Masters Theses

Keywords and Phrases

Bitcoin; Forecasting; Large Language Models; Machine Learning; Natural Language Processing; Transformer

Abstract

"In the context of a Master's thesis in applied mathematics, this work investigates compact transformer-based language models as an instrument for Bitcoin-related predictions and decision support. The work connects three topics: the structure of the Bitcoin system and its data, the mathematical and algorithmic foundations of deep autoregressive transformers, and the design of practical training pipelines for financial applications. On this foundation, a reproducible ETL pipeline for Bitcoin data is developed and two forecasting experiments are conducted with the compact language model nanochat. The first experiment approaches the prediction of next-day price movements through autoregressive next-token generation based on structured input. In the second experiment the role of the language model changes: it is used to extract hidden news embeddings, which serve as features for a downstream predictor. Performance is evaluated with formal statistical tests. Beyond forecasting, the thesis sketches a proof-of-concept path toward language-based market interpretation and decision making in cryptocurrency markets"-- Abstract, p. iii

Advisor(s)

Hu, Wenqing

Committee Member(s)

Olbricht, Gayla R.
Singler, John R.

Department(s)

Mathematics and Statistics

Degree Name

M.S. in Applied Mathematics

Publisher

Missouri University of Science and Technology

Publication Date

2026

Pagination

xi, 141 pages

Note about bibliography

Includes_bibliographical_references_(pages 133-140)

Rights

© 2026 Erich Gozebina , All Rights Reserved

Document Type

Thesis - Open Access

File Type

text

Language

English

Thesis Number

T 12618

Download

Share

 
COinS