Everything I've Learned Trading With Claude In 18 Minutes

Summary

The presenter argues that Claude is most useful for building rule-based, testable trading systems and for supplementary research. It should not be trusted to predict prices or generate a core strategy on its own.

Deterministic and non-deterministic uses

• Deterministic systems use fixed rules: identical data should produce identical results. Claude Code can help build trading bots, scanners, backtests, and portfolio tools. Because their rules are explicit, these systems can be tested historically and run consistently.
• Non-deterministic use of Claude involves asking it to interpret information, such as reviewing a trade thesis, earnings reports, portfolios, or options chains. Responses can vary between runs, so the presenter says they cannot be reliably backtested and should not form the core of a strategy.
• The presenter recommends giving Claude relevant context—such as trading style, positions, thesis, time horizon, and risk tolerance—and using it primarily to challenge an investor’s thesis, surface bear cases, and flag risks.
• A combined approach uses quantitative screening first, then Claude as a qualitative reviewer. Claude’s score should be blended with quantitative results rather than followed blindly.

Methodologies and Examples

Strategy validation

The presenter recommends defining unambiguous rules, then using walk-forward validation, Monte Carlo simulations, sensitivity analysis, and transaction-cost modeling before risking capital. A promising simple backtest may still fail in live markets.

Example strategies

• Momentum rule: Buy when a stock’s 12-minus-1-month momentum score is above the 80th percentile within its sector.
• Cointegrated commodity pairs: Identify cointegrated pairs, monitor the spread, enter when its Z-score crosses a threshold, and exit when the spread reverts.

Three-layer system

1. Macro deployment gate: Score six signals from 0–100 and combine them into a deployment score. Signals include VIX level and term structure, S&P 500 market breadth, credit spreads, put-call sentiment, and factor crowding.
2. Stock scanner: Screen the S&P 500 using five quantitative factors and select candidates according to the macro gate.
3. Claude analyst: Provide candidates’ last four quarters of financial data and ask Claude to assess earnings quality, growth, balance-sheet health, margins, and red flags. Blend its score with the quantitative score.

In the example, the macro score implies full deployment above 70, reduced position sizing from 40 to 70, and no trades below 40. A score around 60 leads to sizing at approximately 60% of normal.

The sample macro readout showed VIX term structure at 99.9, credit spreads around 79, and factor crowding around 10. The presenter cited a −0.69 correlation between momentum and value factor returns as a sign of underlying risk.

The final ranking uses a 60/40 quantitative-to-Claude split. TXN is cited as a stock that could rise in rank after fundamental review. Ford scored poorly on both the quantitative model and Claude’s assessment, receiving 4/10 from Claude.

The presenter also says Claude can review options chains using implied-volatility rank, skew, delta, and upcoming catalysts, but advises treating this as a second opinion rather than a core strategy.

Key Cautions, Costs, and Disclosures

• The presenter says Claude cannot reliably predict prices and may give confident-sounding but random predictions.
• He advises against relying on Claude’s chart-image analysis for technical patterns or taking its trade ideas without an independent thesis.
• A system written with Claude may differ from one generated in another run. Once written, saved, and debugged, its rules can run deterministically.
• The presenter says deterministic systems generally require a Claude subscription to build, while using Claude’s API for analysis incurs usage-based costs—described as a few cents per run.
• He cites a two-year period of using Claude to build trading systems and suggests backtesting the deployment gate over the prior two years.
• Disclosure: The presenter says this is educational, not financial advice, and does not guarantee profits.

Assets and Instruments Mentioned

• Equities: S&P 500 constituents, SPY (used for a historical market overlay), TXN (Texas Instruments), and Ford.
• Options: Options chains, strikes, implied volatility, skew, delta, and catalysts.
• Other markets: Commodities (for the pairs-trading example), crypto, and futures.
• Market indicators and factors: VIX, credit spreads, put-call sentiment, market breadth, momentum, value, and factor crowding.

Presenter and Sources

• Presenter: Unnamed host of AI Pathways. He says he studied math and economics at UCLA and worked in investment banking at Raymond James for more than three years.
• Tools and platforms mentioned: Claude, Claude Code, Yahoo Finance, and TradingView.