Time series forecasting of stock price changes using large language models: A foundation for financial decision-making
Go, Eun
This item's files can only be accessed by the System Administrators group.
Permalink
https://hdl.handle.net/2142/129687
Description
Title
Time series forecasting of stock price changes using large language models: A foundation for financial decision-making
Author(s)
Go, Eun
Issue Date
2025-04-15
Director of Research (if dissertation) or Advisor (if thesis)
Banerjee, Arindam
Department of Study
Siebel School Comp & Data Sci
Discipline
Computer Science
Degree Granting Institution
University of Illinois Urbana-Champaign
Degree Name
M.S.
Degree Level
Thesis
Keyword(s)
Time series forecasting
Large language models
Abstract
Forecasting stock price changes is a fundamental task in financial decision-making, yet it remains highly challenging due to the noisy, non-stationary nature of financial time series. This thesis investigates the use of a foundation model—Chronos, a large language model (LLM)-based forecaster—for multi-horizon stock price change prediction. Unlike traditional models that focus on absolute price forecasting, we formulate the task around predicting future price differences, which are more actionable in trading and portfolio optimization contexts. We evaluate Chronos under multiple configurations, comparing zero-shot and fine-tuned settings across several input formats, including raw prices, daily differences, and horizon-based changes. Extensive experiments on U.S. stock data demonstrate that fine-tuning Chronos significantly improves predictive performance, especially at longer horizons. Among the formats, daily price differences yield the most stable and interpretable results. An ablation study on context length further reveals trade-offs between longer historical windows and increased noise. While no control policy is implemented in this work, we propose a future integration of Chronos into a model predictive control (MPC) framework for multi-period financial planning. This thesis concludes that LLM-based forecasters like Chronos are promising tools for time series prediction in finance, especially when paired with domain-specific fine-tuning and careful input design.
Use this login method if you
don't
have an
@illinois.edu
email address.
(Oops, I do have one)
IDEALS migrated to a new platform on June 23, 2022. If you created
your account prior to this date, you will have to reset your password
using the forgot-password link below.
