This paper explores the capability of large language models (LLMs) to generate coherent textual reports from time series data, using financial reports from stock data as the use case. We conduct a comprehensive multi-aspect evaluation across four model families, including linguistic quality, content source attribution, automated metrics, and expert human assessment. We evaluate models using four major stock indices and two synthetic time series to assess generalization. We assess reports based on single and multiple time series data, and experiment with plain text and multi-modal prompting. We examine temporal effects by analyzing report quality as data approaches model knowledge cutoffs and testing synthetic future intervals. Our evaluation shows that LLMs are capable of creating high-quality financial analyst reports, with larger models demonstrating superior performance, however even those require human oversight and have potential for temporal logic errors. Our findings reveal model-specific behavioral patterns that enable tailored generation pipelines and inform future research about model pitfalls in time series-to-text generation tasks.

Large Language Models (LLMs) offer the potential for automatic time series analysis and reporting, which is a critical task across many domains, spanning healthcare, finance, climate, energy, and many more. In this paper, we propose a framework for rigorously evaluating the capabilities of LLMs on time series understanding, encompassing both univariate and multivariate forms. We introduce a comprehensive taxonomy of time series features, a critical framework that delineates various characteristics inherent in time series data. Leveraging this taxonomy, we have systematically designed and synthesized a diverse dataset of time series, embodying the different outlined features, each accompanied by textual descriptions. This dataset acts as a solid foundation for assessing the proficiency of LLMs in comprehending time series. Our experiments shed light on the strengths and limitations of state-of-the-art LLMs in time series understanding, revealing which features these models readily comprehend effectively and where they falter. In addition, we uncover the sensitivity of LLMs to factors including the formatting of the data, the position of points queried within a series and the overall time series length.