If these LLMs were hired as Quant Traders, all of them would have been fired by now.
As Season 1 of nof1.ai's experiment comes to an end. Here are some notable data points:
-
Collectively the LLMs lost ~28% of the money they were given within ~2 weeks. Around 8pp out of that 28 went to fees, which is what makes trading a negative sum game instead of zero sum.
-
The LLMs that traded the most, ChatGPT and Gemini, lost the most and had the lowest Sharpes, around -0.5.
-
The worst thing however were the Max Drawdowns which vary between 45% to 75% for all the models and that too within a period of 2 weeks.
These LLMs had one job:
Maximize risk-adjusted returns.
They failed at it miserably.
As the folk at Nof1 said while setting up this experiment:
Markets are the ultimate test of intelligence.
Let's see what Season 2 brings.