Stock price prediction is a complex task in financial analysis, traditionally addressed by statistical models or language models.
A new framework called VISTA (Vision-Language Inference for Stock Time-series Analysis) has been introduced for training-free stock forecasting.
VISTA combines textual representations of historical stock prices and line charts to predict future price values using Vision-Language Models (VLMs).
Experimental results indicate that VISTA outperforms standard baselines by up to 89.83% in stock time-series analysis, showcasing the effectiveness of multi-modal inference in financial forecasting.