Nexus : An Agentic Framework for Time Series Forecasting

2026-05-14

AI Summary

Existing time series foundation models struggle to integrate unstructured textual context, while large language models (LLMs) exhibit instability and poor contextual anchoring in zero-shot forecasting. To address these limitations, this work proposes Nexus—the first framework that formulates time series prediction as a multi-agent reasoning problem. Nexus decomposes macro- and micro-level temporal dynamics and dynamically incorporates contextual information to collaboratively synthesize forecasts, without relying on external statistical anchors or fixed prompt templates. This approach substantially unlocks the intrinsic predictive capabilities of LLMs while generating interpretable analyses of driving factors. Experiments demonstrate that Nexus consistently outperforms state-of-the-art time series foundation models and strong LLM baselines on Zillow housing price and highly volatile stock datasets—both strictly postdating the LLMs’ knowledge cutoff—and produces high-quality reasoning trajectories.

Abstract

Time series forecasting is not just numerical extrapolation, but often requires reasoning with unstructured contextual data such as news or events. While specialized Time Series Foundation Models (TSFMs) excel at forecasting based on numerical patterns, they remain unaware to real-world textual signals. Conversely, while LLMs are emerging as zero-shot forecasters, their performance remains uneven across domains and contextual grounding. To bridge this gap, we introduce Nexus, a multi-agent forecasting framework that decomposes prediction into specialized stages: isolating macro-level and micro-level temporal fluctuations, and integrating contextual information when available before synthesizing a final forecast. This decomposition enables Nexus to adapt from seasonal signals to volatile, event-driven information without relying on external statistical anchors or monolithic prompting. We show that current-generation LLMs possess substantially stronger intrinsic forecasting ability than previously recognized, depending critically on how numerical and contextual reasoning are organized. Evaluated on data strictly succeeding LLM knowledge cutoffs spanning Zillow real estate metrics and volatile stock market equities, Nexus consistently matches or outperforms state-of-the-art TSFMs and strong LLM baselines. Beyond numerical accuracy, Nexus produces high-quality reasoning traces that explicitly show the fundamental drivers behind each forecast. Our results establish that real-world forecasting is an agentic reasoning problem extending well beyond only sequence modeling.

Problem

• time series forecasting
• contextual reasoning
• large language models
• foundation models
• unstructured data

Innovation

• multi-agent framework
• time series forecasting
• contextual reasoning
• large language models
• agentic reasoning