Evaluating Exponential Smoothing Models
You'll learn how to measure and compare the accuracy of different exponential smoothing forecasts using three key error metrics.
Why Evaluation Matters
After fitting Simple, Double, or Damped exponential smoothing models, you need objective ways to judge which performs best. Just like grading student essays with a rubric instead of gut feeling, we use standardized metrics to assess forecast quality. This helps you choose the optimal smoothing parameter α or decide between model types.
The Three Core Metrics
Mean Squared Error (MSE) averages the squared differences between actual and predicted values. Squaring penalizes large errors heavily, making MSE sensitive to outliers. Lower MSE means better fit.
Mean Absolute Error (MAE) averages the absolute differences without squaring. It treats all errors proportionally and is easier to interpret in original units. A MAE of 5 means your forecasts are off by 5 units on average.
Mean Absolute Percentage Error (MAPE) expresses errors as percentages of actual values, making it scale-independent. A MAPE of 10% means your forecasts are typically 10% off. This lets you compare forecasts across different datasets or units (dollars vs units sold).
Comparing Smoothing Approaches
Calculate all three metrics for each candidate model. A Simple Exponential Smoothing with α=0.2 might have MAE=12, while Holt's method yields MAE=8—the latter wins. Test multiple α values systematically, evaluate each, and select the configuration minimizing your chosen error metric. Often you'll examine all three metrics together since each highlights different aspects of accuracy.
Key Takeaway: MSE, MAE, and MAPE transform subjective "Does this look good?" questions into quantifiable comparisons, letting you objectively select the best exponential smoothing model and parameter values for your time series.