Investment Strategies Ideation Using Large-Language Models and Structured Multi-Modal Data

“How do you come up with your strategies? Where do you find your ideas?” We’ve asked many professional investment managers, but most don’t really know. They read voraciously, watch the market, and talk to peers. They’re “plugged in.” They’re immersed. Oftentimes, this results in spurious ideas twisted by behavioral biases that are more akin to numerology than science. Other times, the ideas contain brilliant insights.

How do we go about finding strategies to test and implement without decades of hard-earned experience? This is not easy to do, and most graduate students find naive and impractical papers.

There are too many papers to read, and too few are useful. Most people, even professionals, need more experience, domain knowledge, and intuition to sort them out. A system to identify worthwhile research would be a great boon.

By converting social media and academic research into a graph vector database, enhancing it with machine learning, and then querying with prompts researched by dozens of graduate students over several semesters, we create a pipeline for ideas that can then be backtested and stress tested.