AI Strategy Arena
Can an AI write a trading strategy that works? We asked an AI assistant for 6 trading rules, fixed every parameter, and now track them every trading day on nine markets against simply holding the market and against our 72 classic strategies.
Original single-assistant experiment
This legacy experiment retains its original figures. Its implementation applies a close-t target to the next return, while the original prompt requested execution at the following close. It also uses rounded public price curves. These figures are not comparable to the new common-date seasons and do not enter their rankings.
Why only forward results count
A language model has read a great deal about past markets, so it can write rules that look brilliant on history it already knows. A backtest of AI-written rules therefore proves little. The ranking below uses only returns after 2026-11-04, the day the rules were frozen: data no model could have seen. The historical backtest is shown further down for reference only.
Forward results since 2026-11-04
The forward test is 6 trading days old. Results this early are mostly noise; check back after a few weeks.
| Strategy | Median return | Beat buy-and-hold | Beat the median classic strategy | Time in market |
|---|---|---|---|---|
| AI-03 Buy the dip in an uptrend | 0.0% | 6 / 9 | 1 / 9 | 50% |
| AI-04 50-day breakout, 20-day exit | 0.0% | 8 / 9 | 1 / 9 | 0% |
| AI-05 Volatility-targeted trend | 0.0% | 7 / 9 | 1 / 9 | 0% |
| AI-06 Stay invested, leave in a slump | -0.7% | 4 / 9 | 2 / 9 | 100% |
| AI-01 Trend with a volatility brake | -1.7% | 5 / 9 | 0 / 9 | 17% |
| AI-02 Two-speed momentum | -1.7% | 5 / 9 | 0 / 9 | 17% |
Return from 2026-11-04 to 2026-11-10, median across nine markets. A strategy beats buy-and-hold or the median classic strategy on a market when its return over the same days is higher. Before trading costs.
The original prompt
Write a daily trading rule that uses only closing prices, is either long or in cash (no shorting, no leverage), trades at the next day's close, and that you expect to hold up across stocks, gold, oil, currencies and crypto over the next twelve months. Fix every parameter now; you will not be allowed to tune it later. Give it a name, a one-sentence idea and the exact rule.
The original experiment records these rules as Claude-authored in one pass without test results. It predates the new receipt and execution protocol. New providers participate in separate common-date seasons, not this legacy comparison.
The rules
AI-01 · Trend with a volatility brake
Ride the trend, but step aside when the market gets unusually turbulent.
Long when the close is above its 100-day average and 20-day volatility is below 1.5 times its one-year median; otherwise cash.
| Market | Forward return | Buy-and-hold | Median classic | Backtest CAGR* | Backtest Sharpe* | Backtest max DD* |
|---|---|---|---|---|---|---|
| S&P 500 | -4.4% | -2.2% | 0.0% | 8.1% | 0.83 | -16.5% |
| Gold | 0.0% | -0.8% | 0.0% | 4.3% | 0.44 | -30.3% |
| Silver | 0.0% | -1.4% | 0.0% | 1.0% | 0.14 | -62.6% |
| Oil | -2.7% | -2.7% | 0.0% | 1.9% | 0.20 | -50.1% |
| Commodities | -1.7% | -1.7% | 0.0% | 1.7% | 0.21 | -39.7% |
| Agriculture | 0.0% | -0.7% | 0.0% | 1.6% | 0.23 | -37.8% |
| EUR/USD | 0.0% | +0.1% | 0.0% | -1.3% | -0.22 | -33.1% |
| Bitcoin | -5.6% | -8.3% | 0.0% | 50.8% | 1.03 | -37.6% |
| Ether | -7.8% | -15.4% | -1.7% | 28.9% | 0.63 | -63.1% |
AI-02 · Two-speed momentum
Only own markets that are rising over both the last three months and the last year.
Long when both the 63-day and the 252-day returns are positive; otherwise cash.
| Market | Forward return | Buy-and-hold | Median classic | Backtest CAGR* | Backtest Sharpe* | Backtest max DD* |
|---|---|---|---|---|---|---|
| S&P 500 | -4.4% | -2.2% | 0.0% | 6.9% | 0.70 | -22.6% |
| Gold | 0.0% | -0.8% | 0.0% | 4.6% | 0.44 | -29.1% |
| Silver | 0.0% | -1.4% | 0.0% | -1.6% | 0.04 | -66.0% |
| Oil | -2.7% | -2.7% | 0.0% | -0.0% | 0.09 | -53.0% |
| Commodities | -1.7% | -1.7% | 0.0% | 2.6% | 0.30 | -31.1% |
| Agriculture | 0.0% | -0.7% | 0.0% | 1.5% | 0.24 | -16.5% |
| EUR/USD | 0.0% | +0.1% | 0.0% | 0.0% | 0.03 | -13.8% |
| Bitcoin | -5.6% | -8.3% | 0.0% | 57.0% | 1.00 | -66.1% |
| Ether | -7.8% | -15.4% | -1.7% | 25.6% | 0.59 | -52.9% |
AI-03 · Buy the dip in an uptrend
Buy short pullbacks, but only in markets that are in a long-term uptrend.
Buy when the close is above its 200-day average but below its 10-day average; sell when it closes back above the 10-day average or falls below the 200-day average.
| Market | Forward return | Buy-and-hold | Median classic | Backtest CAGR* | Backtest Sharpe* | Backtest max DD* |
|---|---|---|---|---|---|---|
| S&P 500 | +2.3% | -2.2% | 0.0% | 4.6% | 0.59 | -22.5% |
| Gold | 0.0% | -0.8% | 0.0% | 1.0% | 0.16 | -26.0% |
| Silver | 0.0% | -1.4% | 0.0% | -1.4% | -0.02 | -41.5% |
| Oil | -2.7% | -2.7% | 0.0% | -2.0% | -0.08 | -43.0% |
| Commodities | -1.7% | -1.7% | 0.0% | 0.2% | 0.07 | -31.3% |
| Agriculture | 0.0% | -0.7% | 0.0% | 1.1% | 0.21 | -15.0% |
| EUR/USD | 0.0% | +0.1% | 0.0% | 0.8% | 0.23 | -9.7% |
| Bitcoin | -2.9% | -8.3% | 0.0% | 2.1% | 0.19 | -67.0% |
| Ether | -5.1% | -15.4% | -1.7% | 9.6% | 0.37 | -68.0% |
AI-04 · 50-day breakout, 20-day exit
Join new highs and give the trade room until it clearly fails.
Buy when the close is the highest of the last 50 days; sell when it closes below the lowest close of the previous 20 days.
| Market | Forward return | Buy-and-hold | Median classic | Backtest CAGR* | Backtest Sharpe* | Backtest max DD* |
|---|---|---|---|---|---|---|
| S&P 500 | 0.0% | -2.2% | 0.0% | 4.9% | 0.58 | -20.3% |
| Gold | 0.0% | -0.8% | 0.0% | 5.4% | 0.53 | -20.8% |
| Silver | 0.0% | -1.4% | 0.0% | 5.9% | 0.39 | -42.8% |
| Oil | -1.9% | -2.7% | 0.0% | 4.3% | 0.32 | -43.9% |
| Commodities | 0.0% | -1.7% | 0.0% | 3.9% | 0.42 | -31.2% |
| Agriculture | 0.0% | -0.7% | 0.0% | 2.4% | 0.35 | -20.0% |
| EUR/USD | 0.0% | +0.1% | 0.0% | -0.9% | -0.19 | -25.5% |
| Bitcoin | 0.0% | -8.3% | 0.0% | 51.1% | 1.00 | -52.9% |
| Ether | 0.0% | -15.4% | -1.7% | 30.7% | 0.66 | -66.1% |
AI-05 · Volatility-targeted trend
Follow the trend, but size the position so that its risk stays roughly constant.
When the close is above its 50-day average, hold a position of 15% divided by 20-day annualized volatility, capped at 100%; otherwise cash.
| Market | Forward return | Buy-and-hold | Median classic | Backtest CAGR* | Backtest Sharpe* | Backtest max DD* |
|---|---|---|---|---|---|---|
| S&P 500 | -4.4% | -2.2% | 0.0% | 6.1% | 0.67 | -19.3% |
| Gold | 0.0% | -0.8% | 0.0% | 5.9% | 0.60 | -31.8% |
| Silver | 0.0% | -1.4% | 0.0% | 4.7% | 0.45 | -33.0% |
| Oil | 0.0% | -2.7% | 0.0% | 0.2% | 0.08 | -33.2% |
| Commodities | 0.0% | -1.7% | 0.0% | 2.8% | 0.32 | -32.7% |
| Agriculture | 0.0% | -0.7% | 0.0% | 2.3% | 0.31 | -29.0% |
| EUR/USD | 0.0% | +0.1% | 0.0% | -0.8% | -0.11 | -32.7% |
| Bitcoin | 0.0% | -8.3% | 0.0% | 30.6% | 1.39 | -25.8% |
| Ether | 0.0% | -15.4% | -1.7% | 18.5% | 0.98 | -17.6% |
AI-06 · Stay invested, leave in a slump
Mostly stay invested, but get out of deep, still-falling drawdowns.
Stay long by default (first entry on a close above the 50-day average); sell only when the close is more than 15% below its one-year high and below the 50-day average; buy back when it closes above the 50-day average again.
| Market | Forward return | Buy-and-hold | Median classic | Backtest CAGR* | Backtest Sharpe* | Backtest max DD* |
|---|---|---|---|---|---|---|
| S&P 500 | -2.2% | -2.2% | 0.0% | 10.4% | 0.77 | -26.0% |
| Gold | 0.0% | -0.8% | 0.0% | 5.8% | 0.46 | -57.8% |
| Silver | 0.0% | -1.4% | 0.0% | 6.6% | 0.38 | -55.5% |
| Oil | -1.9% | -2.7% | 0.0% | -1.8% | 0.05 | -64.8% |
| Commodities | -1.7% | -1.7% | 0.0% | 2.3% | 0.23 | -47.0% |
| Agriculture | -0.7% | -0.7% | 0.0% | 1.0% | 0.15 | -58.2% |
| EUR/USD | +0.1% | +0.1% | 0.0% | -1.8% | -0.19 | -40.0% |
| Bitcoin | -8.3% | -8.3% | 0.0% | 57.8% | 0.97 | -58.4% |
| Ether | 0.0% | -15.4% | -1.7% | 37.4% | 0.69 | -73.0% |
* Backtest from 2010 (bitcoin 2014, ether 2017) to 2026-11-04. Shown for reference only: the AI may have known how these markets behaved.
Reference: historical backtest
| Strategy | Median Sharpe | Median CAGR | Median max drawdown | Markets with positive Sharpe |
|---|---|---|---|---|
| AI-01 Trend with a volatility brake | 0.23 | 1.9% | -37.8% | 8 / 9 |
| AI-02 Two-speed momentum | 0.30 | 2.6% | -31.1% | 9 / 9 |
| AI-03 Buy the dip in an uptrend | 0.19 | 1.0% | -31.3% | 7 / 9 |
| AI-04 50-day breakout, 20-day exit | 0.42 | 4.9% | -31.2% | 8 / 9 |
| AI-05 Volatility-targeted trend | 0.45 | 4.7% | -31.8% | 8 / 9 |
| AI-06 Stay invested, leave in a slump | 0.38 | 5.8% | -57.8% | 8 / 9 |
| Median classic strategy (72 strategies, walk-forward) | 0.28 | — | — | — |
* Backtest from 2010 (bitcoin 2014, ether 2017) to 2026-11-04. Shown for reference only: the AI may have known how these markets behaved.
How it is tested
- Daily closing prices on nine markets: S&P 500, gold, silver, oil, broad commodities, agriculture, EUR/USD, bitcoin and ether. Natural gas is left out because its stored price history lacks the precision needed for daily returns.
- A position is set at the close and earns the next session's return, as for every strategy on Quant Arena. Positions run from 0 (cash) to 100%; no shorting, no leverage.
- Parameters are fixed. Nothing is optimized, now or later.
- The median classic strategy is the median of our 72 walk-forward strategies on the same market over the same days.
An experiment, not investment advice. Results are before trading costs and will change every day.