Background & Research Question

Background

Stock return forecasting is essential for portfolio optimization in financial markets. Traditional models like ARIMA offer statistical rigor but struggle with nonlinear dependencies and long-term trends in financial data. Recent advancements in deep learning have introduced transformer-based models such as Chronos-Bolt, which leverage large-scale pretraining to improve forecasting accuracy. This study evaluates ARIMA and Chronos-Bolt for stock return prediction in the S&P 500 IT sector, incorporating sentiment analysis from social media to assess its impact on portfolio performance.

Research Question

Can Chronos-Bolt, when enhanced with technical indicators and sentiment-derived market signals, provide more accurate stock return predictions and improve portfolio optimization compared to ARIMA?

---