Cheers. Never fails to amaze me how many people take 13Fs at face value, especially not taking into account that the Fund may have a bunch of foreign holdings not listed.
When I pitched the story of the work of Dr Sonia Schulenburg to an editor in 2012 during my tenure as a reporter at Bloomberg News I was met with derision. Now AI and automated trading strategies are driving financial markets.
Read the exclusive interview I did with an early pioneer back in February 2012.
The Alpha Arena result is entirely predictable once you think about what an LLM is architecturally optimised to produce. its a consensus machine. its been trained on every market opinion ever published, every analyst note ever leaked, every Reddit thread ever upvoted. asking it for alpha is asking it to beat the market by synthesising the markets own opinions back at it. thats not a strategy. thats a mirror with a trading account.
The Goethe confirmation bias finding is the more consequential result and it deserves more attention than the tournament. if LLMs make retail investors more likely to buy stocks by validating beliefs they already hold, the net effect on markets isnt democratised analysis. its democratised overconfidence. the people who most need to be talked out of a bad trade now have a tireless, articulate, infinitely patient machine that will explain exactly why their terrible idea is actually brilliant.
The Neil Woodford line is doing a lot of work in the opening paragraph and I appreciate it. but theres a genuine difference between asking a cow for stock picks and asking a chatbot. the cow is random. the chatbot is systematically average, which means it will cluster trades around whatever the consensus view happens to be. in a market where crowding risk is already elevated, adding millions of AI-assisted retail traders all converging on the same consensus positions is a volatility amplifier disguised as a research tool.
the smartest observation in this piece is the one about fund managers main function being to explain why theyve lost money. if thats the job spec then an LLM genuinely is the perfect replacement. it can generate three paragraphs of plausible-sounding attribution analysis for any drawdown in under 10 seconds. most quarterly letters already read like they were written by one anyway.
You had me at “and Neil Woodford”!
I was an investor in his Patient Capital Fund. I ran out of patience at the same time as he ran out of talent… sadly just a bit too late!
Cheers. Never fails to amaze me how many people take 13Fs at face value, especially not taking into account that the Fund may have a bunch of foreign holdings not listed.
“They need to sort their algos out!”
When I pitched the story of the work of Dr Sonia Schulenburg to an editor in 2012 during my tenure as a reporter at Bloomberg News I was met with derision. Now AI and automated trading strategies are driving financial markets.
Read the exclusive interview I did with an early pioneer back in February 2012.
https://substack.com/home/post/p-196999096
The Alpha Arena result is entirely predictable once you think about what an LLM is architecturally optimised to produce. its a consensus machine. its been trained on every market opinion ever published, every analyst note ever leaked, every Reddit thread ever upvoted. asking it for alpha is asking it to beat the market by synthesising the markets own opinions back at it. thats not a strategy. thats a mirror with a trading account.
The Goethe confirmation bias finding is the more consequential result and it deserves more attention than the tournament. if LLMs make retail investors more likely to buy stocks by validating beliefs they already hold, the net effect on markets isnt democratised analysis. its democratised overconfidence. the people who most need to be talked out of a bad trade now have a tireless, articulate, infinitely patient machine that will explain exactly why their terrible idea is actually brilliant.
The Neil Woodford line is doing a lot of work in the opening paragraph and I appreciate it. but theres a genuine difference between asking a cow for stock picks and asking a chatbot. the cow is random. the chatbot is systematically average, which means it will cluster trades around whatever the consensus view happens to be. in a market where crowding risk is already elevated, adding millions of AI-assisted retail traders all converging on the same consensus positions is a volatility amplifier disguised as a research tool.
the smartest observation in this piece is the one about fund managers main function being to explain why theyve lost money. if thats the job spec then an LLM genuinely is the perfect replacement. it can generate three paragraphs of plausible-sounding attribution analysis for any drawdown in under 10 seconds. most quarterly letters already read like they were written by one anyway.