Date Approved

2026

Degree Type

Open Access Senior Honors Thesis

Department or School

Accounting and Finance

First Advisor

Yu Zhang, Ph.D.

Second Advisor

Jodonnis Rodriguez, Ph.D.

Third Advisor

Charles Teague III, Ph.D.

Abstract

This research seeks to determine the effectiveness of machine learning models in predicting revenues of corporations compared to conventional models of financial forecasting. The investigation is based on a group of 21 companies that maintained membership in the Dow Jones Industrial Average (DJIA) between 2010 and 2025. Two machine learning algorithms are assessed: Random Forest and XGBoost. They are compared with the naïve growth model and deterministic linear trend, which serve as benchmarks. To check for statistical significance, the Wilcoxon signed-rank tests are used to compare the mean absolute percentage error (MAPE) and the root mean squared error (RMSE) of the models. The results show that machine learning models always do better than the linear trend benchmark in terms of both average accuracy and statistical reliability. But the improvements over the simple growth model are small and not statistically significant for the whole sample. Further analysis shows that the benefits of machine learning are not universal. Machine learning models deliver substantial improvements in forecasting accuracy for firms with high revenue volatility, which assumes revenue dynamics are more complex and less predictable. On the flip side, for firms with stable revenue patterns, simple growth-based models outperform more complex algorithms. In summary, it is observed that although ML can perform better in certain environments, specifically for non-linear datasets, traditional methods can be still very competitive under static conditions. This experiment demonstrates the necessity of fitting forecasting techniques to the data features.

Download

EMU Library Accessibility Request

Share

COinS