Automating Alpha: How Large Language Models Are Quietly Rewriting Quantitative Investing

From Human Alpha Factories to Machine Strategy Discovery

Strategies are born from human intuition, refined through backtests, and often decay once deployed. Even sophisticated machine learning pipelines still depend heavily on manually designed factors and static modeling assumptions. The study (https://www.quantpaper.com/paper/b6cfec3b-2dc3-4e4b-a174-dac02e719f8e) behind this article proposes a fundamentally different idea: what if alpha discovery itself could be automated, adaptive, and self-updating, without constant human intervention?

The framework introduced in this research reframes the role of large language models. Instead of treating LLMs as forecasting tools, it positions them as strategy engineers. The system does not ask an LLM to predict prices. It asks the model to read financial research, extract executable alpha formulas, organize them into economically meaningful categories, and continuously refine them through market feedback. This distinction matters. Prediction is brittle. Strategy construction, when done right, compounds.