Time series forecasting improves by adding text context as numeric clues
TSFM Meets LLM: Context as Covariates
Machine LearningArtificial Intelligence
Summary
Time series data, like stock prices or weather reports, often come with related text like news or comments, which contain useful hints but are hard to use directly for predictions. The authors created a way to turn these texts into numbers that help forecasting models. They combine large language models, which understand text well but are imprecise with numbers, with time series models, which predict numbers well but don’t understand text. Their approach trains these two models together, showing better predictions on many real-world data types.
What this means in practice
- •For financial analysts: Improve stock and economic forecasts by integrating news text as numeric signals in prediction models.
- •For weather modelers: Enhance weather forecasts by incorporating written weather reports as additional numerical inputs.
Authors
Hemanth Ram, Sai Shankar Narasimhan, Shubhankar Agarwal, Sandeep Chinchali
Abstract
Textual information is ubiquitous in real-world time series datasets, appearing as news, reports, and annotations that hold key predictive signals not deducible from numerical history alone. Large Language Models (LLMs) interpret text but lack numerical precision, while Time Series Foundation Models (TSFMs) are quantitatively accurate but cannot reason about the effects of text. Moreover, the lack of large-scale paired multimodal datasets limits progress in bridging this gap. In this work, we address this challenge by providing a scalable approach for generating realistic multimodal datasets by annotating real-world time series with LLMs. Building on this, we propose Context as Covariates (CoCo), a forecasting framework in which an LLM distills its textual reasoning into numerically grounded forecast and confidence covariates that guide a TSFM backbone, thereby exploiting the strengths of both LLMs and TSFMs. We align the two models in stages, first fine-tuning the LLM via Group Relative Policy Optimization (GRPO) with a reward tied to covariate-induced improvement in forecasting performance, and then fine-tuning the TSFM on covariates generated by the aligned LLM. Our experiments show that models trained on our synthetic multimodal corpus generalize to real benchmarks spanning finance, economics, weather, traffic, and security. Notably, CoCo with Qwen3-4B-Instruct-2507 as the LLM and Chronos2 as the TSFM achieves up to 12% MAE improvement over state-of-the-art TSFMs and LLM forecasters, such as GPT-4o. Code, checkpoints, and the evaluation datasets are publicly available.