MultiCast: Zero-Shot Multivariate Time Series Forecasting Using LLMs

TL;DR

MultiCast leverages large language models for zero-shot multivariate time series forecasting by introducing novel token multiplexing and quantization techniques, achieving competitive accuracy and efficiency on real-world datasets.

Contribution

This work presents the first zero-shot LLM-based method for multivariate time series forecasting, with innovative token multiplexing and quantization schemes.

Findings

Competitive RMSE performance on real datasets

Reduced token usage improves practicality

Faster execution times compared to state-of-the-art

Abstract

Predicting future values in multivariate time series is vital across various domains. This work explores the use of large language models (LLMs) for this task. However, LLMs typically handle one-dimensional data. We introduce MultiCast, a zero-shot LLM-based approach for multivariate time series forecasting. It allows LLMs to receive multivariate time series as input, through three novel token multiplexing solutions that effectively reduce dimensionality while preserving key repetitive patterns. Additionally, a quantization scheme helps LLMs to better learn these patterns, while significantly reducing token use for practical applications. We showcase the performance of our approach in terms of RMSE and execution time against state-of-the-art approaches on three real-world datasets.