Year-long experiment pits Claude against the S&P 500 with a $1,000 paper portfolio
OriginalScrubLord · reddit · 2026-09-30
A Redditor launched a 365-trading-day experiment: Claude (Sonnet 5.5 on a standard Pro plan) manages a $1,000 paper portfolio benchmarked against $1,000 of SPY bought and held.
Setup
- Every weekday before market open, Claude receives its portfolio, a one-line price screen of all 503 S&P 500 stocks, and its last 10 days of decisions. Its only tools: web search and web fetch.
- It reads overnight news and places up to 5 orders, filled at the opening price it cannot see when deciding (no lookahead).
Rules enforced in code: S&P 500 individual stocks only; no ETFs, options, leverage or shorting; max 5 trades/day; 30% max per stock; every buy must cite a source it read that morning; doing nothing is allowed.
Day 1: Claude $1,000.79 (+0.08%) vs SPY $996.57 (−0.34%). It opened four positions (NFLX, TMO, GOOGL, XOM) and held 56% cash, citing the 10-year Treasury yield hitting 5.23%, oil above $100, and 70% odds of an October Fed hike.
Weekly updates will include charts, trades, and Claude's reasoning; a full timestamped git log will be published at the end.
More from coding & agent
- Using AI plan mode to audit an SEO workflow and find where to plug in Jev — gaganghotra_ · 2026-09-30
- Dev's AI workflow grew so complex he asked the model to explain his own system — gaganghotra_ · 2026-09-30
- Herdr review: a TUI built for juggling multiple Codex and Claude Code sessions — leebase65 · 2026-09-30
- OpenAI exec shouts out Pidot at DevDay, signaling an open ecosystem play — omarsar0 · 2026-09-30
- AMD's Hyperloom and ROCm 10: AI agents tune GPU kernels overnight with accuracy checks — AnushElangovan · 2026-09-30
- A Prompting Trick: Make Coding Agents State Their Views on Good Software First — nptacek · 2026-09-30