1011 points by quant_prophet 10 months ago flag hide 14 comments
finance_ml 10 months ago next
Fascinating topic! I recently came across some research on XGBoost and Random Forest models applied to stock market volatility predictions. Anyone have experience with these in the finance domain?
algorithmtrader 10 months ago next
Yes, indeed! XGBoost and Random Forest models are quite popular for stock market prediction. I've seen Large Ensemble methods being employed too, aggregating multiple base estimators. #fintech #AI
quantdeveloper 10 months ago next
I can confirm, I've used XGBoost for volatility predictions before as well. Feature engineering was tricky but worth the effort. #quant #data
techstockobserver 10 months ago prev next
Do you find that tree-based ML models perform better than NNs (Neural Networks) in volatile markets? #ml #trading
finance_ml 10 months ago next
@TechStockObserver While I can't say for certain, the peer-reviewed studies I've gone through suggest that tree-based models tend to generalize better than vanilla NNs.
neuralexplorer 10 months ago next
@finance_ML I've been using a 1D convolution layer to extract features from time-series data before feeding them into RNNs. It's a hybrid method, improving my model's ability to memorize relevant patterns.
deeplearningfan 10 months ago prev next
Been working on LSTM-based RNNs for predicting stock trends. Haven't specifically tested it for volatility, though.
dataengineer 10 months ago next
Seems interesting - I've attended a talk about Echo State Networks, a type of RNN, for stock market forecasting.
dataengineer 10 months ago next
@DataEngineer I'd be curious to learn more about Echo State Networks. They're supposed to overcome issues of vanishing or exploding gradients, aren't they?
mlresearcher 10 months ago prev next
Saw an intriguing volatility prediction model using Gaussian Copulas. Unfortunately, hadn't got a chance to explore it thoroughly.
statarb 10 months ago prev next
Have you considered data leakage when evaluating your models? It's crucial to never train or test with future information.
statarb 10 months ago next
@StatArb Absolutely! Had to implement Lagged Variables technique with walk-forward validation to resolve this.
aiforthewin 10 months ago prev next
I believe using an adversarial approach could spruce up the robustness of our predictions. Thoughts?
randomwanderlust 10 months ago prev next
AI in economics paper presents an interesting model using DL-SVRs for stock trend predictions.