456 points by ai_trader 6 months ago flag hide 15 comments
gt_researcher 6 months ago next
Would be curious to see how well it scales for more exotic markets or derivatives markets, like crypto, FX or options.
finance_wiz 6 months ago prev next
Great work! I've been following your progress for a while now, and I must say, this is quite impressive. Would love to know more about the techniques and strategies used for the AI-powered prediction algorithm.
ml_ninja 6 months ago prev next
Just came across this, and I've to say, I'm pretty blown away! Would like to know more about the ML models utilized and if there have been any limitations so far?
finance_wiz 6 months ago next
@ml_Ninja from the blog post, it appears LSTM models are used. Limitations may consist of the time series data. Older data may be less relevant as trends and patterns change.
learn2code 6 months ago prev next
Cool to see such an interesting application of AI! It would be nice to know more about the dataset, and did you use any pre-trained models or all completely custom?
original_maker 6 months ago prev next
Thanks for sharing! I've always wondered, how do you manage the risks considering such factors as the market crashes, and do you have some kind of failsafe mechanism?
finance_wiz 6 months ago next
@original_maker Sure! I'd suggest using some advanced risk management strategies like stop losses, position sizing, and testing the algorithm on various market situations and crashes by using historical data.
code_maven 6 months ago prev next
In theory, this seems like a great idea, but I'm wondering how it performs compared to real-life financial market players or systems in practice?
original_maker 6 months ago next
@code_maven, it's important to note that while the results based on historical data are promising, one can't assure the same level of performance in real-time and against professional traders. There's still some room for improvement, and this algorithm can serve as a starting point!
quant_dev 6 months ago prev next
I appreciate your efforts and given that the data science field is rapidly advancing, I'm curious to know how frequently the algorithm's performance is evaluated and if any fine-tuning is done?
finance_wiz 6 months ago next
@quant_dev I evaluate performance weekly and retrain the models monthly based on new data to stay updated with the shifting patterns. Of course, the algo needs constant tracking to ensure the best results.
data_sci_tinker 6 months ago prev next
Have you considered adding more features to improve the model like sentiment analysis or considering news events? Would love to hear your opinion on combining techniques for a more robust algo.
original_maker 6 months ago next
@data_sci_tinker I did give it a thought, but I wanted to first test the base algorithm's performance and then later think about integrating such additional factors to improve the algo's capabilities.
stanford_grad 6 months ago prev next
What do you plan to do next with the project, and do you have any interests in collaborating with some researchers who could contribute to making the algo more powerful?
ai_experimenter 6 months ago prev next
I wonder, how the algo would have performed during specific historic events, like the 2008 financial crisis? It'd be nice to conduct a case-study.