LSTPrompt: Large Language Models as Zero-Shot Time Series Forecasters by Long-Short-Term Prompting

TL;DR

LSTPrompt introduces a novel zero-shot prompting method for large language models that decomposes time-series forecasting into sub-tasks, improving adaptability and performance over existing prompting strategies.

Contribution

The paper proposes LSTPrompt, a new prompting approach that decomposes TSF into sub-tasks and guides LLMs to reassess forecasting, enhancing zero-shot TSF performance.

Findings

LSTPrompt outperforms existing prompting methods in zero-shot TSF tasks.

LSTPrompt achieves competitive results compared to specialized TSF models.

Decomposing TSF into sub-tasks improves model adaptability and accuracy.

Abstract

Time-series forecasting (TSF) finds broad applications in real-world scenarios. Prompting off-the-shelf Large Language Models (LLMs) demonstrates strong zero-shot TSF capabilities while preserving computational efficiency. However, existing prompting methods oversimplify TSF as language next-token predictions, overlooking its dynamic nature and lack of integration with state-of-the-art prompt strategies such as Chain-of-Thought. Thus, we propose LSTPrompt, a novel approach for prompting LLMs in zero-shot TSF tasks. LSTPrompt decomposes TSF into short-term and long-term forecasting sub-tasks, tailoring prompts to each. LSTPrompt guides LLMs to regularly reassess forecasting mechanisms to enhance adaptability. Extensive evaluations demonstrate consistently better performance of LSTPrompt than existing prompting methods, and competitive results compared to foundation TSF models.