My forecast is better than yours! What does that even mean?
summary
Shows that probabilistic forecasts and their point estimates drive evaluation outcomes, not high-level topics. BlueYonder implements concrete fixes and compares RMSE, MAPE and RMAE against Poisson and Poisson vs Negative Binomial assumptions. Demonstrations include a toy case where aggregation changes apparent error from 100% to 50% and a statistical limit of 0.17 RMAE for low-volume items. Practical takeaways: choose the metric tied to the business loss, pick the matching point estimator, and group by current‑available features.
AI-written summary · about
chapters (12)
- 0:00Intro
- 1:28What is traditional forecasting
- 6:30What is probabilistic forecasting
- 8:47Matrix selection
- 10:55Examples
- 14:02Scaling problem
- 16:22Conclusions
- 17:27Limits
- 19:04Ultimate performance
- 19:43Grouping your data
- 21:28Constant bias
- 22:24Conclusion
related talks
30mKnowing what you don’t know matters- Uncertainty-aware model rating
24mThe Forecast Whisperer: Secrets of Model Tuning Revealed
39mForecasting With Classical and Machine Learning Methods
33mBacktesting and error metrics for modern time series forecasting
29mProbabilistic demand forecasting at scale
29mEvaluation Design
12mOn the Use of ML for Blackbox System Performance Prediction
59mPredictability Check: Demand Forecasting