A novel deep learning framework for automated stock trading signal generation is proposed, leveraging Exponential Moving Average (EMA) crossover, average traded volume, and Parabolic SAR as feature inputs. The model integrates technical indicators to reduce noise (EMA), validate momentum (volume), and identify trend reversals (Parabolic SAR). These sequential features are processed by a bi-LSTM neural network, which then produces discrete trading signals. Backtesting on daily stock market data post-2020 demonstrates a statistically significant improvement in risk-adjusted returns, higher Sharpe ratios, and lower drawdowns compared to baseline strategies using EMA crossover alone. The inclusion of average traded volume helps filter false signals during low-liquidity periods, while Parabolic SAR enhances early trend reversal detection, especially in trending markets. These findings suggest that hybridizing technical indicators within deep learning architectures can yield superior automated trading performance. The proposed method offers a promising direction for algorithmic trading systems.
Adhikary et al. (2025) studied this question.