学術雑誌論文

本学関連学会

text

大阪大学経済学. 2025, 74(4), p. 45-60

Earnings prediction using machine learning : A survey

Peng, Yuanchao

英語

text

2025-03

This survey investigates the application of machine learning (ML) techniques in predicting corporate earnings. By reviewing literature spanning 2019 to 2024, this paper aims to provide a comprehensive overview of the methodological trends, strengths, and limitations of current ML approaches in the context of earnings prediction. While most research focuses on U.S. firms, a smaller portion examines international firms, including one study on Japan. A key trend is the preference for predicting directional changes in earnings (binary classification) over actual earnings levels, as classification models leverage high-dimensional data more effectively and yield economically meaningful insights. For example, portfolios based on predicted earnings changes outperform traditional models in generating abnormal returns. This paper also points out the shortcomings of existing research, 1) there is a lack of sufficient international evidence to prove that ML is superior to traditional models, 2) most earnings forecasts are short-term, and 3) there is a lack of exploration of non-financial data or forward-looking data. These shortcomings point to promising directions for future research. Another notable trend is that large language models (LLMs) have been shown to outperform traditional methods and human analysts in predicting the direction of corporate earnings. This emerging approach demonstrates impressive predictive performance in analyzing financial ratios and trends without extensive retraining.

10.18910/100638

100638

Earnings prediction using machine learning : A survey

Peng, Yuanchao

大阪大学経済学会・大阪大学大学院経済学研究科

英語

2025-03

大阪大学経済学

74

4

45

60

https://hdl.handle.net/11094/100638

10.18910/100638

https://doi.org/10.18910/100638