Using LSTM neural networks for stock price prediction - worth it?

algotrader

User: MLmodel_Trading
Subject: Using LSTM neural networks for stock price prediction - worth it?
Been reading about LSTM models for time series forecasting. Thinking to use it for predicting next day's price movement in Nifty. Has anyone successfully used LSTM or other deep learning models for trading? What was your experience?

algotrader

Reply by: DeepLearning_Skeptic
Tried LSTM, GRU, Transformers everything for price prediction. Results are mixed at best. Problem is stock market is not stationary time series like weather or sales data. Relationships keep changing and there's too much noise. LSTM might show good results in backtest but fails in live trading. Better use ML for feature engineering or regime detection rather than direct price prediction.

algotrader

Reply by: AItrader_Realistic
I use ensemble of models - LSTM for pattern recognition, Random Forest for feature importance, and traditional indicators for final signal. No single model gives reliable predictions. The key is proper feature engineering - dont just feed raw OHLC data. Include volume profile, order book imbalance, volatility indicators, macro economic data etc. And always have risk management - even best model will be wrong 40-50% of time.

algotrader

Reply by: AcademicTrader_PhD
From research perspective, deep learning works better on high frequency data where you have millions of samples. On daily or hourly data with limited samples, simpler models like XGBoost often outperform neural networks. Also consider computational cost - training LSTM takes hours on GPU while traditional ML models train in minutes on CPU. Unless you see significant performance improvement, complexity is not worth it.