LLMs Handling High-Dimensional Financial Time Series

Ascertain whether large language models can effectively analyze and model high-dimensional financial time series data, including multivariate dependencies and complex dynamics, and quantify their performance in forecasting and inference compared to specialized time series models and potential hybrid architectures.

Background

The authors note that LLMs excel at long-context language understanding but emphasize that financial time series pose unique challenges due to high dimensionality and complex dependencies. They suggest exploring hybrid models, domain-specific pretraining, and integration with time series techniques.

Given these difficulties, the paper explicitly states uncertainty regarding LLM performance on high-dimensional financial time series, identifying a concrete unresolved question for future research.