Deep Adversarial Learning for Financial Time-Series Forecasting: An RNN-Based Sentiment-Aware GAN Approach (SE-GAN)

Stock price prediction is a challenging task due to the inherent volatility of the market and the complexity of price movement. Traditional models still encounter limitations when faced with restricted historical data, which leads to over-fitting problems and poor performance on unseen data. Furthermore, market sentiment is a crucial factor for price fluctuations, yet several models still fail to capture this important factor adequately. To address the issues, the research proposes a novel deep learning network architecture. The approach leverages generative adversarial networks (GANs) to expand the training set and generalize on unseen data by generating synthetic historical data. As the GAN model generates predictions of closing prices, the generated prices can be compared with actual market movements to assess its performance. To capture the impact of market sentiment, the model also integrates sentiment analysis from financial news. This integration allows a more comprehensive technique for price forecasting, considering both quantitative historical data and qualitative sentiment indicators. The model was trained on historical data of Microsoft stock (MSFT) from January 2013 to December 2023. To conduct market sentiment analysis on financial news, FINBERT will be utilized to extract sentiment scores from headlines. Additionally, various indices of Microsoft stock (MSFT) are integrated in the training data. The proposed model integrates sentiment analysis with generative adversarial networks (GANs) to generate more robust and accurate stock price predictions. The model demonstrates the lowest Root Mean Square Error (RMSE) for the overall experiment setting compared to baselines models, including Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU), as well as TimeGPT. The improvement suggests that the SE-GAN approach offers a more effective method for predicting stock prices, leveraging the combined strengths of sentiment analysis and generative adversarial networks (GANs).