Auburn researchers' new machine learning-based method highlights the importance of algorithmic approaches for addressing class imbalance and supporting theory testing.

A team of Harbert College of Business researchers have developed a novel algorithmic approach for handling imbalanced data sets—a common occurrence in predictive analytics—and they aim to apply the method to the world of mergers and acquisitions (M&A).

Predictive analytics tools allow businesses to make data-driven decisions, mitigate potential risks, and increase the odds of a successful M&A deal.

As a point of reference, management consulting firm Bain & Company reported that the value of M&A deals globally was more than $3 trillion in 2023.

Business Analytics faculty members Kangbok Lee and Sumin Han created the method with former Auburn doctoral student Yeasung Jeong, who is now a faculty member at the State University of New York at Albany, and Young Woong Park, an assistant professor at Iowa State University.

According to Lee, their method, which they call the systemic approach for imbalanced learning, may someday be used to predict whether an announced corporate M&A will successfully result in a deal.

“We examined whether the acquiring and target firms indeed merged after the deal was announced,” said Lee, the EBSCO associate professor of business analytics. “Completing the deal is crucial for stakeholders like merger arbitrageurs, target shareholders and acquiring firms. Their primary interest lies in predicting whether a deal will go through to effectively manage financial risks and opportunities.”

These deals can fail for various reasons—shareholders get cold feet and vote against the deal, another company steps in with a better offer, or regulatory issues prohibit the deal, for example. But the bottom line is investors would like to be able to predict whether a deal will close because a failed M&A will affect the reputation and stock price of the companies involved.

A nature-inspired algorithm

The algorithm-based model that Lee and his colleagues developed to predict M&A success was inspired by nature.

“Our systemic approach for an imbalanced learning algorithm is inspired by the survival story of sea turtle hatchlings,” Lee explained. “Many hatchlings die on the beach due to predators, but once they enter the ocean their survival chances dramatically go up."

Lee and his colleagues applied this two-stage scenario—beach and sea—to M&A by creating an automated algorithm that contains two latent equations—a split probit equation in the first stage and an ordinary logit equation in the second stage.

“In the first stage of the M&A survival story, completed cases are primarily influenced by M&A-related predictors, as highlighted in the literature,” Lee said. “Hostile deals often encounter resistance from target firms, making non-hostile deals more likely to succeed, while tender offers signal confidence in the transaction. We also considered factors like the relative risk between acquirer and target firms and pre-announcement increases in the target firm’s share price, which can reduce the likelihood of bid competition and price revision.”

In the second stage, financial predictors play a significant role in M&A deal that fail to go through.

“We analyzed financial characteristics of both acquirer and target firms, including dividend per share, inventory-to-total-assets ratio, market-to-book ratio, price-to-earnings ratio, sales growth rate, capital expenditure-to-operating revenue ratio, invested capital turnover, dividend yield, and total assets logarithm,” he said.

Conventional prediction methods, in contrast, typically include both success and failure outcomes of an M&A deal in a single binary equation, Lee said.

“This means that the same set of factors is used to predict both success and failure simultaneously, which may limit the accuracy of the model by treating the two outcomes as driven by the same underlying dynamics,” he added. “In contrast, our approach separates the distinct outcomes into two equations—one equation focuses on explaining the factors leading to successful deals, while the other explains the factors contributing to failed deals.”

According to Lee, their experimental results show that their automated proposed algorithm yields better predictive accuracy compared to other leading algorithms.

“Given that our approach highlights the crucial importance of accounting for class imbalance in M&A data to ensure accurate empirical analysis, it is clear that future research must integrate advanced methods, such as algorithmic approaches, to effectively address this issue in theory testing,” Lee said. “Traditional logit or probit models, while useful, are insufficient for testing hypotheses and capturing the complexities inherent in imbalanced datasets. As such, the adoption of more sophisticated techniques, like the two-regime process combined with boosting, is essential for producing more reliable and accurate theory testing in M&A research.”

Lee published his results and the algorithm in Machine Learning, a highly regarded peer-reviewed scientific journal primarily read by computer science researchers. He and his colleagues also introduced algorithmic approaches for causal inference in Pattern Recognition, a highly respected peer-reviewed journal in computer science.

They are currently working on a follow-up research paper detailing their algorithm-based approach, which is currently under revision for the Journal of Operations Management, a leading business journal.

###

Learn more about business analytics research and education in the Harbert College of Business.