming trends. If it's 72°F today, SES predicts 72°F forever forward.
Understanding the Limitations
SES is not appropriate when your data shows:
- Clear upward or downward trends (you'd need Holt's method)
- Seasonal patterns (you'd need Holt-Winters)
- Strong cyclical behavior
SES works best for data that fluctuates around a slowly changing level without systematic patterns — like daily inventory levels with irregular demand.
Key Takeaway: SES generates flat forecasts where all future values equal the latest smoothed level, making it suitable only for data without trend or seasonality; multi-step forecasts provide no additional information beyond the one-step forecast.