Machine learning and information-theoretic approaches for financial applications

Ikegwu, Kelechi Moise

Permalink

https://hdl.handle.net/2142/112947 Copy

Description

Title

Machine learning and information-theoretic approaches for financial applications

Author(s)

Ikegwu, Kelechi Moise

Issue Date

2021-05-14

Director of Research (if dissertation) or Advisor (if thesis)

Brunner, Robert

Doctoral Committee Chair(s)

Brunner, Robert

Committee Member(s)

• Sougiannis, Theodore
• McMullin, Jeff
• Yun, Joseph

Department of Study

Illinois Informatics Institute

Discipline

Informatics

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

• Machine Learning
• Information Theory
• Random Walk
• Network Science
• Finance
• Accounting
• Financial Accounting

Abstract

Firm-specific information contributes to resource allocation in capital markets. The prediction and inference of future firm-specific information are often critical factors that can lead to a prosperous economy. Machine Learning and Information Theory offer new, alternative methods to aid in the prediction and inference of future firm-specific information. Thus, we explore approaches from machine learning and information theory to improve existing research areas in accounting and finance. In Chapter 2, we utilized random forest trees to predict the annual direction of profitability for firms with a minimal amount of information. We demonstrate that we can out-perform benchmark models typically used in financial accounting research. Major public firms announce their annual earnings during the first quarter of the year. The embedding information in these announcements effects the announcing firm and is transferred to or incorporated by other firms. Existing literature in finance and accounting focuses on measuring information transfers as effects stemming from firm-specific information releases. In Chapter 3, we discuss a solution to estimate information transfer via transfer entropy between random processes. We show that our solution to estimate information transfer scales better with data set size and is up to 1,072 times faster than all existing, open source solutions for large datasets. In Chapter 4, we present an alternative approach to estimate information transfer between firms centered around earnings announcements. We show that information transfer between firms is stronger for firms on days with unexpected earnings news. We also demonstrate that communities of firms significantly differ depending on the presence of earnings releases.

Graduation Semester

2021-08

Type of Resource

Thesis

Permalink

http://hdl.handle.net/2142/112947

Copyright and License Information

Copyright 2021 Kelechi Ikegwu

Owning Collections