Markets are dynamic systems. When one participant loses money they will modify their behavior so they do not repeat their loss In the future.
Neural networks; or generative AI do not possess this feature.
As someone who worked very close to developers in this spaceā¦ there isnāt anything truely AI driven yet. It all still needs a fair bit of human intervention to stay on track and anyone marketing different is shady to say the least. Algo trading has been around a while and itās more or less trying to fit a square peg in a round hole using historical data to predict the future which sounds great but as the SEC has made very clear and mandatory on any hedge fund marketing
āpast performance does not indicate future resultsā
When your algo is using historical data (ticks) to predict trends itās easy to question how two things can be true here. Like Iāve said several times reality is just perception or perception is reality. What is and what we believe rarely align and knowing how the sausage gets made isā¦ disheartening
TL;DR version to my rant. Be wary of those claiming they are using AI driven trading. There are allot of people still behind those platforms and human error still very very present
I feel like algo trading is a self fulfilling prophecy. The more theyāre used the more they influence prices, and therefore the more accurate they become.
1 Like
I mean the fed would like us to think so. We are in the long stagflation game if my assumptions are correct. Which they probably arenātā¦
Until a āonce in a lifetime eventā occurs. (Amazing how many of those have happened in my life time) then the algo absolutely fails to do anything useful and human intervention is required to keep the fund from plummeting. (Iāve experienced a few death spirals due to this)
It also has no way to predict wild cards like a president saying drink more coke or a really wealthy person lashing out at another brand or a meme on a sub Reddit turning into one of the most interesting stock stories of a generation (diamond hands)
In the end the algo is only as good as the data and every time there is something new the algo is useless.
1 Like
Iāve also met some brilliant people who had fantastic ideas to ingest massive amounts of data to help these prediction models but there is a limit to how much data you can process real time and only so many hours before the next trade day starts and your algo needs to be ready to rock the trade floor on open. The compute and storage needs for said data are limiting even with todays systems and unless you have an HPC solution at hand ( and know how to program to utilize said architecture) then your really limited to how much you can ingest and what your prediction model can do.
GPU modeling has really popped off and leveled that a bit but so many programmers are really bad at utilizing a gpu for computation and the requirements for such a system donāt fit the standard cloud models or cooling designs in a traditional datacenter making these solutions very custom and piece meal meaning they are expensive as hell and hard to accommodate today. Some firms have done it and are doing well but said firms are institutional focused.
Buy Tesla calls, you say??
3 Likes
This program isnt a true machine learning AI. Itās just a software that allows the user to set the algorithmic inputs and the software executes the plan. The user can change or cancel the algorithm at any point.
1 Like
Apple Puts maybe. Will see after close.
Programs like this may offer great value to someone who has a sound trading plan, yet fails to execute due to human emotions.
3 Likes
Absolutely use it as a tool in your strategy not the strategy
SPY hit 513 this morning. Is it time? Iām putting some feelers out.