ICCV 2025 Workshops

MORFI: Multimodal Zero-Shot Reasoning for Financial Time-Series Inference

Tina Khezresmaeilzadeh, Parsa Razmara, Mohammad Erfan Sadeghi, Seyedarmin Azizi, Erfan Baghaei Potraghloo

University of Southern California

MORFI pipeline: a price history shown as a chart and as numbers, read by a VLM to forecast the next five days

Abstract

MORFI is a training-free framework that uses vision-language models (VLMs) to predict stock prices by combining line-chart images with numerical time-series data. Prompting the VLM with both modalities, optionally with chain-of-thought reasoning, outperforms text-only LLMs by 40–80% in MSE.

Citation

@InProceedings{Khezresmaeilzadeh_2025_ICCV,
  author    = {Khezresmaeilzadeh, Tina and Razmara, Parsa and
               Sadeghi, Mohammad Erfan and Azizi, Seyedarmin and
               Potraghloo, Erfan Baghaei},
  title     = {MORFI: Mutimodal Zero-Shot Reasoning for Financial
               Time-Series Inference},
  booktitle = {Proceedings of the IEEE/CVF International Conference
               on Computer Vision (ICCV) Workshops},
  month     = {October},
  year      = {2025},
  pages     = {4236-4245}
}