A Framework for Time-Updating Probabilistic Forecasts
10 hours ago
- #probabilistic forecasts
- #Kelly betting
- #Bayesian inference
- Proposes a new method for evaluating probabilistic forecasts that change over time.
- Models are treated as Kelly bettors in an iterative betting contest.
- Bankroll growth or decline serves as the evaluation metric.
- Allows real-time updates of market consensus probabilities and model credibilities.
- More accurate than traditional methods like average log-loss and Brier score.
- Shows a direct mathematical and conceptual analogue to Bayesian inference.