Created by Dr. Pedram Jahangiry | Enhanced with Claude
Interactive Teaching Tool for Understanding Level-Only Exponential Smoothing
This tool walks through simple exponential smoothing one step at a time on monthly airline-passenger data, where each forecast is nothing more than a weighted blend of the last observation and the prior level. Drag the smoothing parameter or the starting level to watch the fitted line, the level path, and the training and test error recompute in real time.
The forecast for any future period is simply the current level
Level at time t is a weighted average of current observation and previous level
It asks one question: which value of α make the one-step-ahead forecasts fit the training months best? For every candidate it runs the same recursion you watch in the animation across the 115 training months, adds up the squared errors between each forecast ŷt and the actual yt, and keeps the candidate with the smallest total (the SSE). That is exactly the criterion statsmodels and R's ets() use when errors are additive, so the answer you get here is the answer you would get in Python or R.
statsmodels: 0 ≤ α ≤ 1. With α = 1 the level is simply the last observation and the forecast becomes the naive "tomorrow = today" rule; with α = 0 the level never moves. Results are snapped to the sliders' 0.01 grid.What to look for. On the airline data the best training fit is α = 1. That is the naive forecast, and it is a sign that a level-only model is missing the trend and the seasonality rather than a recommendation to use it. Compare its test RMSE with a hand-picked α such as 0.3, then move on to Holt and Holt-Winters, where the optimizer has something better to find.
Smooth & Stable: Creates very smooth forecasts that change slowly. The model gives more weight to historical data, making it resistant to noise but slow to adapt to genuine changes.
Responsive & Reactive: Quickly adapts to recent changes in the data. The model prioritizes recent observations, making forecasts more volatile but responsive to new patterns.
When optimal α is very high (>0.95), the model is essentially saying "I can only trust the most recent observation." This typically indicates that Simple Exponential Smoothing is inadequate for your data, likely because the data contains trend or seasonal patterns that SES cannot capture. Time to consider Holt's or Holt-Winters methods!
The α parameter controls the exponential decay of historical weights. With α = 0.3, the current observation gets 30% weight, the previous gets 21% (0.7 × 0.3), the one before gets 14.7% (0.7² × 0.3), and so on. This creates the characteristic "exponential smoothing" effect where influence decreases exponentially with age.