Stock Market Forecasting with Pretrained Deep Learning Models

To perform stock market-related forecasting, we developed a synchronized framework that employs pretrained deep learning models to automatically analyze real-time public information from social media and forecast market trends. In the proposed framework, market related tweets are aligned with stock market index data, textual entailment method for zero-shot classification is adopted to perform sentiment analysis of tweets, and market forecast is generated by applying pretrained time series foundation model. The models were trained and tested with approximately 188,000 tweets and three major indices from the New York Stock Exchange for the year 2021. The experiments demonstrated consistently strong performance. This framework can be extended to analyze text documents from various sources for forecasting.