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

Corey Hoffstein and Ben Wellington discuss the intricacies of feature generation in quantitative investment strategies. They explore the definition of features, their role in the quant process, and how Two Sigma's shared platform model supports feature ideation. The conversation delves into managing feature collinearity, the impact of forecast horizon, and the life cycle of features in relation to market efficiency. They highlight the creative thinking necessary in quant research teams and the use of AI tools for rapid prototyping. The episode also covers the risks of AI, trade-offs in infrastructure, and the importance of addressing p-hacking. Ben shares insights into future skills for feature engineers and his personal interests in public data.

Speakers

AI Chat

Corey Hoffstein

Host • 27%

Ben Wellington

Guest • 72%

Unidentified

Chapters

0:00	Introduction, disclaimer, and guest introduction
1:20	Feature generation, AI, and Ben Wellington's background
5:23	Definition and examples of features
8:53	Two Sigma's shared platform model
11:32	Where alpha lives in the quant process
16:29	Ideating new features
20:02	Managing feature collinearity and orthogonality
22:24	Forecast horizon relevance in feature selection
23:22	Life cycle of features and market efficiency
24:51	Creative thinking in quant research teams
27:03	NLP to LLMs: Turning numbers into language
30:03	Rapid prototyping with AI tools
33:33	Risks of AI lowering output entropy
37:15	Balancing difficulty and technological advancement
39:26	Trade-offs in AI infrastructure choices
41:38	Addressing p-hacking and overfitting risks
44:28	Feature engineering and data updates
48:10	Idiosyncratic scale and AI
50:40	Skills for future feature engineers
53:45	Personal obsessions and public data

Transcript

Clip

Loading transcript...