I am since AI picked up its game (past year). quant framework but still manual execution . similar?
Ignored
All my stuff is executed automatically. All of my code and algos are proprietary, I trade my own cash in a spread betting account and copy my trades to the likes of Darwinex and FTMO. My job is system testing and software development, I use DeepSeek to code more quickly, although I've got a lot of professional software development experience which makes my prompts more effective.
Nice. One step ahead, that's where I want to end up.
I've always been proprietary stuff since this year but because of the insanity of what's possible I've done what everyone else is doing (outside retail). The models are effing amazing - all published stuff but applied properly, incredibly useful for gating et al.
Nice. One step ahead, that's where I want to end up. I've always been proprietary stuff since this year but because of the insanity of what's possible I've done what everyone else is doing (outside retail). The models are effing amazing - all published stuff but applied properly, incredibly useful for gating et al.
Ignored
Probably I should look at this "quant framework" where can I find it ?
Nice. One step ahead, that's where I want to end up. I've always been proprietary stuff since this year but because of the insanity of what's possible I've done what everyone else is doing (outside retail). The models are effing amazing - all published stuff but applied properly, incredibly useful for gating et al.
gpt the hell out of it - there aren't any good vids on it. you basically only get it served if you're using the language of geeks but essentially, hourly vol forecasting. I naively thought it'd take me a few weeks but it's hell of lot complex; daily forecast > events > seasonality > regime > hourly plus any conditioning. It's a ridiculously long pipeline. First output is a box (all these predicted ahead of time), then you can get fancy with over/undershoots as I'm currently wrestling with.
Actually makes things far harder IMO. Quantconnect provides a platform and language but it's adding an extra step in the process; AI can natively python interrogate local data which makes it way easier and less steps. I'd dabbled with QC but I think it's had its day save for HFT style algos.
gpt the hell out of it - there aren't any good vids on it. you basically only get it served if you're using the language of geeks but essentially, hourly vol forecasting. I naively thought it'd take me a few weeks but it's hell of lot complex; daily forecast > events > seasonality > regime > hourly plus any conditioning. It's a ridiculously long pipeline. First output is a box (all these predicted ahead of time), then you can get fancy with over/undershoots as I'm currently wrestling with. {image}
Ignored
so have you traded any of this stuff? Demo or otherwise?
{quote} so have you traded any of this stuff? Demo or otherwise?
Ignored
Some not all. So:
daily vol - mad useful, drives me to the most active pair ahead of it happening. Easily biggest win
regimes - fairly useful, flags up compression/expansion
hourlies- any hour soon, literally. It doesn't dovetail with my existing easily, so needs to be bloody good to get in the decision making process. working on it now
{quote} Some not all. So: daily vol - mad useful, drives me to the most active pair ahead of it happening. Easily biggest win regimes - fairly useful, flags up compression/expansion hourlies- any hour soon, literally. It doesn't dovetail with my existing easily, so needs to be bloody good to get in the decision making process. working on it now
Ignored
Can you use the ChatGPT api and then execute without manual intervention via MT4/5 for example ? Need some Python glue in the middle no doubt.
I'm thinking a Python script probably on a Linux VPS utilising the ChatGPT API to get the analysis and send signals into MT5 (or 4) although MT5 a little faster, but MT4 has more broker options
{quote} Can you use the ChatGPT api and then execute without manual intervention via MT4/5 for example ? Need some Python glue in the middle no doubt. I'm thinking a Python script probably on a Linux VPS utilising the ChatGPT API to get the analysis and send signals into MT5 (or 4) although MT5 a little faster, but MT4 has more broker options
Ignored
Theoretically yes but it'll be a guaranteed car crash. GPT isn't good enough to let it run riot on your trading decisions; it needs heavily structured, parsed data and even then... you're best off programming the logic or at the very least, give it very tight guardrails. use AI to get there sure. Not to actually do it, not yet anyway.
Witness the surge in AI EAs, absolute shit show the lot of them.
Mt5 - you need - calendar data. Essential. Python, optional, though I wish I'd used it when I started
{quote} Theoretically yes but it'll be a guaranteed car crash. GPT isn't good enough to let it run riot on your trading decisions; it needs heavily structured, parsed data and even then... you're best off programming the logic or at the very least, give it very tight guardrails. use AI to get there sure. Not to actually do it, not yet anyway. Witness the surge in AI EAs, absolute shit show the lot of them. Mt5 - you need - calendar data. Essential. Python, optional, though I wish I'd used it when I started
Ignored
So at this point you are mainly using ChatGPT to generate trading ideas essentially? Forgive me I haven't played with this so I maybe misunderstanding exactly what it is giving you
{quote} So at this point you are mainly using ChatGPT to generate trading ideas essentially? Forgive me I haven't played with this so I maybe misunderstanding exactly what it is giving you
Ignored
Got you. No, it's different
There's basically an overlay used everywhere by market makers, hedge funds et al that underpins all strategies, which is forecasting volatility and identifying regimes. It helps everything. Even a donkey of a strategy can be made usable with a bit of gating. GPT will tell you the process, Codex/Claude to code, and then you've a hell of a lot of realtime useful info that you can apply to your strategy or create one from fresh. I'm not asking GPT for trade ideas, we know where that leads, but it's quant skills are absolutely amazing. There's no chance I could have done this even 12 months back - I'm simply not that smart.
{quote} Got you. No, it's different There's basically an overlay used everywhere by market makers, hedge funds et al that underpins all strategies, which is forecasting volatility and identifying regimes. It helps everything. Even a donkey of a strategy can be made usable with a bit of gating. GPT will tell you the process, Codex/Claude to code, and then you've a hell of a lot of realtime useful info that you can apply to your strategy or create one from fresh. I'm not asking GPT for trade ideas, we know where that leads, but it's quant skills are...
Ignored
I need to have play, I agree about the advance of the LLMs in general, creative they are not, but in terms of producing stuff that is known in a usable form they are incredible. I can produce code in an afternoon that might have taken weeks to do manually, I treat DeepSeek like my own personal team of programmers, I talk to it like I might have talked to a team of coders back in the 90s, we have specification, design and code reviews I just love the productive edge it gives me, and for free no less!
same here, I'm all over it. The differential is basically the training data- quant is a known discipline so it draws on the best of academic research and spits out damn near perfect specs. Decent trading strats - especially quant ones - are not generally published so it's got nothing to go on. it'll change but for now, it's down to us.
