Forecasting Time Series with LLMs via Patch-Based Prompting and Decomposition

TL;DR

This paper introduces PatchInstruct, a prompt-based method leveraging time series decomposition and patching techniques to enable Large Language Models to perform accurate forecasting without extensive retraining.

Contribution

The paper presents a novel prompt-based approach, PatchInstruct, that improves LLM time series forecasting by utilizing decomposition, patching, and neighbor augmentation, avoiding heavy fine-tuning.

Findings

Enhanced forecasting accuracy with minimal preprocessing.

Effective use of decomposition and patching strategies.

Maintained simplicity and flexibility in LLM-based forecasting.

Abstract

Recent advances in Large Language Models (LLMs) have demonstrated new possibilities for accurate and efficient time series analysis, but prior work often required heavy fine-tuning and/or ignored inter-series correlations. In this work, we explore simple and flexible prompt-based strategies that enable LLMs to perform time series forecasting without extensive retraining or the use of a complex external architecture. Through the exploration of specialized prompting methods that leverage time series decomposition, patch-based tokenization, and similarity-based neighbor augmentation, we find that it is possible to enhance LLM forecasting quality while maintaining simplicity and requiring minimal preprocessing of data. To this end, we propose our own method, PatchInstruct, which enables LLMs to make precise and effective predictions.