Claude Code Trading Bot: Full Step-by-Step Tutorial

Summary

The presenter demonstrates a process for building an automated Nasdaq futures trading system by describing instructions to Claude in VS Code rather than writing code directly. The proposed bot would test strategies, monitor the market, send Discord alerts, and—after testing and safeguards—place trades automatically.

A central theme is that a promising backtest is not enough. The presenter says many strategies fail when realistic trading costs, unseen market periods, randomized-entry comparisons, and live paper testing are introduced. The video is educational, not financial advice, and emphasizes that futures trading is risky.

Main Ideas and Methodology

1. Set Up the Tools

• Install VS Code and Claude Code.
• Use plain-language prompts to have Claude create and run the code.
• The presenter says coding experience is not required, though Claude may request permissions while working.

2. Get Historical Market Data

• Obtain one-minute Nasdaq futures data for backtesting.
• The presenter uses Databento and says 16 years of data cost about $40; fewer years may cost less.
• He also mentions researching free CFD Nasdaq data, but does not present it as equivalent or recommend it outright.
• Protect API credentials: do not paste keys directly into VS Code. Store them securely and provide access carefully.

3. Describe the Trading Knowledge and Rules

Explain the strategy, trading hours, instruments, risk limits, and other relevant rules to Claude. Include trading-journal entries and rules such as whether to avoid CPI, FOMC, or other major news days, and limits on trades per day.

The presenter recommends speaking or dictating the information at length. He mentions Whisper Flow as a transcription tool.

4. Build an Honest Backtester

Ask Claude to test strategies using historical one-minute data. Require the backtester to:

• Include commissions and slippage on every trade.
• Fill orders only at prices the market actually traded.
• Avoid using future information.
• Investigate unusually good results for possible bugs.

The presenter says an early backtest overstated profits because it omitted costs; he estimates that 74% of those profits were fake.

5. Develop and Refine Strategies Cautiously

Test existing strategies or ask Claude to examine published, well-known day-trading strategies. Try filters such as volatility conditions, breakout rules, or moving-average direction, but beware of overfitting: tweaking a strategy until it fits past data may make it fail in the future.

6. Use Out-of-Sample Testing

The presenter suggests developing strategies on 2010–2020 data, then testing them on 2021–2026 data that was not used during development. Compare results from both periods. A strategy that only works on the data used to build it may have merely memorized past conditions.

7. Apply the Coin-Flip Test

Compare the strategy against 500 randomized-entry versions with the same stop-loss and number of trades. Ask what proportion of those random versions performed better.

The presenter’s rule is that a strategy that cannot beat this comparison should be discarded.

8. Keep a Permanent Research Log

Record each idea, its rules, the date tested, and every version. Do not edit or delete past entries. Track how many variations have been tested, so repeated experimentation does not create a misleading impression that a result is robust.

9. Combine Strategies and Paper Trade

Combine strategies that pass the tests into a portfolio, since they may perform differently on different days. The presenter describes a personal portfolio of roughly five to seven strategies, named Ace, King, Queen, Jack, and Ten.

Build a “forward tester” that watches the market on new trading days, records hypothetical entries and exits, and sends updates to Discord without placing real orders. Review the results over time.

10. Test Position Size Against the Firm’s Rules

Simulate the portfolio under the exact prop-firm evaluation rules across many historical starting dates and different position sizes. Include randomized-entry comparisons at each size.

The presenter recounts losing two accounts while trading 10 micros, then reducing to five micros and passing two evaluations two days later. This is his personal account, not a guarantee of results.

11. Automate Only After Testing and Add Safeguards

Connect the trading system to the prop-firm account through an API only after prior testing. Start in test mode and require explicit approval before live orders.

• Attach stop-losses—and, where applicable, take-profits—to orders.
• Schedule the bot’s trading hours.
• Set a daily loss limit.
• Create a desktop kill switch.
• Send Discord alerts for entries, exits, and problems.

The Presenter’s Five Rules

1. Use real data and a backtester that accounts for costs and does not cheat.
2. Test on years the strategy has not seen.
3. Require the strategy to beat a coin-flip comparison.
4. Forward-test with fake money before risking real money.
5. Size small and let the bot execute only after the system has been tested.

Speakers and Sources Featured

• Speaker: The video’s presenter, associated with the Steven Ace Trades channel. He describes his own testing, trading, and prop-firm experiences.
• Tools and services mentioned: Claude / Claude Code, VS Code, Databento, Whisper Flow, Discord, and an unnamed prop-firm account and API.