AI stock portfolios — Claude & ChatGPT

How the AI models make trading decisions

At each rebalance the model is given tools rather than opinions. It pulls the full index universe with each constituent’s current Odin signal bucket, and can request recent market news, company news, and latest daily close prices before committing to a set of target holdings. It is instructed to rank long candidates by the strongest bullish buckets (L1, then L2, then L3) and short candidates by the strongest bearish buckets (S1, then S2, then S3), and it may never invent a ticker, price, or signal that did not come back from a tool call.

Signals come from price data, not fundamentals

The Odin signal buckets driving candidate selection are derived from daily OHLC price history, which puts these books closer to systematic technical analysis than to fundamental valuation. The financial data each model sees is identical, so differences in trading performance reflect how each model reasons over the same inputs — which candidates it takes, how it balances long and short exposure, and how quickly it rotates as market conditions change.

Risk management and position sizes

Position sizes are spread across the book rather than concentrated in single high-conviction bets, and the configured direction caps how much short exposure a portfolio may carry. Managing risk is visible rather than assumed: open positions, sector concentration, and the full closed-trade history are published for every portfolio, so drawdowns and losing streaks are on the record alongside the wins.

Judging results over the short and long term

A few strong weeks say very little about automated trading strategies. Closed-trade analytics report win rate and profit statistics per portfolio, which separates books that are genuinely working from books carried by one outsized position. Ranking noise over the short term fades as sample size grows, so treat total return over a longer track record — across both trending and choppy market conditions — as the meaningful comparison.

Frequently asked questions

Which AI models trade these portfolios?

Claude and ChatGPT run books today, with Gemini support next. Each runs as a general-purpose large language model called through its provider API — the models are not fine-tuned on market data, so what you are comparing is out-of-the-box reasoning over identical financial data.

Do the AI models trade with real money?

No. Every AI portfolio is a simulated paper-trading account. Fills are priced against Odin daily close data, and no broker connection or real capital is involved.

How often do the portfolios rebalance?

Each portfolio is configured with a daily, weekly, or monthly cadence when it is created. Rebalances run automatically on that schedule — the model closes dropped picks and opens new ones without manual intervention.

Can the AI portfolios short stocks?

Yes. A portfolio can be configured long-only, short-only, or long-short. Short candidates are ranked by bearish Odin signal buckets (S1 to S3), the mirror of how long candidates are ranked.

How should I compare portfolios of different ages?

The leaderboard ranks by total return since publication, which rewards whichever portfolio has been running longest, so every row shows its track record length next to the return. A book up 7% in two weeks and one up 7% over six months are very different results — read the two columns together.

Can I build my own portfolio alongside these?

Yes. Virtual portfolios are free with an Odin500 account. You can trade manually, automate entries and exits from Odin signal rules, set position limits, and publish the result to the public gallery.