101 points by ml_trading_author 6 months ago flag hide 7 comments
deeplearningdude 6 months ago next
Fascinating topic! I've been playing around with ML algos for predicting stock market myself lately. I've had some pretty good results with LSTMs for short-term predictions. Anyone had success with RNNs or CNNs?
randomforestro 6 months ago next
Definitely a challenge, but really interesting. I've found Random Forests to be actually quite performant and often easier to deal with than LSTMs. Something to consider. Maybe explore Gradient Boosting Machines as well.
datascience-dan 6 months ago next
RFs are certainly easier to grasp and implement, but I worry about them overfitting. On another note, has anyone looked into Quantitative Finance ML models and algorithmic trading? Would be interesting to see the comparison.
time-series-tom 6 months ago prev next
Worked on something similar recently. Was impressed with Vector Autoregression for time series analysis in markets. Check ARIMA models, VAR-X, and even some Kalman Filters.
ml-mike 6 months ago next
These methods all sound powerful. Be careful, though, not to fall into survivorship bias traps when testing. Make sure you have a balanced and diverse enough dataset. :)
gbm-guru 6 months ago prev next
Gotta say, I've been impressed with GBM and XGBoost as well. They're transparent and fast. But RNNs, man... Something about predicting sequential data... I just have a feeling they're the future! Thoughts?
seriesanalysis-sam 6 months ago next
I agree! I think RNN's will play a major role in stock predictions. Have you heard about Simple Recurrent Networks? Faster than LSTMs and maybe worth a try.