How to use it
- 1. Choose a patternStable, trending or seasonal demand, with a noise slider and a reseed button.
- 2. Tune the methodsSliders set the window and the smoothing constants; every forecast redraws instantly.
- 3. Read the raceAll five one-step-ahead forecasts overlay the actual series in their own colour.
- 4. Crown a winnerThe leaderboard ranks by MAPE; each method carries a one-line reason it wins or loses.
The arena
Demand pattern
Method parameters
Dots are actual demand; each line is a method's one-step-ahead forecast. Errors are scored only after enough history exists for every method (grey shaded warm-up is excluded).
Swipe the chart and the table sideways to see everything.
| Method | MAPE % | MAE | Bias | Why it wins / loses |
|---|
Try these experiments
- On the stable series, lower SES α and find the point where SES beats MA(3) - smoothing more helps when there is only noise.
- Switch to seasonal demand and watch Holt (no seasonality) develop a systematic lag while Holt-Winters tracks it.
- Set noise high on a stable series and see the naive forecast fall furthest behind - it copies each noisy value, while smoothing averages the noise away.
About this tool
The concept
Every forecast trades responsiveness against stability. The right method depends on whether the series has level, trend and season.
How it works
Each method makes a rolling one-step-ahead forecast; we score them out-of-sample with MAPE, MAE and bias on the same held horizon.
Winning strategy
Match the method to the pattern: SES for level only, Holt adds trend, Holt-Winters adds season. Naive only wins when demand behaves like a random walk.