Abstract: Predicting future stock prices accurately can significantly help investors maximize their profits. Advances in Artificial Intelligence (AI) have shown great promise in the financial sector, despite the stock market's inherent volatility and unpredictability. To address the interconnectedness of stocks within the same industry, we propose a hybrid relational approach for predicting stock prices in the American, Indian, and Korean economies. Our method uses market data from peer companies and a data-driven feature selection approach called Random Forest Feature Permutation (RF2P) to improve prediction accuracy. The proposed hybrid prediction module, combining Temporal Convolution and Linear Model (TCLM), outperforms existing methods and results in higher profitability through a trading strategy based on the predicted results.
Abstract: This study examines current research on deep learning models for stock market forecasting, including Artificial Neural Networks, Convolution Neural Networks, Sequence to Sequence models, Generative Adversarial Networks, Graph Neural Networks, and Transformers. The study investigates datasets, characteristics, evaluation factors, and technique outcomes. Graph Neural Networks and Transformer models can better analyse dynamic and non-linear financial time series data. The study examined current finance research and suggests ways to improve prediction.
Abstract: This study aims to improve the forecasting of future price movements of equities listed on the National Stock Exchange (NSE) of India. To achieve this, the study proposes a new model called SM2PNet that incorporates the interdependence between highly correlated firms in the same industry, along with temporal dependencies. The study demonstrates that SM2PNet outperforms other deep learning methods in predicting stock price movements.