Using Language Models to Iteratively Develop Quantitative Factors
Summary
This project report describes using a model with the BigQuant dai function’s syntax and formulas as context, then prompting it to build increasingly complex quantitative factors. The author says ordinary prompts and iterative adjustment moved the generated factors from very poor returns toward near-positive returns. Four factors were produced, with complexity increasing through the process.
The work was first tried in Gemini and then in a local VS Code workflow using Gemini, which the author found slow to respond. The report suggests a future experiment: have AI interpret a research report to propose an initial factor, then refine it iteratively. It gives no factor definitions, dataset, evaluation period, benchmark, or risk-adjusted performance, so the reported return improvement is anecdotal and does not establish out-of-sample predictive value. The document also includes software-service promotion that is separate from its research workflow.
Key ideas
- Providing a model with a quantitative platform’s function syntax can guide it to construct factors.
- Iterative prompts can make generated factors more complex and adjust their reported performance.
- The author produced four factors but does not describe their formulas or evaluation setup.
- A proposed extension is to generate an initial factor from a research report and refine it.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.