Integrating Hybrid Decomposition Methods and LSTM for Enhanced NIFTY Index Prediction and Trading Strategies

Predicting the next-day closing price of the NIFTY 50 index presents challenges due to its complexity, which complicates investment decisions. This study explores advanced methodologies—Dynamic Mode Decomposition (DMD), Long Short-Term Memory (LSTM) with Empirical Mode Decomposition (EMD)—aimed at enhancing prediction accuracy. Examination of historical data and subsequent testing showed that LSTM models, including those without decomposition (LSTM-Without) and incorporating EMD (EMD-LSTM), outperformed traditional DMD in accuracy, with EMD-LSTM notably excelling in capturing lower frequency components.For addressing the challenges posed by high-frequency components, a new hybrid model was developed, the naive-EMD-LSTM. Using this model, naive methods and LSTM are combined to enhance the trading strategy performance and improve prediction accuracy. Over a test period, naive-EMD-LSTM generated substantial returns using a simple trading strategy that appreciated from Rs. 1 to Rs. 2.31. Analysis of the trading strategy based on model predictions revealed that naive-EMD-LSTM had higher accuracy, lower return volatility, and a Sharpe Ratio of 1.17. The findings of this study highlight the potential of the naive-EMD-LSTM technique in improving the accuracy of stock market forecasts, suggesting its potential for broader use in financial predictions.