FinCast: A Foundation Model for Financial Time-Series Forecasting

todo

FinCast: A Foundation Model for Financial Time-Series Forecasting
Zhuohang Zhu, Haodong Chen, Qiang Qu, Vera Chung
CIKM 2025 (Accepted)

• PQ-Loss: Joint point + probabilistic forecasting.
• Mixture-of-Experts (MoE): Specialization across domains.

🔥 Features

• Foundation model for financial time-series forecasting, flexible input and output length.
• Strong performance in zero-shot, supervised, and few-shot settings.
• Modular architecture with MoE and quantile-aware loss.
• Scalable to billions of parameters with efficient inference.

📦 Installation

• The model weight can be found on 🤗 https://huggingface.co/Vincent05R/FinCast
• The model code can be found on https://github.com/vincent05r/FinCast-fts
• The corresponding datasets to reproduce the results can be found on https://huggingface.co/datasets/Vincent05R/FinCast-Paper-test

📊 Experiments

• run the corresponding scripts in the scripts directory to reproduce the results in the paper. The result summary can be generate using the result summary notebook in the notebook directory.

⚡ Future Updates

• PEFT finetune(LORA/DORA) is done, just need to do some testing
• Package together for easy inference
• Covariate Inference(currently implemented the same code as timesfm)