Descriptions:
A quantitative finance practitioner with a math and economics degree from UCLA and over three years of investment banking experience at Raymond James shares a comprehensive, two-year perspective on using Claude to build and operate trading systems. The central framework distinguishes between deterministic systems — rule-based, backtestable algorithms like automated trading bots, walk-forward backtesting engines, and regime detection models — and non-deterministic uses, where Claude acts as an analyst rather than a code generator.
For deterministic work, the presenter argues Claude Code can handle nearly all development without a full engineering team, allowing a solo trader to prototype and test multiple strategies in hours. He walks through what separates a robust system from a fragile one: unambiguous entry/exit rules (e.g., buy when 12-minus-one-month momentum exceeds the 80th percentile in its sector), proper walk-forward validation, and position sizing logic. For non-deterministic analysis, he demonstrates feeding Claude earnings transcripts, 10-Ks, and options chains — using metrics like IV rank, skew, and delta — to generate second opinions and stress-test existing theses, while being explicit that these outputs cannot be backtested.
The video also covers clear failure modes: Claude cannot predict prices, misidentifies chart patterns from images, and generates plausible-sounding but generic trade ideas without grounding. Pricing is addressed directly — Claude Code’s monthly subscription covers deterministic builds, while non-deterministic API calls cost cents per query when embedded in live systems.
📺 Source: AI Pathways · Published May 17, 2026
🏷️ Format: Workflow Case Study
