Model A says 54% and Model B says 60% for the same game. Which one do you bet on, or do you use neither? Ensemble forecasting is the practice of blending forecasts with weights you set before the event, then testing the blend on data neither model has seen. You'll work the example where 54% and 60% become 55.8%. The bigger lesson is when blending fails: models trained on the same data behave like one model, and an overconfident component drags the whole mix down. You'll also separate disagreement between models from real uncertainty, then let that uncertainty shape how much you put at stake on a play.
Learning objectives
- Explain why an ensemble helps only when its models are genuinely different and calibrated.
- Blend two forecasts with set weights (54% and 60% become 55.8%) and test the blend on a holdout.
- Separate model disagreement from real uncertainty, and size exposure with that uncertainty in mind.
