Real-World Forecasting with GPT-4o and MSCFT: 7 Resolved Cases
Over a 12-year forecasting career with more than 1,600 structured predictions logged, I’ve implemented the MSCFT Template (v3.1C – BIN Integrated) to align with GPT-4o’s capabilities in structured reasoning, information filtering, and scenario-based forecasting.
Results from 7 Resolved Forecasts:
- All 7 forecasts correct or in the correct probability bucket
- Average Brier Score: 0.0831
- Career Brier Score improved: 0.497 → 0.494
- Ranked 1st out of 66 in U.S. 10-Year Treasury yield forecast
Other domains include S&P 500, Bitcoin, nuclear risk, geopolitical leadership
MSCFT’s design includes:
- A strict structure for Bias, Information, and Noise analysis (BIN)
- Defined bucket logic or binary outcomes
- Optional clause modules for yield curve asymmetries and conditional weighting
These forecasts were produced using GPT-4o (ChatGPT Plus) with strict adherence to the MSCFT 3.1C forecasting template.
GitHub repository (public):
find the project at captbullett65 MSCFT
If you’re working on structured forecasting with LLMs, this may offer a stable, reproducible approach.