I don’t understand this constant fascination with having language models trade stocks. Language models are very useful tools but not aligned at all with generating alpha.
I use the Interactive Brokers MCP pretty heavily. I don't do any cool automatic fun "trading", but instead I use it to have "pseudo-QQQ".
I didn't like the relatively high fees for QQQ, and I realized that Invesco releases the weights for QQQ for free. I also think Tesla is too overvalued, and I want to avoid the SpaceX IPO. With the Interactive Brokers MCP, I just feed it the CSV of QQQ's weights, tell it to remove and redistribute Tesla, and then I tell it to buy "$1000 of pseudo-QQQ", in the form of raw stocks.
Doing this, I still basically get the same exposure as QQQ, without any fees.
This is absolutely and unfathomably terrible to such a great degree that I think it reinforces OPs point. It seems like using an LLM has given you the confidence to make an incredibly ill-informed decision that will cost you dearly.
Every single time you rebalance your portfolio, you will need to pay short-term capital gains taxes on any gains, as opposed to an ETF in which you simply pay for the gains when you sell your stock which can be years/decades from now.
Furthermore, assuming you rebalance your portfolio monthly, which is the minimum you need to rebalance in order to remain even somewhat aligned with QQQ, you're basically going to be paying a MINIMUM of 30-40 bucks a month in commissions to Interactive Brokers, or 400 dollars a year.
Compare that to QQQ which only costs you 18 dollars a year for every $10000 invested.
I've read some incredibly foolish investment advise on HackerNews, but I think this one just about takes the cake.
You don't need AI for this though. I was doing something like this with a python script and a crypto meta etf I created years ago. I even had some simple heuristics for selecting what coins and quantity to purchase given trading volume and spot price. Its like 175 lines of python. Probably could be a lot leaner too.
I agree I don't need it, I actually wrote a program to automatically buy and sell stuff years ago using Alpaca [1].
I just found it a bit of a pain in the ass to manage a service to do that automatically, vs thirty seconds of chatting and getting results immediately, and having something that can be supplemented by RAGs in the process.
[1] I swear I had a blog post about how I did it somewhere but I seem to have misplaced it.
It sounds like you are just pulling weights of qqq and buying based on that though. What more management do you have to do? Just pull and parse the weights wherever they might be stored, break the investment up based on that weight. Should work until the heat death of the universe.
QQQ gets the leverage from, among other things, swap agreements and futures. I don’t think what you have could be reasonably considered “pseudo-QQQ”. It’s like copying a cake recipe, but leaving out the flour and eggs because they are too expensive.
I am in the "false confidence" stage of Dunning Kruger Syndrome for finance stuff, so I personally would do swap agreements, but I'm not an average case.
I mean, even still, my point stays the same; if you have access to their strategies, I don't see why you can't just get the MCP to directly mimic that.
It’s achievable. It’s called “direct indexing”, and there are some extra costs associated with it, so for most investors, I think it is cheaper to get QQQ. You can flip that around with tax loss harvesting but I don’t understand that strategy and I can’t explain it.
You also don’t need AI to do this. Before AI, the main barrier to direct indexing was the amount of capital you need. That is still true.
I have enough capital to where I can do everything with the incremental share threshold of Interactive Brokers; as such I don't have to deal with the fees associated with normal direct indexing.
I feel like you could probably have the AI write a script that uses the API to do the same thing, except this time you have code you can test rather than relying on the probabilistic machine every time you do a trade.
I don't let it buy anything without confirming, and I will load the CSV into Google Sheets to make sure that the numbers more or less correspond to what I think they will. It's just easier to directly use the MCP and set up some custom skills for what I want to do.
I have thought about this but snag on rebalancing, because it would create a taxable event, or be drawn out over months/years.
Although maybe a bit spicier, VGT is half the cost of QQQ, so that is what my "NASDAQ" has been. I also blend in VTI to cut the volatility a bit, which is 1/3 the cost of VGT.
I'm doing the same strategy for rebalancing that QQQ does, and I figure that the headache of tax time is a "Tom in 11 months from now"'s problem :)
Some tax software nowadays will allow you to simply upload the tax documents with all the transactions and it will tabulate everything for you, so I don't think it will be too hard for me.
I'll admit that there's primarily just kind of a coolness factor to be able to say that I ripped off and copied QQQ without any fees, but I do genuinely like the idea that I can avoid companies that I think are terrible in the process.
Alpha is ultimately the result of analysis, of better analysis than others.
LLM's can actually be exceptionally good at research and pattern recognition, i.e. analysis. And while they aren't great at running numbers themselves, they can do exceptional work passing off Python scripts to an interpreter to generate the numerical results they need.
I'm quite sure the Robinhood AI is going to be trash, i.e. just a gimmick.
But, it's not crazy to think that with the right harness, there are big opportunities for identifying profitable strategies. Especially relying on unparalleled and essentially unlimited research capacity based on public information. More analysis than any single firm could ever hire.
And even for Robinhood users, it's entirely plausible that AI-traded stocks will perform much better than the trades a majority of users would make, since most investors are really unsophisticated.
>LLM's can actually be exceptionally good at research and pattern recognition, i.e. analysis.
No they aren't, they're good at imitating analysis based on representations of analysis in their training data. Also, Its likely that out dated techniques would over represented in training data.
Do you think Jane Street would have the returns they do if they just imitated all their competitors and everyone was using the same strategies?
They’re possibly great at generating alpha in highly complex systems that compose LLMs with tabular machine learning and other analytical techniques at a large scale. So yea, certainly not for these users.
An LLM may be bad at trading stocks, but an LLM may be good at analyzing the wider context, like the news feed, to inform automated trading driven by a more sophisticated model, called by the LLM as a tool.
I don't think that this contraption should necessarily perform tolerably, but the use of an LLM is not necessarily a wrong move.
As much as I hate the idea of enabling the desperate masses to gamble like this, LLMs are very aligned tools for sentiment analysis, which can be the foundation of a trading strategy. I think it's extremely irresponsible to use them for execution, though.
That is sidestepping the point: 70-90% of retail traders lose money. The question should be: is AI trading enough of an improvement to justify its non-subsidized costs?
Just like any useful tools, there would be an expert super tool user who could probably generate enough profit based on the tools. The majority would not profit from it in the long run (the monopolistic tool makers would reap any profit from the value chain.)
AI agents for trading, as well as 24/7 trading are no different than offering sports gambling and prediction markets to the masses; it is a vacuum for the fiat of the unsophisticated. The goal is more trading volume to generate more fees, similar story with private equity wanting access to 401ks to unload PE at peak valuations to bag holders.
Cryptoscammers everywhere rubbing their hands together. There's so many ways this could go wrong. Everything from prompt injection, to being tricked in running a specific scammers setup, to which they can pump and dump specific stocks, and all sorts of other manmade horrors.
I was definitely wondering this. As I understand it they make money on order flow and don't charge for transactions (is that right?). But allowing LLMs to trade dilutes the true information in the order flow.
On the other hand maybe it's just chasing trends, like their previous forays into blockchains. It pays because it keeps their name in the news.
There needs to be a ton of regulation of this eventually. this will not be a problem from a safety perspective today, but a smarter-than-human agent trained on long-horizon tasks should not be given access to influence the market unless this is done very carefully.
I was a fan of Robinhood's mission of democratizing finance and prioritizing UX for casual traders. They seem to jump on every hype train though, crypto, prediction markets, now agentic trading, whether it is ethical or not or good for their customers or not, and it seems like the distance between "democratizing finance" and "finding new suckers" is closing. Disappointing but not surprising.
Robinhood's crypto offering is extremely deceptive. They offer "commission-free cryptocurrency trading" but don't make it clear that you pay a 0.95% fee[1] on every trade (technically a 'spread' and not a 'commission' but there is hardly a difference). They also take 25% of staking rewards. These are absurdly high fees.
They were adopted by professionals long ago, and those highly tuned and validated proprietary models are going to kick the butt of the models that you have access to every day of the week.
If there was anything missing from the average American’s economic wellbeing, it was the ability to create bespoke financial products to scalably make bets against informed professional traders while they sleep.
Quite ironic. The original Robin Hood took from the rich and gave to the poor. Robinhood, the app, seems to do the exact opposite: it helps the rich get richer at the expense of regular folk.
This should be limited to giving advice (education, warning, explicit consent), unless there's harm to third parties.
Because, you know, certain actions and even thoughts can lead to eternal damnation in Hell, according to what a society may think. Would you prefer the society to hold you off from that?
A child is not a fully autonomous person. I would of course take the loaded gun from the child, unload it, and explain its dangers to the child.
Money, in any form, may be as dangerous as a loaded gun, trading stocks or not. Most adults are careful with money, as they are with loaded guns. The problem is that some parties may try to make trading stocks (even leveraged) look much easier and safer than it is. It's like giving somebody a real loaded gun, while making it look like a toy gun, safe even for a child. And this of course needs to be regulated: not the trading, but the disclosure. This is not a toy.
Who are you to decide that a child is not a fully autonomous person? Sounds like you're imposing a normative rule based on societally derived presuppositions of right and wrong, which is exactly the point. We're just haggling over where the line should be drawn and you think it should be drawn somewhere further back than others do, but there's no truth to be found here.
This was always the Robin Hood play (versus being a grown up brokerage), they are simply griftmaxxing now in a "low regulatory environment". Like Coinbase, they need volume to succeed economically, not buy and hold investors. Crypto volume is down, so Coinbase revenue is down. Young people have little to no cashflow, but they have high intent to gamble in a crushing and financially nihilistic macro, which Robin Hood serves.
I disagree, AI agents could help level the playing field. Citadel doesn't have any AI models that are better than what you or I have. Market data is more accessible than ever. As LLMs get better at trading, the difference in capability between you and a professional trader gets smaller.
Also, Claude knows about a lot of the traps that consumers can fall into: spread, execution, risk concentration, etc. -- high chance that if I tell Claude I'm thinking of going all in on AMC because some Reddit post told me to, it'll say "slow down cowboy"
What is the point of having a speculative market if everyone has access to the same information and capabilities? You might as well just direct deposit a proportional share of all economic growth relative to investment into every citizens account and be done with it.
Will it be is a different thing though. And if it’s not, who exactly is accountable?
With funds and portfolio managers that run them, there’s a clear accountability model (if the fund sucks, the manager loses their job and the company loses credibility)
With AI agents doing the management, who is accountable when the fund sucks? If it’s the customer, we’ve moved accountability from someone who at least in theory, knows what they’re doing to someone who has little to no clue.
You have to be accountable for what you have the model do on your behalf. I hear what you're saying, but there are also issues with the hedge fund accountability model. There are certainly swaths of fund managers who are only there because they got lucky or had the right pedigree, and more that are better traders but never became a fund manager because they got unlucky or had other passions.
An individual investor can invest with their risk appetite on their time horizon and not be subject to Citadel's "5% draw down in a quarter and you're fired" culture which can be toxic to returns over time.
I believe that your individual ability to execute an order is constrained such that some of the difference is removed. On the other hand, the overall thesis has merits IMHO
Even better for America's well being will be if thousands of individual investors have identical or near identical bots for sophisticated financial institutions to exploit while they sleep.
And not just informed professional traders -- also insiders with privileged information about world events that let them trade before the news hits. Now AI agents are going to be chasing phantom signals that look like they might be evidence of an insider's move.
LOL. This is the outcome when a Product Manager sits there and says "You know, people just aren't losing enough money on sports betting and gambling apps. How can we fix this?"
This was not due to malice but instead, incompetence. They didn’t have enough cash to clear their trades.
I have ranted on here before about the SV startup mindset of “I don’t need to know anything about the industry I’m ‘disrupting’ nor do I need to play by their rules” and this was an example of that. On that day, everybody who was actually in capital markets went, “what f-ing idiots those guys are”
I didn't like the relatively high fees for QQQ, and I realized that Invesco releases the weights for QQQ for free. I also think Tesla is too overvalued, and I want to avoid the SpaceX IPO. With the Interactive Brokers MCP, I just feed it the CSV of QQQ's weights, tell it to remove and redistribute Tesla, and then I tell it to buy "$1000 of pseudo-QQQ", in the form of raw stocks.
Doing this, I still basically get the same exposure as QQQ, without any fees.
Every single time you rebalance your portfolio, you will need to pay short-term capital gains taxes on any gains, as opposed to an ETF in which you simply pay for the gains when you sell your stock which can be years/decades from now.
Furthermore, assuming you rebalance your portfolio monthly, which is the minimum you need to rebalance in order to remain even somewhat aligned with QQQ, you're basically going to be paying a MINIMUM of 30-40 bucks a month in commissions to Interactive Brokers, or 400 dollars a year.
Compare that to QQQ which only costs you 18 dollars a year for every $10000 invested.
I've read some incredibly foolish investment advise on HackerNews, but I think this one just about takes the cake.
I just found it a bit of a pain in the ass to manage a service to do that automatically, vs thirty seconds of chatting and getting results immediately, and having something that can be supplemented by RAGs in the process.
[1] I swear I had a blog post about how I did it somewhere but I seem to have misplaced it.
I am in the "false confidence" stage of Dunning Kruger Syndrome for finance stuff, so I personally would do swap agreements, but I'm not an average case.
Still, as you said, just mimicking regular QQQ is achievable.
You also don’t need AI to do this. Before AI, the main barrier to direct indexing was the amount of capital you need. That is still true.
I don't let it buy anything without confirming, and I will load the CSV into Google Sheets to make sure that the numbers more or less correspond to what I think they will. It's just easier to directly use the MCP and set up some custom skills for what I want to do.
Dunno, it seems to work fine.
Although maybe a bit spicier, VGT is half the cost of QQQ, so that is what my "NASDAQ" has been. I also blend in VTI to cut the volatility a bit, which is 1/3 the cost of VGT.
Some tax software nowadays will allow you to simply upload the tax documents with all the transactions and it will tabulate everything for you, so I don't think it will be too hard for me.
I'll admit that there's primarily just kind of a coolness factor to be able to say that I ripped off and copied QQQ without any fees, but I do genuinely like the idea that I can avoid companies that I think are terrible in the process.
LLM's can actually be exceptionally good at research and pattern recognition, i.e. analysis. And while they aren't great at running numbers themselves, they can do exceptional work passing off Python scripts to an interpreter to generate the numerical results they need.
I'm quite sure the Robinhood AI is going to be trash, i.e. just a gimmick.
But, it's not crazy to think that with the right harness, there are big opportunities for identifying profitable strategies. Especially relying on unparalleled and essentially unlimited research capacity based on public information. More analysis than any single firm could ever hire.
And even for Robinhood users, it's entirely plausible that AI-traded stocks will perform much better than the trades a majority of users would make, since most investors are really unsophisticated.
No they aren't, they're good at imitating analysis based on representations of analysis in their training data. Also, Its likely that out dated techniques would over represented in training data.
Do you think Jane Street would have the returns they do if they just imitated all their competitors and everyone was using the same strategies?
I don't think that this contraption should necessarily perform tolerably, but the use of an LLM is not necessarily a wrong move.
https://en.wikipedia.org/wiki/Parable_of_the_broken_window
https://www.bogleheads.org/wiki/Getting_started
Obviously how much the average user will profit / compile debt from this change is a lot more variable.
On the other hand maybe it's just chasing trends, like their previous forays into blockchains. It pays because it keeps their name in the news.
This feels like when everything became webified for no reason, or everyone added features like 3D TVs that were clearly not necessary.
This is only about removing friction for the non-professionals to rapidly burn their money...
1. https://cdn.robinhood.com/assets/robinhood/legal/rhc-fee-sch...
If you’d like to make dubious trades that’s your prerogative and who am I to stop you.
Because, you know, certain actions and even thoughts can lead to eternal damnation in Hell, according to what a society may think. Would you prefer the society to hold you off from that?
If you see a child playing with a loaded gun, you won’t stop it?
Money, in any form, may be as dangerous as a loaded gun, trading stocks or not. Most adults are careful with money, as they are with loaded guns. The problem is that some parties may try to make trading stocks (even leveraged) look much easier and safer than it is. It's like giving somebody a real loaded gun, while making it look like a toy gun, safe even for a child. And this of course needs to be regulated: not the trading, but the disclosure. This is not a toy.
https://www.npr.org/2026/04/05/nx-s1-5762276/teens-getting-h...
https://kyla.substack.com/p/gen-z-and-financial-nihilism
https://web.archive.org/web/20240226104327/https://youngmone...
https://web.archive.org/web/20240226104327/https://coinmarke...
Also, Claude knows about a lot of the traps that consumers can fall into: spread, execution, risk concentration, etc. -- high chance that if I tell Claude I'm thinking of going all in on AMC because some Reddit post told me to, it'll say "slow down cowboy"
Will it be is a different thing though. And if it’s not, who exactly is accountable?
With funds and portfolio managers that run them, there’s a clear accountability model (if the fund sucks, the manager loses their job and the company loses credibility)
With AI agents doing the management, who is accountable when the fund sucks? If it’s the customer, we’ve moved accountability from someone who at least in theory, knows what they’re doing to someone who has little to no clue.
An individual investor can invest with their risk appetite on their time horizon and not be subject to Citadel's "5% draw down in a quarter and you're fired" culture which can be toxic to returns over time.
Maybe if you prompt it to be highly critical of you, the user.
Otherwise it will absolutely right you out of money.
I am sure there are some very happy people in the larger firms due to this news.
I have ranted on here before about the SV startup mindset of “I don’t need to know anything about the industry I’m ‘disrupting’ nor do I need to play by their rules” and this was an example of that. On that day, everybody who was actually in capital markets went, “what f-ing idiots those guys are”
https://en.wikipedia.org/wiki/GameStop_short_squeeze
Will be waiting for the notice to say that 70% of users lose money to now 90% of users lose money.