Yes. People have been trying to use AI as a statistical method for making money for literally fucking decades. Neural nets, genetic algorithms, statistical sampling, etc. etc. etc.
All you end up making 99% of the time is a volatility and/or momentum bot. Either a bot that makes tons of money when the stocks are predictable, or a bot that makes tons of money when stocks have wild unpredictable swings. When the opposite happens (ex: volatility is less than expected, or greater than expected), the bots collapse and you lose like $10 million bucks and everyone shuts down the bot. Every single time.
Its not even clear how you’re supposed to “test” a trading bot. Everyone’s got ridiculous ideas and “new AI” algorithms that try every few weeks, and they all fail before the unpredictability that is the market.
And then it turns out that you could have traded on volatility by just buying VIX and holding it anyway. So if you want a glorified volatility trader, there’s easier ways to do so than spending $millions on developers and $millions on computers and hooking it up to a $100-million bank account and praying for the best.
Just put the $100 million into VIX (or short-VIX) and bam. You roughly accomplish the same thing except it didn’t cost you $million developers or $million computers.
The stuff that makes money are like, Black Scholes differential equations (which are extremely fast. No "AI’ here, just numerical methods that directly compute a price). Of course, Bvlack Scholes is what they teach in college so everyone knows it. The secret sauce is the stuff that all the firms add to their computers and keep literally secret.
When it starts to happen, it will be a paper in a research journal. Then it will be years as people analyze the paper and come to undersstanding of what the new stuff can do.
Things don’t just “pop up” magically without warning. There’s papers, journals, discussions. If we aren’t even at the “discussion” point yet, its kind of worthless to spend more thought on it.
All this “ChatGPT” thing is an advanced neural network. Those things were first discovered in 1960s, Tensors (ie: applications of neural nets to SIMD compute) was 90s / 00s thing, and NVidia GPU optimizations to the models were researched through the 2010s.
I can reliably count on research taking decades. Because computers, algorithms, and AI is very difficult. Anyone paying attention in this field will see it coming.
Again, I’m not talking about ChatGPT. That is an entirely different type of AI from which I am talking about, something you should understand since you “work in computers.”
There wouldn’t be scientific studies of using AI, again- NOT ChatGPT- to predict the stock market if there were no way to do it. Or at least, those studies would say so. They don’t.
Again, I’m not talking about ChatGPT. That is an entirely different type of AI from which I am talking about, something you should understand since you “work in computers.”
You know that ChatGPT is a neural net tied to a large language model, right? Or the ANN fr that article you posted.
The other thing I’ve seen was hooking language models to predict positive vs negative news from news feeds, Twitter, and other sources of online discussion. Which is 100% in the realm of language models.
This is why it’s important to be specific about the algorithm you are talking about and not to just spitball. There’s lots of theoretical applications but no one has made much progress on making money as much as the HFT arbitrage bots.
In any case, you’re spitballing. Its all theoretical talk without any actual algorithm of note. You’re not talking about how Wall Street is organized or what HFTs are doing, which was the point of the post at the root of this discussion.
Its about Support Vector Machines (a statistical method) and ANNs (of which ChatGPT is one type of).
Did you read the link? Or did you just pick up the first hit from Google when you noticed this discussion wasn’t going the way you hoped? It doesn’t seem to have anything to do to counter my discussion point from earlier.
Also, I love that you keep telling me I’m spitballing as if I was claiming I wasn’t and I didn’t say in my very first post that it likely hasn’t happened yet.
Yes. People have been trying to use AI as a statistical method for making money for literally fucking decades. Neural nets, genetic algorithms, statistical sampling, etc. etc. etc.
All you end up making 99% of the time is a volatility and/or momentum bot. Either a bot that makes tons of money when the stocks are predictable, or a bot that makes tons of money when stocks have wild unpredictable swings. When the opposite happens (ex: volatility is less than expected, or greater than expected), the bots collapse and you lose like $10 million bucks and everyone shuts down the bot. Every single time.
Its not even clear how you’re supposed to “test” a trading bot. Everyone’s got ridiculous ideas and “new AI” algorithms that try every few weeks, and they all fail before the unpredictability that is the market.
And then it turns out that you could have traded on volatility by just buying VIX and holding it anyway. So if you want a glorified volatility trader, there’s easier ways to do so than spending $millions on developers and $millions on computers and hooking it up to a $100-million bank account and praying for the best.
Just put the $100 million into VIX (or short-VIX) and bam. You roughly accomplish the same thing except it didn’t cost you $million developers or $million computers.
The stuff that makes money are like, Black Scholes differential equations (which are extremely fast. No "AI’ here, just numerical methods that directly compute a price). Of course, Bvlack Scholes is what they teach in college so everyone knows it. The secret sauce is the stuff that all the firms add to their computers and keep literally secret.
Got it, it hasn’t happened yet so it will never happen.
I work in computers.
When it starts to happen, it will be a paper in a research journal. Then it will be years as people analyze the paper and come to undersstanding of what the new stuff can do.
Things don’t just “pop up” magically without warning. There’s papers, journals, discussions. If we aren’t even at the “discussion” point yet, its kind of worthless to spend more thought on it.
All this “ChatGPT” thing is an advanced neural network. Those things were first discovered in 1960s, Tensors (ie: applications of neural nets to SIMD compute) was 90s / 00s thing, and NVidia GPU optimizations to the models were researched through the 2010s.
I can reliably count on research taking decades. Because computers, algorithms, and AI is very difficult. Anyone paying attention in this field will see it coming.
Again, I’m not talking about ChatGPT. That is an entirely different type of AI from which I am talking about, something you should understand since you “work in computers.”
There wouldn’t be scientific studies of using AI, again- NOT ChatGPT- to predict the stock market if there were no way to do it. Or at least, those studies would say so. They don’t.
For example:
https://www.sciencedirect.com/science/article/pii/S2590291124000615
You know that ChatGPT is a neural net tied to a large language model, right? Or the ANN fr that article you posted.
The other thing I’ve seen was hooking language models to predict positive vs negative news from news feeds, Twitter, and other sources of online discussion. Which is 100% in the realm of language models.
This is why it’s important to be specific about the algorithm you are talking about and not to just spitball. There’s lots of theoretical applications but no one has made much progress on making money as much as the HFT arbitrage bots.
I am not talking about LLMs either.
Maybe look at the link.
ANNs form the basis of LLMs dude.
In any case, you’re spitballing. Its all theoretical talk without any actual algorithm of note. You’re not talking about how Wall Street is organized or what HFTs are doing, which was the point of the post at the root of this discussion.
Any time you want to read the link, let me know.
Its about Support Vector Machines (a statistical method) and ANNs (of which ChatGPT is one type of).
Did you read the link? Or did you just pick up the first hit from Google when you noticed this discussion wasn’t going the way you hoped? It doesn’t seem to have anything to do to counter my discussion point from earlier.
Also, I love that you keep telling me I’m spitballing as if I was claiming I wasn’t and I didn’t say in my very first post that it likely hasn’t happened yet.