LoFT-LLM: Low-Frequency Time-Series Forecasting with Large Language Models

2025-12-22

AI Summary

Addressing the challenge of time-series forecasting under low-data, high-noise, and highly dynamic conditions in finance and energy domains, this paper proposes a frequency-aware forecasting framework. Methodologically, it introduces the novel “low-frequency-first” paradigm: stable low-frequency trends are extracted via spectral patching, while residual networks model high-frequency noise; auxiliary variables and structured domain knowledge are integrated through fine-tuned large language models (LLMs), supported by joint time-frequency modeling and semantic calibration mechanisms. The key contributions include effective decoupling of trend and noise components, enhanced few-shot robustness, and improved interpretability. Extensive experiments on multi-source financial and energy datasets demonstrate that our method consistently outperforms state-of-the-art baselines—under both full-data and few-shot settings—in prediction accuracy, noise resilience, and interpretability.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Planning, Routing, and Scheduling: Planning with Language Models

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

Time-series forecasting in real-world applications such as finance and energy often faces challenges due to limited training data and complex, noisy temporal dynamics. Existing deep forecasting models typically supervise predictions using full-length temporal windows, which include substantial high-frequency noise and obscure long-term trends. Moreover, auxiliary variables containing rich domain-specific information are often underutilized, especially in few-shot settings. To address these challenges, we propose LoFT-LLM, a frequency-aware forecasting pipeline that integrates low-frequency learning with semantic calibration via a large language model (LLM). Firstly, a Patch Low-Frequency forecasting Module (PLFM) extracts stable low-frequency trends from localized spectral patches. Secondly, a residual learner then models high-frequency variations. Finally, a fine-tuned LLM refines the predictions by incorporating auxiliary context and domain knowledge through structured natural language prompts. Extensive experiments on financial and energy datasets demonstrate that LoFT-LLM significantly outperforms strong baselines under both full-data and few-shot regimes, delivering superior accuracy, robustness, and interpretability.

Problem

• Forecasting with limited data and noisy dynamics
• Underutilizing auxiliary variables in few-shot settings
• Integrating low-frequency trends and semantic calibration

Innovation

• Low-frequency trend extraction from spectral patches
• Residual learner models high-frequency variations
• LLM refines predictions with auxiliary context