claude code / ai trading / algorithmic trading / backtesting / developer tools
Claude Code Trading Strategy: 1,100% in 11 Years, and a 59% Drawdown
I built a trading strategy by talking to Claude Code in a terminal, and this video is the whole method in four steps: open a terminal, run one command, ask for a strategy, and deploy it to paper trading. The command is `claude mcp add --transport http nexustrade https://nexustrade.io/api/mcp`, and it hands the model real prices and real backtests instead of a guess at market history.
The part worth watching is what got thrown away. The model searched my backtest corpus first, 222,010 prior runs, to see which indicators have actually worked rather than which ones sound convincing. It tried a momentum rule, and that rule made money through the bull run and then lost about a quarter of the account in the 2021 to 2023 bear market, so it was rejected. What survived is a single regime rule: buy TQQQ when QQQ's last price is above its 200-day simple moving average and the position is under 95% of the book, and sell the whole position when QQQ closes back below that average.
Backtested from January 2015 to September 2026 with fees on, $10,000 becomes $119,942, a return of 1,100.59% across 98 closed trades, against $45,086 for buying and holding the S&P 500 over the same window. Buying and holding TQQQ itself returned 3,542.90%, and the strategy does not beat that. TQQQ also wins on Sharpe, 0.803 against 0.694, and on Sortino, 1.115 against 0.927. What the strategy wins on is every drawdown measure in the table: a 59.27% worst drawdown against TQQQ's 81.86%, a 20.05% average drawdown against 21.95%, and an ulcer index of 24.80 against 31.15. That is the trade, stated as numbers.
The strategy was deployed to paper trading on 8 September 2026 and is running on live prices with no money at risk. It is down about 4% after two days, which on a 3x leveraged fund is noise rather than evidence. Everything above is a backtest, not a live track record. The portfolio is public, including the notepad with the full ten-step derivation and the backtests that failed.
Transcript
0:00I built a trading strategy that returned almost three times what the S&P 500 did, and I did it by talking to Claude Code in a terminal.
0:07Here is exactly how I did it, in four steps.
0:10Step one.
0:11Open a terminal.
0:12That is genuinely the hardest part of this, and it is not hard.
0:16Step two.
0:17Run this one command.
0:18It connects Claude Code to NexusTrade, so the model can pull real market data and run real backtests instead of guessing at them.
0:25Step three.
0:27Type claude, and ask it for a strategy.
0:29I told it to find me something profitable and to prove it with backtests rather than tell me about it.
0:34It searched my backtest corpus first to see which indicators have actually worked.
0:38It tried a momentum rule, and that one made money in the bull run and then lost a quarter of the account in the bear market, therefore it got thrown out.
0:46What survived was a single rule on a moving average.
0:49Step four.
0:50Deploy it to paper trading and let it run on live prices with no money at risk.
0:54Over eleven and a half years that rule turned $10,000 into $119,942, on 98 trades.
1:01That is two and a half times what the index did.
1:03And against holding the leveraged fund straight, it cut the worst drawdown from 82% down to 59%.
1:10The whole thing, including every backtest I threw away, is public on the link.
1:15You do not have to code, learn Python, or be a quant to deploy a strategy that beats the market.
1:21Follow for more.
Join the conversation
Loading conversation…