•
Kelly, Pruitt & Su (2019) introduce Instrumented PCA (IPCA): firm characteristics instrument latent factor loadings. Plain-language intro. Not investment advice.
Takeaway. In IPCA, firm characteristics instrument latent factor loadings—so “characteristics are covariances,” not only cute stock sorts.
Characteristics Are Covariances: A Unified Model of Risk and Return (Bryan Kelly, Seth Pruitt, Yinan Su, 2019 / NBER w24540) is a cornerstone of modern empirical asset-pricing stacks.
Fit for learners: Instrumented PCA links firm characteristics to latent factors—core for modern factor Python stacks.
This Hub note is a plain-language intro. Not investment advice. Bridge from omitted factors toward deep factor models.
Classic PCA on returns ignores the rich characteristics panel (size, value, investment, …). Pure characteristic sorts ignore latent factor structure. You want one framework that uses both.
sklearn.decomposition.PCA on a return matrix; start by reading the identification story, then code small panels.Do not confuse IPCA with “dump all characteristics into a neural net and hope.”
IPCA vocabulary shows up in many 2020s quant research repos. Understanding instruments → loadings → factors prevents cargo-cult copying of someone else’s estimator.
Tiny evening exercise: build a tiny T×N toy return matrix and a matching characteristic; run ordinary PCA on returns and separately regress returns on the characteristic; write three bullets on what each method cannot see—learning only.
Landing: NBER w24540
Not investment advice. This Hub note is educational only. IPCA is an econometric framework—not a stock-picking app.
Single course CTA: SoloQuant Academy’s Introduction to Python Programming—matrices and loops first; then latent-factor papers stop feeling opaque.