Honest Factor Research

The Problem

Factor models in quantitative finance look great on paper. Most are inflated by hidden pathologies: sector mirrors, proxy contamination, regime shifts, and overconfident statistics.

The Solution

Built a framework that systematically audits 4 pathologies across 2,944 US stocks, 28 factors, and 20 years of data. Uses RidgeCV regression, Politis-Romano block bootstrap, and VIX-stratified regime analysis.

Technical Highlights

• Politis-Romano block bootstrap implemented from scratch in NumPy
• Gram-Schmidt residualization to isolate proxy effects between factors
• VIX-stratified regime betas — factor stability across crash/normal/boom markets
• YAML-driven factor catalog — 28 factors configurable without code changes
• Proper Python packaging (pyproject.toml), installable as library
• GitHub Actions CI matrix (Python 3.11 + 3.12), ruff, mypy, pytest

Tech Stack

Key Numbers