Position: Beyond Model-Centric Prediction -- Agentic Time Series Forecasting

TL;DR

This paper advocates for a shift from traditional static time series forecasting to an agentic approach that incorporates perception, planning, and reflection, enabling adaptive and interactive forecasting systems.

Contribution

It introduces the concept of agentic time series forecasting (ATSF), emphasizing an agent-based workflow that interacts with tools, learns from feedback, and adapts over time.

Findings

Proposes three implementation paradigms: workflow-based, reinforcement learning, and hybrid.

Highlights opportunities for adaptive, interactive, and continual learning in forecasting.

Discusses challenges in transitioning from model-centric to agentic approaches.

Abstract

Time series forecasting has traditionally been formulated as a model-centric, static, and single-pass prediction problem that maps historical observations to future values. While this paradigm has driven substantial progress, it proves insufficient in adaptive and multi-turn settings where forecasting requires informative feature extraction, reasoning-driven inference, iterative refinement, and continual adaptation over time. In this paper, we argue for agentic time series forecasting (ATSF), which reframes forecasting as an agentic process composed of perception, planning, action, reflection, and memory. Rather than focusing solely on predictive models, ATSF emphasizes organizing forecasting as an agentic workflow that can interact with tools, incorporate feedback from outcomes, and evolve through experience accumulation. We outline three representative implementation paradigms -- workflow-based design, agentic reinforcement learning, and a hybrid agentic workflow paradigm -- and discuss the opportunities and challenges that arise when shifting from model-centric prediction to agentic forecasting. Together, this position aims to establish agentic forecasting as a foundation for future research at the intersection of time series forecasting.