Flirting with Models Chris Carrano – Designing Practical Factor Models (S7E20)
76 snips
Sep 2, 2025 In this conversation, Chris Carrano, Vice President of Strategic Research at Venn by Two Sigma, shares his extensive experience in factor modeling and risk assessment. He explains the philosophy behind using 18 orthogonalized factors and the blend of Lasso and OLS to prevent overfitting. The discussion covers the challenges of analyzing private markets, the importance of interpretability in models, and actionable insights from factor results. Carrano even touches on personal interests, like his idea for a Pokémon card index as a market benchmark.
AI Snips
Chapters
Transcript
Episode notes
Asset Classes Hide Common Risks
- Asset-class labels can hide common underlying risks like interest rates that span bonds, real estate, and equities.
- A factor lens reveals shared exposures and enables true total-portfolio analysis across sleeves.
Less Is More For Factor Models
- Parsimony reduces overfitting, increases interpretability, and keeps models resilient out of sample.
- Venn builds 18 'greatest hits' factors focused on explaining maximum portfolio risk rather than chasing unique alpha.
Select Then Estimate With Lasso + OLS
- Use a lasso to select stable factors first, then run OLS on the survivors to estimate betas and t-stats.
- Choose the lasso penalty with corrected AIC to balance fit and model complexity, especially for small samples.