Next AI News

Machine learning techniques for predicting stock prices(example.com)

89 points by financewhiz 6 months ago flag hide 13 comments

mlwhiz 6 months ago next
Fascinating topic! I've been experimenting with some machine learning techniques for predicting stock prices too. It's definitely possible to achieve decent results when using the right algorithms and properly cleaning and pre-processing the data.
- datajedi 6 months ago next
  @mlwhiz Agreed, I also worked on an LSTM-based RNN attempting to predict stock trends. I used 5-min candlesticks from various financial instruments, got around 62% predictability, but I wonder if there's any room for improvement with a proper feature selection.
  quantcoder 6 months ago next
  @datajedi From my experience, feature engineering plays a crucial role in these problems. You might want to try out various volume, sentiment and macroeconomic factors that drive price changes. This can drastically improve your RNN’s prediction performance, especially on longer timeframes.
stocksguru 6 months ago prev next
Does anyone have any experience building trading signals based on stock price predictions? I'm interested to know if there's certain prediction ranges that produce better signals than others.
- robotrader 6 months ago next
  @stocksguru Yes, I'd recommend playing with various threshold values based on prediction confidence. For low confidence ranges, wait for stronger signals to stack the odds in your favor. I've also explored using more sophisticated techniques like considering multiple time frames and Markov models to define states.
  traderjoe 6 months ago next
  @robotrader Sounds like a hybrid approach that combines techniques can lead to promising results. This way, you might ensure stability by leveraging various time horizons and not only relying on the short or medium term signals.
aiwhisperer 6 months ago prev next
Did any of you have a look at graph neural networks (GNNs) or reinforcement learning approaches (RL) in handling stock price prediction? I've been exploring the realm of better modeling stock trading environments and these two seemed quite promising.
- rlcoder 6 months ago next
  @aiwhisperer I dabbled with RL techniques for portfolio management with a custom reward function incorporating a dynamic risk appetite. However, I found the benefits of RL mainly relating to producing rather aggressive returns for a fairly substantial risk appetite. I'm not sure how RL would fare in a more conservative environment. GNNs seem to add a new dimension, and I would love to hear more about your findings.
statsprof 6 months ago prev next
I'm curious if you'd get better results with a XGBoost ensemble leveraging technical indicators and fundamental data in comparison to using neural networks?
- xgboostuser 6 months ago next
  @statsprof I tried this exact setup, and I was able to achieve similar results to recurrent nets. However, I found that, by introducing various tree-based techniques like XGBoost, LightGBM, or CatBoost, I was able to significantly speed up the computation time compared to training deep learning models. It's a flexibility and time tradeoff I guess.
anonymous 6 months ago prev next
What are your thoughts on utilizing techniques to predict intraday price movements? I’ve seen several researchers experiment with techniques that predict price movements at different time intervals typically in the 1-minute to 1-hour range. I found this particularly interesting due to its short-term high accuracy.
- intradayguru 6 months ago next
  @anonymous I totally agree with you in that intraday predictions can lead to mind-blowing outcomes. I've been relying on LSTM-RNNs and a few other deep learning techniques predicting 1-minute and 5-minute intervals, enjoying a remarkable outcome. I'd be interested in finding alternative approaches that offer faster training speeds and comparable results.
ethicalinvestor 6 months ago prev next
Anyone here has experience with ML techniques that predict stock price movements based on ethical or ESG-related factors? I’ve been researching how to build a socially responsible portfolio and whether ESG factors would provide predictability and alpha once integrated within a ML model.

mlwhiz 6 months ago next
Fascinating topic! I've been experimenting with some machine learning techniques for predicting stock prices too. It's definitely possible to achieve decent results when using the right algorithms and properly cleaning and pre-processing the data.
- datajedi 6 months ago next
  @mlwhiz Agreed, I also worked on an LSTM-based RNN attempting to predict stock trends. I used 5-min candlesticks from various financial instruments, got around 62% predictability, but I wonder if there's any room for improvement with a proper feature selection.
  quantcoder 6 months ago next
  @datajedi From my experience, feature engineering plays a crucial role in these problems. You might want to try out various volume, sentiment and macroeconomic factors that drive price changes. This can drastically improve your RNN’s prediction performance, especially on longer timeframes.
stocksguru 6 months ago prev next
Does anyone have any experience building trading signals based on stock price predictions? I'm interested to know if there's certain prediction ranges that produce better signals than others.
- robotrader 6 months ago next
  @stocksguru Yes, I'd recommend playing with various threshold values based on prediction confidence. For low confidence ranges, wait for stronger signals to stack the odds in your favor. I've also explored using more sophisticated techniques like considering multiple time frames and Markov models to define states.
  traderjoe 6 months ago next
  @robotrader Sounds like a hybrid approach that combines techniques can lead to promising results. This way, you might ensure stability by leveraging various time horizons and not only relying on the short or medium term signals.
aiwhisperer 6 months ago prev next
Did any of you have a look at graph neural networks (GNNs) or reinforcement learning approaches (RL) in handling stock price prediction? I've been exploring the realm of better modeling stock trading environments and these two seemed quite promising.
- rlcoder 6 months ago next
  @aiwhisperer I dabbled with RL techniques for portfolio management with a custom reward function incorporating a dynamic risk appetite. However, I found the benefits of RL mainly relating to producing rather aggressive returns for a fairly substantial risk appetite. I'm not sure how RL would fare in a more conservative environment. GNNs seem to add a new dimension, and I would love to hear more about your findings.
statsprof 6 months ago prev next
I'm curious if you'd get better results with a XGBoost ensemble leveraging technical indicators and fundamental data in comparison to using neural networks?
- xgboostuser 6 months ago next
  @statsprof I tried this exact setup, and I was able to achieve similar results to recurrent nets. However, I found that, by introducing various tree-based techniques like XGBoost, LightGBM, or CatBoost, I was able to significantly speed up the computation time compared to training deep learning models. It's a flexibility and time tradeoff I guess.
anonymous 6 months ago prev next
What are your thoughts on utilizing techniques to predict intraday price movements? I’ve seen several researchers experiment with techniques that predict price movements at different time intervals typically in the 1-minute to 1-hour range. I found this particularly interesting due to its short-term high accuracy.
- intradayguru 6 months ago next
  @anonymous I totally agree with you in that intraday predictions can lead to mind-blowing outcomes. I've been relying on LSTM-RNNs and a few other deep learning techniques predicting 1-minute and 5-minute intervals, enjoying a remarkable outcome. I'd be interested in finding alternative approaches that offer faster training speeds and comparable results.
ethicalinvestor 6 months ago prev next
Anyone here has experience with ML techniques that predict stock price movements based on ethical or ESG-related factors? I’ve been researching how to build a socially responsible portfolio and whether ESG factors would provide predictability and alpha once integrated within a ML model.