# Time Series Forecasting with LLMs: Understanding and Enhancing Model Capabilities

**Authors:** Mingyu Jin, Hua Tang, Chong Zhang, Qinkai Yu, Chengzhi Liu, Suiyuan Zhu, Yongfeng Zhang, Mengnan Du  
**arXiv:** [2402.10835](https://arxiv.org/abs/2402.10835)  
**Date:** 2024-02-16  
**Summary page:** https://www.summarizepaper.com/en/paper/2402.10835v1/

## Key points

- Large Language Models (LLMs) are being applied in time series forecasting
- LLMs excel at predicting time series data with clear patterns and trends
- Challenges arise for LLMs when dealing with datasets lacking periodicity
- Researchers designed prompts to help LLMs identify periods within datasets to improve performance
- Incorporating external knowledge and natural language paraphrases enhances the accuracy of LLM predictions for time series data
- Understanding the advantages and limitations of using LLMs in time series forecasting is crucial for optimizing model capabilities

## Summary

In their paper titled "Time Series Forecasting with LLMs: Understanding and Enhancing Model Capabilities," Mingyu Jin, Hua Tang, Chong Zhang, Qinkai Yu, Chengzhi Liu, Suiyuan Zhu, Yongfeng Zhang, and Mengnan Du explore the application of Large Language Models (LLMs) in time series forecasting. The authors highlight the rapid development of LLMs across various fields and how they have recently impacted time series prediction. By comparing LLMs with traditional models, the study uncovers several key properties of LLMs in this domain. One significant finding is that LLMs demonstrate proficiency in predicting time series data with clear patterns and trends. However, they encounter challenges when dealing with datasets lacking periodicity. To address this issue, the researchers designed prompts that required LLMs to identify the period within the datasets. This approach shed light on how LLMs perform under different conditions and provided insights into their preferences. Moreover, the study delves into the impact of input strategies on predictive performance. It was discovered that incorporating external knowledge and utilizing natural language paraphrases positively influenced the accuracy of LLM predictions for time series data. This suggests that leveraging additional information and refining input methods can enhance the capabilities of LLMs in forecasting future trends. Overall, this research contributes valuable insights into both the advantages and limitations of using LLMs for time series forecasting. By understanding these nuances and exploring ways to optimize model capabilities, practitioners can make more informed decisions when applying LLM technology in this field.