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

Aug 5, 2026

Upload by Liam

Defining and finalizing Mergers and Acquisitions (M&A) depends on complex human judgment, making it difficult to identify optimal partners and anticipate deal formation. The study introduces the MASS algorithm, an interpretable similarity measure tailored to company technological relatedness derived from patenting activity data. Applied to Zephyr and Crunchbase, MASS forecasts M&A events and improves over LightGCN, while LightGCN performs better when similar firms exhibit disjoint patent activities. The approach supports decision-makers with explainable, practical insights.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages16

Reading Time40 min

AccessFree to download

Introduction

Motivation for M&A forecasting using patent activity

Relatedness and patent-based similarity metrics

Sapling Similarity and the proposed MASS approach

Machine learning-based similarity measure to forecast M&A from patent data⋆,⋆⋆ G Alboraa , M Straccamorea,b,∗ and A Zaccariac,a a Centro Ricerche Enrico Fermi, Via Panisperna 89/A, Rome, 00184 RM, Italy bSony CSL-Rome, Joint Initiative CREF-SONY, Centro Ricerche Enrico Fermi, Via Panisperna 89/A, Rome, 00184 RM, Italy cIstituto dei Sistemi Complessi (ISC) -CNR UoS Sapienza, P.le A. Moro, 2, Rome, 00185 RM, Italy | ARTICLE INFO | | AB STRACT | | --- | --- | --- | | Keywords:
Merging & Acquisition Patent
Technology innovation Sapling Similarity | | Defining and finalizing Mergers and Acquisitions (M&A) requires complex human skills, which makes it very hard to automatically find the best partner or predict which firms will make a deal. In this work, we propose the MASS algorithm, a specifically designed measure of similarity between companies and we apply it to patenting activity data to forecast M&A deals. MASS is based on an extreme simplification of tree-based machine learning algorithms and naturally incorporates intuitive criteria for deals; as such, it is fully interpretable and explainable. By applying MASS to the Zephyr and Crunchbase datasets, we show that it outperforms LightGCN, a "black box" graph convolutional network algorithm. When similar companies have disjoint patenting activities, on the contrary, LightGCN turns out to be the most effective algorithm. This study provides a simple and powerful tool to model and predict M&A deals, offering valuable insights to managers and practitioners for informed decision-making. | 1. Introduction In today’s rapidly evolving landscape of technological advancements, companies face the constant challenge of staying at the forefront of innovation. While internal research and development efforts play a significant role, they may not always be sufficient in terms of time and costs to keep up with the swiftly changing technological environment. As a result, many firms seek to expand their technological horizons by engaging in Mergers and Acquisitions (M&A) Bruner and Perella (2004) . Such operations are used extensively as a financial instrument by firms of any region and size and constitute a business that, only in 2019, has almost reached 4 trillion dollars ([https://imaa-institute.org](https://imaa-institute.org)). Such strategic moves allow companies to tap into the innovation capabilities of their target entities, leverage their technologies, and potentially venture into new markets. The choice of the possible best target for a deal is made in a complex evolving landscape of partners and competitors, involving a huge effort in terms of time and human capabilities. In this paper, we propose an automatized, machine learning-inspired approach to quantify the closeness between two firms in terms of their patenting activities, and we test this and other measures in an out-of-sample forecast exercise. Equipped with this tool, decision-makers can assess to what extent to exploit a technology sector a firm already masters or explore new innovation possibilities. In order to build a quantitative measure of the similarity between companies, we draw inspiration from the Economic Complexity framework Hidalgo and Hausmann (2009) . In particular, our investigation centers on the concept of "Relatedness" (Hidalgo, Balland, Boschma, Delgado, Feldman, Frenken, Glaeser, He, Kogler, Morrison et al. (2018); Hidalgo, Klinger, Barabási, and Hausmann (2007); Zaccaria, Cristelli, Tacchella, and Pietronero (2014)), which in our study serves as a measure of the similarity between two firms based on the technological sectors found in their patents. Our similarity metric allows us to compare and contrast the patent portfolios of acquiring and target companies, enabling a deeper understanding of the technological dynamics at play in these strategic transactions. Similarity metrics, such as cosine similarity, are the key to constructing collaborative filtering Schafer, Frankowski, Her

FAQs

What problem does the MASS approach address in M&A forecasting?

It targets the challenge of automatically identifying deal partners and predicting which firms will engage in M&A, using firms’ patenting activities as a measurable basis for similarity.

How does MASS differ from typical similarity measures used in recommendation and link prediction?

MASS introduces an extreme simplification of tree-based ML ideas, incorporates deal-relevant intuitive criteria, and relaxes common assumptions such as symmetry in acquisition probability.

Which algorithms does the study compare MASS against, and what are the outcomes?

The study compares MASS to LightGCN and reports that MASS outperforms LightGCN on Zephyr and Crunchbase; however, when patenting activities are disjoint, LightGCN becomes the most effective method.

Aug 5, 2026

Upload by Liam