Software that sets prices can learn to keep them high without ever being told to cooperate, and it can do so quickly. In simulations of seven different algorithms, prices above the competitive level emerged in less than a thousandth of the time reported in recent studies.
The study is by Bharat Bhole of the Rochester Institute of Technology and Sunita Surana. It was published online in Economic Inquiry on August 1, 2025, in the October 2025 issue, pages 1036 to 1065.
Answering the skeptics
Earlier studies had shown that pricing algorithms could collude tacitly, earning supra-competitive profits with no explicit communication. Critics questioned whether that mattered in practice, and the authors set out to address the main criticisms about the practical relevance of that work.
Bhole and Surana simulated competition among seven algorithms. Their algorithms reached supra-competitive prices in a reasonably short time, taking less than 1/1,000th of the time needed by the algorithms in recent studies.
Collusion survives a mix of rivals
The tacit collusion was also robust to firms choosing different algorithms. High prices did not depend on every competitor running the same code.
The best-performing algorithms shared four properties: niceness, forgiveness, provocability and flexibility.
For competition authorities, the two findings remove the easiest reasons for doubt: the collusion here was neither slow to appear nor dependent on rivals using identical software.
Study Details:
- Title: Tacit collusion by pricing algorithms
- Authors: Bharat Bhole, Sunita Surana
- Journal: Economic Inquiry
- Publication Date: August 1, 2025
- DOI: 10.1111/ecin.70004
