Stock price forecasting using customized decoder-only foundational time series‬ ‭ model

The goal of this praxis is to develop forecasting models using a foundational time‬ series model based on Google's decoder-only architecture (Das et al., 2024). These‬‭ models are customized for stock forecasting purposes and further fine tuned with stock‬‭ data to enhance their predictive performance. The goal is to make these models better at‬ predicting the future values of FAANG stocks (Facebook, Apple, Amazon, Netflix, and‬‭ Google) exceeding the capabilities of both classical and contemporary deep learning‬‭ models currently available for time series analysis, as well as outperform state of the art‬ commercial time series models that people currently use in the market created by Garza‬‭ et al. (2024) and highlighted by nixtla.io company. The intentions of forecasting length‬‭ for FAANG stocks are 5, 10, 15, 20, 25 and 30 days. These duration lengths are corresponding to short to medium term investment strategies.‬‭

Author

• Tran, Khoa N

Language

• en

Keyword

• timeseries
• markets
• machine learning
• stocks
• artificial intelligence
• foundational

Date created

• 2025

Type of Work

• Dissertation

License

• All rights reserved

Rights statement

• In Copyright

GW Unit

• Computer Science

Degree

• D.Engr.

Advisor

• Belkhayat, Mohamed

Committee Member(s)

• Focacci, Aaron
• Raghavan, Vijay

Persistent URL

•  https://scholarspace.library.gwu.edu/etd/6682x490c