Ensemble · Pro
Ensemble mean
Average top-k Free point forecasts when several methods tie on MASE (Pro).
What it assumes
A simple combination can reduce variance of point forecasts.
When to use
When several methods have similar MASE.
When to avoid
One method is clearly better — averaging dilutes it.
Knobs
Pro combination layer.
How it works
A simple mean of several decent forecasts often reduces variance of the point path. It also dilutes a method that is clearly best. The engine uses this as a Pro combination layer on top of Free fits — not a mysterious neural blend. If one row dominates the leaderboard, do not average it away.
Sample forecast question
Eight Monday call counts, same as the SES page. What forward path does the engine’s ensemble-mean method write for the next three Mondays?
Step-by-step on these numbers
| Period | Actual | Fitted |
|---|---|---|
| M1 | 19 | 19 |
| M2 | 21 | 19 |
| M3 | 18 | 20.50 |
| M4 | 20 | 18.67 |
| M5 | 22 | 19.83 |
| M6 | 19 | 21 |
| M7 | 20 | 19.67 |
| M8 | 21 | 20.17 |
| Period | Forecast | 95% interval |
|---|---|---|
| M9 | 21 | 17.61 – 24.39 |
| M10 | 21 | 16.21 – 25.79 |
| M11 | 21 | 15.13 – 26.87 |
Parameters the engine found
last= 21
The question
Eight Monday call counts, same as the SES page. What forward path does the engine’s ensemble-mean method write for the next three Mondays? Sample mondays: M1=19, M2=21, M3=18, M4=20, M5=22, M6=19, M7=20, M8=21 (calls).
What the product intends
On a real Pro run, ensemble mean is a combination layer: average nearby Free point forecasts when several MASE scores tie. It is not a neural blend. If one method dominates, do not average it away.
What this engine does on the canned series
The current module’s in-sample fitted value blends the previous observation with a short mean (last 3). Example at the end: last y = 21, recent mean of 19, 20, 21 = 20. The forward path holds y_n = 21.
Forecast
M9: 21; M10: 21; M11: 21.
Interval
Residual σ = 1.730. For h = 1 the 95% band is [17.61, 24.39] around 21. The engine labels this a residual-Gaussian heuristic (σ√h), not a simulation interval.
Graph of this sample
The chart plots the canned table on this page (actuals, in-sample fitted, forecast, 95% interval). Not your Excel series. Free analysis never uploads raw data.
Educational only. Not investment, weather, or operational advice. In the add-in, rank is rolling-origin MASE — “best supported among candidates on this series.”