Reproduce “Mitigating Retaliatory Algorithmic Collusion in Repeated Games” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Karthik Sivachandran, Rohan Paleja

The authors introduce CURB, a learning intervention that penalizes changes in an agent’s action distribution after a rival defects and adds synthetic examples of unpunished defection. Its theory rules out nontrivial punishment paths at two-player SPC fixed points under assumptions. In two-firm Bertrand Q-learning, CURB reduced the Collusion Index to 0.0029 ± 0.0124, but this is simulated evidence, not a real-market result.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.