- Stack
- Python · PyTorch · FinBERT · yfinance · edgartools
Problem
Combining prices with filings is easy to get subtly wrong. If a filing is joined to prices before it was actually public, the model learns from the future and its backtest means nothing.
Approach
- OHLCV market data combined with FinBERT-processed SEC filings to predict price movements.
- Training on NVIDIA GPUs with PyTorch.
Phase 1 (done)
- Data ingestion for market data and filings.
- Point-in-time-correct temporal alignment of market data with filings, with validation.
Status
In progress. The repository goes public shortly after this site launches.