by David Byrd, Eren Kurshan, Tucker Balch
Continue on SelfMadeQuant
Suggested foundations for text and language models in finance and risk measurement and stress. Specialist prerequisites vary by paper.
Understand how text is converted into numerical inputs.
Check whether labels measure the financial construct of interest.
Represent losses as a distribution rather than one forecast.
Distinguish a loss quantile from the average loss in the tail.
Test whether a text-derived signal survives an honest financial evaluation.
Measure agreement when constructing or checking labels.
Compare risk estimators and their assumptions.
Connect diversification, common risk factors, investor views, and the properties of a risk measure.