---
title: DualCast: A Dual-Path Language Model for Bimodal Financial Time-Series Forecasting
url: https://www.ml-quant.com/papers/arxiv/2609.38197/
site: ML-Quant (https://www.ml-quant.com)
updated: 2026-10-02
license: Summaries CC BY 4.0; links go to the original sources
index: https://www.ml-quant.com/llms.txt
identifier: arXiv:2609.38197
source_url: https://arxiv.org/abs/2609.38197
featured: 2026-10-02
citations: unknown
topic: Econometrics & Forecasting
---


# DualCast: A Dual-Path Language Model for Bimodal Financial Time-Series Forecasting

Dual-path language model extended with financial tokens and news-conditioned revision achieves lowest error in 8 of 12 equity and energy forecasting settings across multiple horizons.

- Source: https://arxiv.org/abs/2609.38197
- Identifier: arXiv:2609.38197
- Released: 2026-10-01
- First featured: Quant Letter No. 133 (2026-10-02): https://www.ml-quant.com/issues/2026-10-02/
- Citations (Semantic Scholar): not tracked
- Published in: not yet
- Topic: Econometrics & Forecasting
- Authors: Wentao Zhao, Hongqiang Wu, Shanghang Liu, Zhaochen Zan, Yu Zhang, Biqing Huang

## Abstract (arXiv, CC0)

Financial time-series forecasting must capture price dynamics across heterogeneous assets while incorporating news available at prediction time. We introduce DualCast, a dual-path framework that extends a frozen language model with a discrete financial vocabulary. Each log-return patch is represented by a learned summary token and three residual shape tokens, preserving local drift and volatility while allowing shape patterns to be shared across assets. To improve codebook utilization, we develop adaptive frequency-equalizing residual vector quantization, which rebalances overloaded codewords without compromising reconstruction accuracy. The fast path trains only the new financial-token embeddings and output heads on a frozen Qwen3-8B backbone. A toggleable LoRA adapter enables a slow path that conditions on the fast forecast and news available at the forecast origin to produce a revised prediction. The reviser is initialized by supervised fine-tuning and further optimized with a return-space group relative policy optimization objective that rewards improvements over the fast forecast. In zero-shot evaluations covering equities and energy prices at five-minute, daily, and weekly resolutions, the slow path achieves the lowest mean absolute percentage error among the compared methods in 8 of 12 dataset-horizon settings, including every longest-horizon setting. News ablations indicate additional gains in most tested settings, although their magnitude varies across markets. DualCast thus combines a fast numerical forecaster with an optional text-conditioned revision mechanism.

## Related

- [TKAN: Temporal Kolmogorov-Arnold Networks](https://www.ml-quant.com/papers/ssrn/4825654/): The article presents Temporal Kolomogorov-Arnold Networks (TKANs), a new neural network design that merges the benefits of Recurrent Neural Networks and Long Short-Term Memory for improved multistep time series forecasting.
- [Robust agents learn causal world models](https://www.ml-quant.com/papers/arxiv/2402.10877/): The research suggests that intelligent agents must learn an approximate causal model to generalize to new domains, impacting fields like transfer learning and causal inference.
- [Directional-shift Dirichlet ARMA models for compositional time series with structural break intervention](https://www.ml-quant.com/papers/arxiv/2601.16821/): The article introduces a new Bayesian model that analyzes compositional time series data, effectively handling structural breaks and enhancing forecasting accuracy during these changes.
- [FinCARE: Financial Causal Analysis with Reasoning and Evidence](https://www.ml-quant.com/papers/arxiv/2510.20221/): KG+LLM for Financial Causal Discovery: Combines SEC knowledge graphs, LLM reasoning, and causal discovery to build more accurate finance‑grounded causal models.
- [FinZero: Launching Multi-modal Financial Time Series Forecast with Large Reasoning Model](https://www.ml-quant.com/papers/arxiv/2509.08742/): Multi-modal Forecasting: FinZero, a pre-trained model fine-tuned by the Uncertainty-adjusted Group Relative Policy Optimization method, is introduced in the article, enhancing the accuracy, adaptability, and scalability of financial time series forecasting.
- [FinCast: A Foundation Model for Financial Time-Series Forecasting](https://www.ml-quant.com/papers/arxiv/2508.19609/): Time-Series Forecasting Model: FinCast, a new model for financial time-series forecasting, outperforms existing methods by effectively capturing diverse patterns without needing domain-specific adjustments.