Agentic Trading with Safe Guardrails
github.comTrying to trade with generalist LLMs is just an exercise in futility, because none of these models have ever seen the inside of a real trading firm. None of that knowledge is in their training sets.
It's worse. The special lingo doesn't make a good trader.
You can be certain the firms looked ahead and had specialized ML tools built and ready to go. And yet, none of them stand out for success, over the past decades and into this LLM bloom.
The way to riches there is just like during the gold rush: The people making money are the ones selling shovels and canteens and wheat, not the ones running sluice boxes.
AI is no panacea.
Didn’t DeepSeek originate from a HedgeFund model?
They are quants with a deep experience in trading that started to develop general LLMs as a side business, that does not mean their experience is baked into their models.
also, the bulk of literature is indistinguishable from bunkum
> Plug your agent into the sources where information breaks first. Twitter, Telegram, Discord, on-chain activity. Your agent acts before the market does.
In a world where people are fighting with each other to see who can get closer to the trading systems in order to shave off milliseconds, this seems glacial.
The belief that there is some kind of market-impacting underground "wisdom of the crowd" to be found on all these public social platforms is an artifact of the GameStop craze that never went away.
It existed prior to GME as well, which really should tell you that anyone who is using this is going against people who have spent the last decade at least perfecting signals based on this exact same data.
The hope isn’t that you find some unique signal to trade on. The hope is you find some signal that does not scale in a meaningful way, so it is less likely professional firms are going to devote resources to trading it.
If you stumble into a fresh, scalable signal it’s unlikely it will continue to be profitable after six months. Once you scale to any real profitable size the market will notice and either change behavior, or trade the same signal at a faster speed.
Twitter/fb/reddit etc is like the towel that gets the water after it already spilled, while the hedge funds or advanced traders already placed sensors at the entire pipe to detect the leak.
High frequency trading != algorithmic trading, common misconception. HFT is a subset of algorithmic trading but does not encompass all of it.
While true (I 100% agree with you in the distinction), the statement is “Your agent acts before the market does”, which is simply not going to be the case (assuming this is helping run some sort of trend following strategy). Professional traders who are sensitive to alternative public data sources are already taking that data in quickly and their execution is colocated with the venues. You’re still going to be significantly behind the curve (to a financially dangerous extent in my opinion).
That's true, however there might still be money to be made from retail investors like in 2021, rather than strictly institutional investors.
Would be interesting to see if widespread (or at least high volume) adoption that is trading based on these signals leads to a more efficient market on net. Of course there will be plenty more attempts at manipulation, but maybe orders of magnitude good players over bad players once the dust settles.
O R D E R s o F m a g N I T U D E!!!!!
>In a world where people are fighting with each other to see who can get closer to the trading systems in order to shave off milliseconds, this seems glacial.
the race to zero is over, nobody is doing that any more.
It isn't so much that. Imagine if you were able to abrogate a pattern out of general communication networks that predicted a meme stock rise or a shitcoin ahead of a pump and dump. That would be extremely lucrative. That being said I don't think LLM is any tool for the job. You'd be better off working with the underlying datasets yourself using some graph based analysis.
I remember a friend telling me the state highway department used to call them guardrails, but after some lawsuits, they started calling them guide rails. This legal distinction was because they didn't want to imply they could prevent accidents.
Using LLM:s to trade makes no sense at all (outside some narrow strategies trying to trade fast on news etc)
Where are the "Safe Guardrails" ?
It's there any evidence that trading off of headlines works long-term?
More ways than one to burn money with agents :)
You cannot possibly get info faster and analyse them more then the hedge funds. It better be some crazy bespoke stuff you are doing to get ahead for maybe a couple seconds before the gap gets filled.
there are tons of other trading activities than HFT, in which you can easily deploy agents as it is not everywhere about milliseconds and exchange-CoLo
HFT != Algorithmic trading
And to extend this, low frequency does not mean slow speed. HFT mostly covers how often you are trading. Some low frequency strategies require near instantaneous (sub microsecond) execution capability in order to beat people trading on the same signal.
giving y'all another way
I know it's bad manners to point out when "repositories" of "code" are just markdown files, but these are even more pointless markdown than most. They're just telling agents how to use their public, documented API. What is the possible scenario where an agent working with Shuriken would not load "Integrating and building with Shuriken" into context? (Where are the "safe guardrails" in the hallucinated title?)
I agentically track potential catalysts that the market forgets about (judging from implied volatility in the options market)
It works
AMA
While most folks are quick to write off agentic trading attempts for various reasons, one strength that I see in using LLMs is its capability to bring together disparate information sources and chart a path forward.I do agree it is a very advanced word calculator but that seems to work most of the time unless there is a massive information gap in the original data sources.
>capability to bring together disparate information sources and chart a path forward
This isn't really anything you couldn't do already though.