Thanks! The way these systems work is that they are portfolio allocation problems. So the equity curve you see is accumulated multiples (1 -> 2 indicates 100% growth).
A 'position' is a set of weights that adds up to 1 such that each time step describes how the proportion of your total capital you're putting into each stock.
Example:
You're trading AAPL, MSFT, GOOG. You start with 25% in AAPL, 25% in MSFT, and 50% in GOOG. Your position will be [0.25, 0.25, 0.5] as it adds up to one.
These algorithms take your current position and modify it based on the rules. So, one strategy might say 'bump up allocation into MSFT' at which point it would proportionally adjust the weights.
If the algorithm weights MSFT more heavily or 'predicts' that we should by 10%, your weights would then look like 20% in AAPL, 35% in MSFT, 40% in GOOG, or [0.2, 0.35, 0.4].
Thanks man! But yeah, I wrote some hardcore GP code for the first preliminary version of this and realized that the LLMs are actually a better route.
Now, I have a ton of potential 'operators' and 'primitives' because I know which ones work. I think from here I could create a GP system and have more success. I also was trying to search across a scope that was just way too large. I did find some really interesting systems, however.
Love it. The alpha mining approach makes me think of me SQX (strategy quant) I used this a lot some years back, but with some LLM flavor - right up my alley 😎. Thanks for the consistently good content!
Excellent work; I have done somewhat simpler loops before the age of AI, and they added value as learning foundations.
I have some ideas where you could take this for likely much better alpha, if you want to chat at some point.
👍🏼👍🏼
Amazing content haha🤙
I’m also curious on how u set position size for ur strategy, look forward to that !
Thanks! The way these systems work is that they are portfolio allocation problems. So the equity curve you see is accumulated multiples (1 -> 2 indicates 100% growth).
A 'position' is a set of weights that adds up to 1 such that each time step describes how the proportion of your total capital you're putting into each stock.
Example:
You're trading AAPL, MSFT, GOOG. You start with 25% in AAPL, 25% in MSFT, and 50% in GOOG. Your position will be [0.25, 0.25, 0.5] as it adds up to one.
These algorithms take your current position and modify it based on the rules. So, one strategy might say 'bump up allocation into MSFT' at which point it would proportionally adjust the weights.
If the algorithm weights MSFT more heavily or 'predicts' that we should by 10%, your weights would then look like 20% in AAPL, 35% in MSFT, 40% in GOOG, or [0.2, 0.35, 0.4].
Hope that's clear!
I seee, thanks alot!
Happy to help!
Thanks for sharing...very interesting strategy generation pipeline....
Remind me of my previous experience of using genetic programming to automate alpha mining, though gp is usually highly uninterpretable...
Thanks man! But yeah, I wrote some hardcore GP code for the first preliminary version of this and realized that the LLMs are actually a better route.
Now, I have a ton of potential 'operators' and 'primitives' because I know which ones work. I think from here I could create a GP system and have more success. I also was trying to search across a scope that was just way too large. I did find some really interesting systems, however.
Maybe I'll get back to it...
Can anybody please provide a selling strategy recommendation or guidance for Indian markets?
Another Fantastic Tool for Retail Traders is Here!
https://vikopine.substack.com/p/another-fantastic-tool-for-retail
Join for LIVE Trading and regular market updates in Substack chat room.
https://substack.com/chat/894242
Love it. The alpha mining approach makes me think of me SQX (strategy quant) I used this a lot some years back, but with some LLM flavor - right up my alley 😎. Thanks for the consistently good content!
Hi, looking forward to testing some out... subscribed a couple days ago and no access to the Google Drive yet. Check messages please.
Hey I think I added you to the drive this morning. Can you double check?
I'm logged into google via that gmail.com addy I sent you, but the link still tells me Access Denied.
And... thanks so very much, the invite came through, the email verification went smoothly and I'm in.
Great! Glad you got it