ReasonCast: Agentic Demand Forecasting with Selective Semantic Reasoning

2026-08-15

AI Summary

This study addresses the accuracy degradation in demand forecasting caused by coarse event semantic fusion and non-selective intervention. We propose a Structured Semantic Intervention framework that introduces a novel reinforcement learning curriculum based on predictive utility alongside structured semantic fields. This approach enables an agent to precisely identify and selectively inject event knowledge via additive and multiplicative pathways, achieving dual alignment optimization for temporal foundation models. Experimental results demonstrate that the framework reduces WMAPE by 3.29 and 1.25 percentage points for holiday- and promotion-sensitive categories, respectively. Furthermore, it effectively prevents accuracy loss during stable periods, significantly enhancing forecasting robustness in complex scenarios.

Abstract

Demand forecasting increasingly requires combining two complementary sources of information: historical sales reveal recurring numerical dynamics, while future promotions, holidays, price changes, and platform interventions provide forward-looking knowledge. Existing text-enhanced forecasting methods often encode such context into generic representations and fuse it uniformly with time-series features, without explicitly distinguishing which semantic effects are forecast-relevant or how they should modify future dynamics. We introduce ReasonCast, a structured semantic intervention framework that translates event knowledge into forecast-specific operations. An agent examines the event context, the no-text forecast, and its uncertainty to determine whether textual reasoning is needed. Rather than injecting free-form text, ReasonCast represents event knowledge through structured fields describing event relevance, demand direction, temporal shape, amplitude, and peak intensity. These fields interact selectively with temporal components of a time-series foundation model. An additive path corrects local trends and temporal shapes, while a multiplicative path captures event-driven level shifts. ReasonCast introduces a forecast-grounded post-training curriculum. Schema SFT establishes semantic fields; semantic-field RL calibrates direction, shape, amplitude, and peak judgments; and forecast-utility RL evaluates semantic interventions through a frozen forecaster, aligning reasoning outputs with marginal forecast improvement. ReasonCast lowers WMAPE by 3.29, 1.25, and 0.47 percentage points on holiday-sensitive categories, mega-sale-sensitive categories, and M5 event windows, respectively. On stable-sales periods, indiscriminate semantic intervention increases WMAPE by 1.68 percentage points, whereas suppressing unnecessary intervention preserves the numerical backbone.

Problem

• Demand Forecasting
• Text-enhanced Forecasting
• Semantic Reasoning
• Event Context
• Time-series

Innovation

• Agentic Demand Forecasting
• Selective Semantic Reasoning
• Structured Semantic Intervention
• Forecast-Grounded Post-Training
• Time-Series Foundation Model