by Baptiste Lefort, Beatrice Guez, David Saltiel, Eric Benhamou, Jean-Jacques Ohana, Thomas Jacquot
Continue on SelfMadeQuant
Suggested foundations for text and language models in finance and trading research and honest backtests. Specialist prerequisites vary by paper.
Understand how text is converted into numerical inputs.
Check whether labels measure the financial construct of interest.
Use only information available at the simulated decision time.
Separate model selection from the final evaluation.
Test whether a text-derived signal survives an honest financial evaluation.
Measure agreement when constructing or checking labels.
Reconcile signals, executions, positions and P&L.
Choose a reproducible strategy, specify the trading rule, and test what remains after costs and estimation uncertainty.