Ben Wellington – Complex Feature Engineering at Two Sigma (S7E32)

Quantitative equity investing increasingly relies on creative feature generation to extract alpha from commoditized datasets. By transforming raw information into unique, economically meaningful signals, firms gain an edge that transcends simple data access. Large language models have accelerated this process, enabling researchers to rapidly prototype idiosyncratic features—such as analyzing CEO non-verbal cues—that were previously too resource-intensive to explore. However, the ease of AI-driven automation poses risks of signal correlation and reduced model diversity. To maintain a competitive advantage, systematic investors must treat AI as an amplification tool for human ingenuity, ensuring that models remain orthogonal and robust. Success in this evolving landscape requires a shift from traditional coding tasks toward high-level conceptualization, where the ability to ask novel questions and manage model life cycles becomes the primary driver of long-term performance.