Findings & Implications

Summary of Study

This study compared ARIMA and Chronos-Bolt for stock return forecasting within the S&P 500 IT sector and evaluated the impact of sentiment analysis on portfolio optimization. The results highlight the strengths and limitations of traditional statistical models versus transformer-based architectures, providing insights into their implications for financial forecasting.

Key Findings

• Chronos-Bolt outperforms ARIMA in predictive accuracy, significantly reducing Mean Absolute Scaled Error (MASE) values. This underscores the ability of deep learning models to capture nonlinear market trends more effectively than statistical approaches.
• Sentiment analysis enhances forecasting accuracy and portfolio performance. Chronos-Bolt, when integrated with market sentiment, achieves the highest returns, demonstrating that investor sentiment serves as a valuable market signal.
• Portfolio optimization results reveal that ARIMA-based strategies suffer from excessive volatility and underperformance, while Chronos-Bolt-based models provide a more stable and profitable allocation strategy.

Implications

• AI-driven models can enhance financial decision-making. Transformer-based architectures, when combined with market sentiment analysis, offer improved market predictions, portfolio allocation, and risk management.
• Hybrid modeling approaches that integrate traditional finance models with machine learning techniques can provide better risk-adjusted returns.
• Future research can further refine predictive capabilities by incorporating additional data sources, alternative sentiment measures, and hybrid modeling strategies.

These findings emphasize the growing role of transformer-based models and AI-driven sentiment analysis in modern asset management and financial forecasting.