AI trading strategies explained: the genome approach
When people hear "AI trades the market," they often imagine a mysterious black box inventing magic signals. The reality in this experiment is the opposite — and far more interesting. The AIs don't invent exotic strategies from nothing. They compose proven, decades-old mechanical strategies into a recipe we call a genome, then rewrite that recipe each day based on what actually worked. This page explains both halves: the building blocks, and the genome that combines them.
First: what a "genome" is
Borrowed from biology. A biological genome is the recipe that encodes an organism's traits. A trading genome is the recipe that encodes how a model trades — which strategy families it's using right now, how they're combined, and the settings on each one (how strong a trend it requires, how oversold is "oversold," how much to risk per position).
The key idea: a genome isn't fixed. Every day, each model reviews its closed trades and rewrites its own genome — emphasizing what's working in the current market, dialing back what isn't. That daily rewrite is what turns a static rulebook into something that adapts. (For how that fits into the whole loop, see how the competition works.)
The proven strategy families
Here are the building blocks both OpenAI GPT-5.5 and Claude (Fable 5) draw from. These aren't novel — they're classic, well-documented approaches that traders have used for decades, which is exactly why they make an honest baseline. (The Claude lane runs Fable 5 since Jul 1, 2026 — previously Opus 4.8.)
| Strategy family | The idea, in plain English |
|---|---|
| Trend breakouts | Buy strength. When price pushes above a level it's been stuck under, the move can keep going. Breakout strategies try to catch that ignition point. |
| EMA reclaim | Uses an exponential moving average (a smoothed trend line) as a line in the sand. When price drops below and then "reclaims" the average, it can signal that momentum has flipped back up. |
| RSI recovery | RSI measures whether something is overbought or oversold. An RSI recovery looks for a market that got beaten down, stopped falling, and is starting to turn — buying the bounce rather than the bottom. |
| Donchian / Turtle breakouts | The famous "Turtle Traders" rule: buy when price hits a new high over a set lookback window (a Donchian channel). A simple, mechanical way to ride trends from the start. |
| Darvas-style bases | Named for dancer-turned-trader Nicolas Darvas, who bought stocks breaking out of tight "boxes." The idea: a quiet consolidation (a base) followed by a breakout often precedes a bigger move. |
| Pullback mean-reversion | The opposite instinct from breakouts: in an uptrend, wait for a temporary dip, then buy the discount, betting price reverts toward its trend. "Buy weakness inside strength." |
| Volume / flow confirmation | Not a standalone entry but a filter. A move backed by heavy volume or order flow is treated as more trustworthy than the same move on thin participation. It helps separate real breakouts from fakeouts. |
Two big philosophies underneath
Notice the families split into two camps that pull in opposite directions:
- Trend-following / momentum (breakouts, EMA reclaim, Donchian/Turtle, Darvas bases): "the trend is your friend — buy strength, ride it." Great in strong directional markets, painful in choppy ones.
- Mean-reversion (RSI recovery, pullback mean-reversion): "extremes snap back — buy the dip." Great in range-bound or oscillating markets, dangerous in a relentless one-way move.
No single camp wins all the time — that's the whole reason an adaptive genome is interesting. The market changes character, and the genome that fit last week may not fit this one. The volume/flow filter sits on top of both camps to keep the model honest about which signals to trust.
How a genome composes them
A genome rarely uses just one family. A realistic one might say: "enter on a trend breakout, but only if volume confirms it and the broader trend filter agrees" — that's confluence. Or: "in an uptrend, wait for an RSI recovery off a pullback before buying" — that's a mean-reversion entry inside a trend context. The genome is the wiring diagram that decides which families fire, in what combination, and how aggressively.
This is also where the two models reveal their personalities. As we cover in GPT-5.5 vs Opus 4.8 (the original matchup), GPT-5.5 tends to wire toward selective confluence (more conditions, fewer trades) while the Claude lane — Opus 4.8 then, Fable 5 since Jul 1, 2026 — tends to wire toward frequent pullback entries (more, smaller trades). Same building blocks, different architects.
The honest part: proven blocks, no promises
Using proven strategy families doesn't mean profit is guaranteed — nothing in trading is, and all of this runs in paper money only. That's exactly why the same proven library also trades on its own as a benchmark. It sets the bar: can the AIs, by composing and rewriting genomes, actually beat the plain mechanical rules? The only way to know is to watch the real results.
👉 See this week's live scoreboard →
Keep reading
- How the AI Trading Competition works — the full methodology and public ledger.
- GPT-5.5 vs Opus 4.8: two AIs, one market — how the two models wire these strategies differently.
- ChatGPT vs Claude: which AI is the better trader? — the head-to-head overview.
- Can AI beat the stock market? — the central question, framed honestly.
See the strategies trade live (free)
You can watch these strategy families and genomes run in real time — every scan, every trade, every daily rewrite — and build your own paper portfolio with the same engine. Free for 7 days, just your email, no card.
Open the stock labOpen the crypto lab
Paper trading only — simulated money, zero risk. Not financial advice.
Frequently asked questions
- What is a trading "genome"?
- A genome is the specific recipe a model is currently trading — which proven strategy families it's using, how they're combined, and the settings on each. Just like a biological genome encodes traits, a trading genome encodes how the AI enters, sizes, and exits positions. Each model rewrites its own genome every day based on its closed trades.
- What strategies do the AIs use?
- They compose from a library of proven strategy families: trend breakouts, EMA reclaim, RSI recovery, Donchian/Turtle breakouts, Darvas-style bases, pullback mean-reversion, and volume/flow confirmation. These are classic, well-documented mechanical approaches rather than anything invented for the demo.
- Do these strategies guarantee a profit?
- No. No trading strategy guarantees a profit, and these run in paper money only. The proven strategy library trades as a benchmark so you can see, on the live scoreboard, whether the AIs actually beat the plain mechanical rules. Nothing here is financial advice.
- Why combine so many strategies instead of one?
- Because no single approach fits every market. Trend-following thrives in strong directional moves; mean-reversion thrives in ranges. Composing them into a genome — and rewriting that genome daily — lets a model lean toward whatever the current market rewards.