Future of Forecasting the Future in...
summary
Bloomberg aggregates crowd forecasts via Whisper and cleans them with a three-pass waterfall that uses mean and median absolute deviation plus regime-detection to separate sign, scaling and outlier errors. The pipeline flags sign and scaling errors for manual review while automating high-precision outlier fixes and reports 97% outlier precision, 100% scaling recall and ~95% sign recall. Normalized timing and direction scoring produce a weighted premium consensus that beats simple median and market-forward baselines.
AI-written summary · about
chapters (31)
- 0:00Introduction
- 1:02About Bloomberg
- 2:12What is Bloomberg
- 3:44Investing in AI and Machine Learning
- 6:16Bloomberg Analytics
- 7:06Overfitting
- 7:39Forecasting Methods
- 9:05Wisdom of the Crowd
- 9:29Whisper
- 10:00Wisdom of crowds
- 10:30Error checking
- 11:40Different types of errors
- 14:36Basic data science algorithm
- 16:10Division metric
- 18:01Ensemble
- 19:39Not always possible
- 21:08Truth is multifaceted
- 23:03Can we handle the truth
- 23:37Data points
- 25:14Regime detection
- 26:45Confusion matrix
- 28:35Rank the analyst
- 28:45Bloomberg
- 29:01Economic forecasts
- 29:40How to rank and score analysts
- 32:56Benefits of scoring
- 33:08Direction
- 33:56Timing
- 35:13Known unknowns
- 35:35Unknown unknowns
- 37:45Conclusion
related talks
28mDemand Forecasting: Accuracy, Stability, Automation
21mDon’t Call It 'The Forecast': Structured Prediction at Scale
40mMulti-Series Forecasting at Scale with StatsForecast
24mPlatform-First Forecasting: Engineering Scalable ML Systems
29mProbabilistic demand forecasting at scale
19mForecasting for Everyone - Foundation Models and the Future of Time Series Prediction
46mPanel about probabilistic forecasting
37mForecasting Time Series Data at scale with the TICK stack