ai trading / public portfolio challenge / monte carlo / options trading / backtesting
Monte Carlo for AI Trading: Stress-Testing My $25K Portfolio
Austin Starks revisits his real $25,000 Public Portfolio Challenge after the account climbed to roughly $32,000, fell to about $26,500 over six weeks, and recovered some by the October 5, 2026 recording. Claude Code helped revise the eligible stocks: Exxon became Bloom Energy, ConocoPhillips became Nvidia, and Okta became Moderna. The foundation remains momentum-driven, long-dated calls, with weekly checks and exits when momentum breaks.
A matched historical year returned about 13% under the old rules and 65% under the updated rules, with about 43% median deployment. The longer daily test favored the old rules. The video explains how 10,000 paired Monte Carlo paths resample historical returns in ten-session blocks to stress-test the comparison. Those paths reuse previously inspected history; they are not a future win probability or a new out-of-sample test. Profit-taking experiments were also completed, with spreads and long calls evaluated separately. Account observations and simulated research are distinct. Follow the linked live account to see what happens next.
Transcript
0:00I gave an AI $25,000 to trade in my real Public account. In under four months of trading, it reached roughly $32,000. Six weeks later, it was near $26,500. It has recovered some, but that giveback made me ask Claude Code to help build a better strategy.
0:17The AI designs the rules, and NexusTrade executes them. The idea was to trade bullish options on my watchlist when momentum was strong. We kept that foundation, but updated the stocks to better match my theses: data-center power, AI compute, and biotech. Exxon became Bloom Energy, ConocoPhillips became Nvidia, and Okta became Moderna. The strategy buys long-dated calls, checks weekly, and exits when momentum breaks. I still want aggressive upside with less cash tied up.
0:44I tested both rule sets with the same account balance, over the same historical year. The old rules returned 13%. The updated rules returned 65%, with median deployment around 43%. The longer daily test favored the old rules. So I also checked the one year results with Monte Carlo. Then I resampled those returns into 10,000 paths, keeping ten day chunks together.
1:09The paths are built from historical returns, so I use them to stress-test the strategy. We also tested taking profits against a spread's maximum possible profit, with long calls evaluated separately. I want aggressive gains I can actually keep. My real money is the scoreboard. Follow the live account on NexusTrade.
Join the conversation
Loading conversation…