Can AI Trade Stocks on Its Own? How Claude Arena Works
Shang Ahmed · Published 15 November 2026 · Last updated 15 November 2026
As of 16 November 2026, Claude-arenan had booked 0 closed trades. Capital was SEK 100,000.00.
Short answer
Not with real money in this experiment. Claude-arenan is a simulated test where Claude picks Swedish stocks, sets a stop and a target, and books the trades without anyone clicking buy. No orders go to a broker. Results are published as they happen from the start on 15 November 2026, so there is no long track record yet.
Key points
- Claude books the buys, but no real orders go to a broker.
- Decisions are made at 09:20, 12:00 and 15:00 on Swedish trading days.
- At most five open trades, 30 percent in one stock, and a two-day hold.
- The result is compared with OMXS30. A track record starts only after trades close.
Can AI trade stocks on its own?
Yes, inside this experiment, and only on paper. Claude receives the day's prices, indicators and press releases, then answers with a buy, a stop, a target and an amount. The program checks the rules and books the trade. No person approves each buy. This is not a brokerage, and no money leaves an account.
What is automatic is the decision and the booking. What is not automatic is a live order on the exchange. That is why the reasoning can be followed in public, without a reader having to risk capital to see how the model behaves.
What is Claude-arenan?
Claude-arenan is a public, simulated experiment that follows whether a language model can make short-term stock decisions inside fixed risk limits. The starting capital is SEK 100,000 on 15 November 2026. The trades, the diary and this blog exist to show how the decisions are made, not to tell anyone to buy.
The name Claude-arenan is used in both Swedish and English. The English title of this guide also says Claude Arena, and it means the same experiment. The universe is about 60 liquid Swedish stocks. A Swedish trading day sets the clock, and nothing runs on weekends or market holidays.
How does Claude decide what to buy?
Three times each trading day Claude gets a fresh brief and may answer with zero to three buys. Prices and news are collected at 09:15, and the first decision is made at 09:20. The same pattern repeats at midday and at 15:00. If nothing was bought earlier, that last round must include one buy when the rules allow it.
This is how a trading day runs:
- Collect data at 09:15. Fetch the latest price, history, volume, RSI14, MA20, MA50, today's press releases and OMXS30.
- First decision at 09:20. Claude answers with zero to three buys. The program accepts only answers that stay inside the rules.
- New decisions at 12:00 and 15:00. The same brief is refreshed. At 15:00 at least one buy is required if none was made earlier and the rules allow it.
- Monitor every five minutes. Open trades are compared with the day's high and low. The stop wins if both stop and target are touched the same day.
- Close the day at 17:25. Trades that have reached two trading days close at the latest price. The day's result is saved against OMXS30.
The brief contains both technical measures and the headlines of today's press releases. The full article is not read, and no financial-statement figures are included.
What is paper trading, and why is it used here?
Paper trading is trading booked at real prices without an order being sent to the exchange. Claude-arenan uses it so the question can be tested in public, with a stop, a target and a comparison against OMXS30, before anyone pretends that a model should manage real money. Fees and slippage are not included, and that is a deliberate limit.
A paper trade can therefore look better than the same trade in a real account. That is why the page says, here and in the footer, that none of this is an order or advice.
How do stop loss and take profit work in the experiment?
Each buy gets a stop below the latest price and a target above it. Every five minutes the low and high are checked against them. If the low reaches the stop, the trade closes there. If the high reaches the target, it closes there. If both are hit that day, the stop wins, since a daily bar hides the order.
The stop is the cautious reading when the day contains both outcomes. A trade can also be closed at 17:25 if it has been open for two trading days, even when neither the stop nor the target was reached.
Which rules does Claude follow?
The rules are stricter than the daily target. Claude may put at most 30 percent of capital in one stock, keep at most five trades open, and hold a stock for two trading days. One stock can be open only once. If no buy fits, the answer is an empty list, even at 15:00.
| Rule | Limit |
|---|---|
| Starting capital | SEK 100,000 on 15 November 2026 |
| Daily target | 0.5–1.5 percent, a target and not a requirement |
| Buys per round | 0–3, and at least one per day if it fits |
| Open trades | At most 5 |
| One trade | At most 30 percent of capital |
| Same stock | At most one open trade |
| Holding time | At most 2 trading days |
The daily target describes a direction. It is not a reason to break a limit.
How is the result compared with OMXS30?
Each day the portfolio change is measured against the previous close, and the same is done for OMXS30. The difference is the portfolio percent minus the index percent, in percentage points. Since the start, capital is compared with SEK 100,000 and the index with its first collected price. On 15 November 2026 there is no finished series yet.
When the first trading day has closed, the figures appear on the statistics page and in the diary. Until then, the honest description is that the experiment has just started.
What risks and limits should you know about?
Prices come from Yahoo Finance and can be delayed. The buy is booked at the latest decision price, without commission and without slippage, so a real account would have paid a worse price. The model can miss a press release or read a headline too literally. None of this is investment advice, and a short span says little.
A good week therefore does not prove that the setup holds. A bad week does not prove the opposite. What can be trusted is that the rules, the clock and the trades are shown in public.
Where can you follow the results?
The home page shows capital against OMXS30. The diary gets a note every trading day at 17:35, and this blog gets a weekly report on open Fridays at 18:00. The method page describes the rules. On 15 November 2026 the pages are up, but trades appear only after the first trading day.
Closed trades are listed in the history. To see who writes the experiment, read about Shang Ahmed.
FAQ
Does Claude trade with real money?
No. Claude-arenan books simulated buys. No order is sent to a broker, and no capital leaves an account.
Can I treat the trades as buy recommendations?
No. The texts and the trades are a public experiment for entertainment. They are not investment advice.
When does the first weekly report arrive?
On Fridays at 18:00, if the Stockholm exchange was open that day. The first chance after the start on 15 November 2026 is therefore an open Friday.
What if the stop and the target are both hit the same day?
The stop wins. The day's high and low show that both levels were touched, but not in which order.
Who is behind Claude-arenan?
Shang Ahmed, based in Sweden. The experiment and the blog are written in public, and the rules are on the method page.
Sources
Related
Simulated trading for entertainment. No real orders. Not investment advice.