As the northern hemisphere summer draws to a close, it beckons in a restarting of everything from the school year to the European football (soccer) seasons. We at Plotinus have decided as part of our ongoing research into AI powered trading, to run a small experiment, in the vein of a football league. Its objective is to determine how well, by creating a trading strategy, some of the large language generative AI models, will performance against each other and standard investment benchmarking.
This experiment will give investors and managers a window on how these various AI models cope with meeting the fluctuations of the stock market. Hopefully this will provide insights on the relative performance each model, whether any of the LLMs are better suited for this purpose than others and crucially of course to see how they fair in the active trading vs passive investment test.
The LLM Trading Teams
- Open AI’s GPT-4 Turbo
- Anthropic’s Claude 3.5 Sonnet
- Alphabet’s Gemini Pro
- Microsoft’s Phi-3 Medium
- Mistral’s Mistral Large
- Meta’s Code Llama 34B
- Open AI’s GPT-3.5 Turbo
- Anthropic’s Claude 3.5 Haiku
Objective and Rules
The task for each model, is to design a trading strategy that makes a singular daily trade for a given instrument. The models are free to construct their approach as each sees fit.
There are of course some rules to ensure a level playing field for all the contenders.
They can only use freely available data, to eliminate any acquired data advantage. The training data used can be from any time up until 31 December 2024. The testing data used is from 1 January 2025 – 29 August 2025. In order to reduce strategy drift, the models will be restricted to retraining their strategy on a three-month basis.
Assessing Performance
Various key metrics will be monitored to evaluate the models’ strategies.
- Accuracy of Prediction (assessing trading signal)
- Sharpe Ratio (assessing risk adjusted return)
- Total Return (cumulative return for the season)
- Maximum Drawdown (assessing peak to trough losses)
The competition will run from September 2 with the winner of the league being declared at the close of the season at the end of May 2026.
We look forward to updating you on the progress of the league and may the best team win! ■
© 2025 Plotinus Asset Management. All rights reserved.
Unauthorized use and/or duplication of any material on this site without written permission is prohibited.
Image Credit: AI at Shutterstock.