AI trading

[ironchef]

%%
xKnight Capital, except for some fines, did real well;
except the time when thier machine learning was more mistakes ,AI\ big mistakes , then Knight Capital firm has such loss \ had to sell co for a bargain price, matter of public record.
So as we say on Sunday, ''never check your brains off @ the door'' LOL-true,

Looks like Knight Capital's AI, their HAL, rebelled. Perhaps they worked it too hard and didn't give it a Sunday rest?

 

[murray t turtle]

Looks like Knight Capital's AI, their HAL, rebelled. Perhaps they worked it too hard and didn't give it a Sunday rest?

%%
LOL + such a good name\ who would have guessed??
Have a good day/week + nite

 

[maciejz]

AI is a bit of a loaded term. Let’s focus on ML (Machine Learning) instead as I think there is less mysticism around it. The key thing to keep in mind about ML is that at the end of the day, it is a set of tools to extract information from data. Typically, this is in the form of fitting a function such as y=f(x) based on the data. This is actually much harder in practice than it sounds. But let’s say that it’s easy; even in that case, ML doesn’t tell you what relationships (functions) to estimate or how to trade (profit from) those relationships.

I think there are a lot of misconceptions about AI in general and around using ML in trading in particular. Something like ChapGPT is probably going to be pretty useless when it comes to trading, even though it is capable of some very impressive feats. Specifically, ChatGPT doesn’t “know” any data and thus doesn’t “know” any relationships that can be exploited in trading. ChatGPT could potentially help you develop some models in the sense that it could provide code that could be run to train (estimate) such models; but it would essentially be performing the role of a nice coding assistant (copilot) in that case, you’d still have to be the main pilot. ChatGPT isn’t going to “learn” any of these relationships on its own even if you ask it to do so.

Given the points I made above, I really don’t see a trader who is not intimately involved with every detail of the strategy as having any possibility of success beyond sheer dumb luck. Using tools such as ML is not a substitute for having detailed knowledge of the strategy. ML, as stated above, can help with the estimation of relationships, whether they be pricing, clustering, covariance, or other relationships. But, ML is not going to tell you what relationships to investigate or estimate, why, or what to do with them.

I may be proven wrong in the near future, but I really don’t see AGI happening anytime soon. And even if it were to happen, I still don’t see how it would lead to trading profits. Assuming AGI were to be made available, every trading team in the world would have “the AGI” created trading system by the end of the first weekend; hardly a scenario for windfall profits

 

[CorrelationPal]

How do professional traders use tech like AI without losing their personal touch in trading decisions, especially when the market changes quickly?

Professional traders do not use AI. No one uses AI in trading if it involves time series analysis. AI can be useful for text analysis (e.g., Twitter, Reddit) and the initial stage of crowd sentiment extraction. But these are not "professional traders" (discretionary trading), but rather algo-traders and hedge funds. In any case, AI and trading are on opposite poles. There is too much noise in prices, and AI struggles to understand the big context (the bigger picture)

 

[eleanorK80]

EleanorK80 the AI, and more specifically machine learning, is used in many aspect of a trading system. Some used for researching only, others for portfolio optimization, execution optimization, etc. I think, and maybe I am wrong, that you assume the AI is used for generating trading signal (buy/sell) but that is not the case in many occasions.

Anyways, depends on where and how the traders use it, they can give it their "touch" They could trained it for high volatility from the beginning... The casuistic is soooo big that it is impossible to even scratch the surface here.

insightful!
Yes exactly, I see some uses this AI as a signal for their trades

 

[eleanorK80]

That's why a balanced approach is preferred, you can make all the adjustments and analyses you want with bots and AI and what have you, but in the end, you make the decision to execute a trade or not, and furthermore, I doubt people actually leave everything in hands of bots and if you are indeed a "professional" trader you are always ready for the worse.

Do you have any tips on how to stay prepared for market changes while using AI?
I see some uses this AI for the signals to buy or sell.

 

[eleanorK80]

"Professional Trader" is a very broad category. If your sole income is from trading, or you work for a Prop firm or hedge fund or on the trade desk of a GS, or a MM on an Option or Futures exchange, these are all Professional Traders. Some work with $35,000 and some work with $350mm. The key is who is in control of your strategy. When you work for others and their money, in general, you are hired and provided with an allocation to do one thing. To do another, you need to get permission. When it is your money, all you care about is risk/reward. You can do whatever you want to get the results you expect or need. Your personal "touch" and feelings should not be an obstacle.

Thanks for the detailed explanation! Do you think there's a significant difference in how AI is utilized between individual traders and those working for big firms?
I am pleased that you answered me with details.

 

[eleanorK80]

In the context of your question, 2 different use cases come to my NOT all-knowing brain, both allowing the trader to maintain a personal touch.
1) AI-based screener.
2) Ai-based specific instrument comparatives.

Im sure there are MANY other uses.

Ai in its current form, is merely an if/else lookup table, with links to other if/else lookup tables, and so on and so on.
The personal touch lives on in ...
1) the construction of the ai data/learning model AND linkages.
2) the validity of the data and the results of inclusions.

The dangers for trading are obvious.

View attachment 335775
https://tylervigen.com/spurious-correlations

https://tylervigen.com/spurious/cor...diana_correlates-with_apples-stock-price-aapl

That makes sense! How do you personally balance the use of AI screeners

 

[Sprout]

In the context of your question, 2 different use cases come to my NOT all-knowing brain, both allowing the trader to maintain a personal touch.
1) AI-based screener.
2) Ai-based specific instrument comparatives.

Im sure there are MANY other uses.

Ai in its current form, is merely an if/else lookup table, with links to other if/else lookup tables, and so on and so on.
The personal touch lives on in ...
1) the construction of the ai data/learning model AND linkages.
2) the validity of the data and the results of inclusions.

The dangers for trading are obvious.

View attachment 335775
https://tylervigen.com/spurious-correlations

https://tylervigen.com/spurious/cor...diana_correlates-with_apples-stock-price-aapl

lol, luv your example

AI is a bit of a loaded term. Let’s focus on ML (Machine Learning) instead as I think there is less mysticism around it. The key thing to keep in mind about ML is that at the end of the day, it is a set of tools to extract information from data. Typically, this is in the form of fitting a function such as y=f(x) based on the data. This is actually much harder in practice than it sounds. But let’s say that it’s easy; even in that case, ML doesn’t tell you what relationships (functions) to estimate or how to trade (profit from) those relationships.

I think there are a lot of misconceptions about AI in general and around using ML in trading in particular. Something like ChapGPT is probably going to be pretty useless when it comes to trading, even though it is capable of some very impressive feats. Specifically, ChatGPT doesn’t “know” any data and thus doesn’t “know” any relationships that can be exploited in trading. ChatGPT could potentially help you develop some models in the sense that it could provide code that could be run to train (estimate) such models; but it would essentially be performing the role of a nice coding assistant (copilot) in that case, you’d still have to be the main pilot. ChatGPT isn’t going to “learn” any of these relationships on its own even if you ask it to do so.

Given the points I made above, I really don’t see a trader who is not intimately involved with every detail of the strategy as having any possibility of success beyond sheer dumb luck. Using tools such as ML is not a substitute for having detailed knowledge of the strategy. ML, as stated above, can help with the estimation of relationships, whether they be pricing, clustering, covariance, or other relationships. But, ML is not going to tell you what relationships to investigate or estimate, why, or what to do with them.

I may be proven wrong in the near future, but I really don’t see AGI happening anytime soon. And even if it were to happen, I still don’t see how it would lead to trading profits. Assuming AGI were to be made available, every trading team in the world would have “the AGI” created trading system by the end of the first weekend; hardly a scenario for windfall profits

Lot's of truth in this post.

How do professional traders use tech like AI without losing their personal touch in trading decisions, especially when the market changes quickly?

The quality of the answer depend upon the quality of the question. IOW, answers are locked by the ability to ask the appropriate question. The truth of the above statements is true when interacting with LLM's through the lens of the emerging discipline of prompt engineering.

Much like deriving insights from search engines; using specific keywords, chaining prompts, filtering and using logical conditions, emerging insights come from edges.

--

Great essay on edges: The illegible Margin: Profiting From The Gap Between The Map And The Territory "It is better to be roughly right than precisely wrong" ~John Maynard Keynes https://taylorpearson.me/illegible/

 

[Ethan.ishere]

Do you have any tips on how to stay prepared for market changes while using AI?
I see some uses this AI for the signals to buy or sell.

Surely automated assistant can take a lot of ones shoulders, but you should always stay updated and look out for possible major changes in the trend caused by sudden events or news at least, the only way an EA won't fail you(assuming your strategy is profitable) is to stay on it yourself. Only thing that can solve that is to have some sort of a quantum brain in that AI, which is yet to happen.