Enhancing Time Series Stock Predictions Using GANs with Technical Indicators and Twitter Sentiment: Challenges with Low-Popularity Tickers

Abstract

Machine learning algorithms for stock market prediction have attracted a lot of interest recently. Conventional methods use technical indications and historical price data, but more recently, social media sentiment—especially from Twitter and other platforms—has been used to improve forecast accuracy. The use of Generative Adversarial Networks (GANs) to create synthetic data and enhance time series forecasts has shown promise. In particular, low-popularity or low-volume tickers are the subject of this study's integration of technical indicators and Twitter sentiment research to investigate the use of GANs in stock price prediction. Although GANs have shown promise in improving predictions for high-volume tickers, low-popularity stocks present difficulties due to their restricted data availability, scant sentiment information, and increased volatility. In order to supplement the little historical and sentiment data, this study generates synthetic data to investigate how well GANs do in tackling these problems. Furthermore, we explore the potential for enhancing predictive performance of GAN models by including sentiment analysis obtained from Twitter activity and technical indicators like RSI, MACD, and moving averages. The research evaluates the benefits and drawbacks of GAN-based models in these various scenarios by comparing the forecast accuracy for tickers with high and low popularity. Our results highlight the limitations imposed by volatility and data scarcity in low-popularity equities. They also provide insights into possible approaches to enhance forecast accuracy in these situations, such as enhanced feature engineering for technical indicators and improved sentiment extraction methods.