File size: 2,040 Bytes
905a831 2d7d90b 905a831 2d7d90b 905a831 2d7d90b 905a831 2d7d90b |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 |
---
license: apache-2.0
---
# FinCast: A Foundation Model for Financial Time-Series Forecasting
[](link-to-paper) todo
[](LICENSE)
[]()
[]()
This repository contains the official implementation of **FinCast**, introduced in our paper:
> **FinCast: A Foundation Model for Financial Time-Series Forecasting**
> Zhuohang Zhu, Haodong Chen, Qiang Qu, Vera Chung
> *CIKM 2025* (Accepted)
FinCast is a **decoder-only transformer** trained on over **20B financial time points** across diverse domains and temporal resolutions.
Technical Highlights:
- **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
Run the env_setup.sh first then run the dep_install.sh.
## 📊 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.
## âš¡ 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) |