Machine learning-inspired similarity measure to forecast M&A from patent data

Area of Science:

• Computational Social Science
• Business Analytics
• Intellectual Property Management

Background:

• Mergers and Acquisitions (M&A) prediction is challenging due to complex human factors.
• Existing methods struggle to automatically identify optimal partners or forecast deal likelihood.
• A need exists for interpretable and effective M&A prediction tools.

Purpose of the Study:

• To propose and evaluate the MASS algorithm for forecasting M&A deals.
• To adapt patent-based company similarity measures for M&A prediction.
• To compare MASS against a graph convolutional network (GCN) approach.

Main Methods:

• Developed the MASS algorithm, a simplified, interpretable tree-based machine learning model.
• Utilized patent data to measure company similarity.
• Applied the algorithm to the Zephyr and Crunchbase datasets for validation.

Main Results:

• The MASS algorithm demonstrated superior performance in predicting M&A deals compared to a GCN model.
• The GCN model showed higher effectiveness for companies with disjoint patenting activities.
• MASS provides an interpretable and explainable approach to M&A forecasting.

Conclusions:

• MASS offers a simple yet powerful tool for modeling and predicting M&A deals among patent-active firms.
• The algorithm provides valuable insights for managers and practitioners in strategic decision-making.
• This work highlights the utility of patent-based similarity for M&A analysis.