The 23rd International Conference on Principles and Practice of Multi-Agent Systems, pp. 352-359, Nov. 19, 2020
The 23rd International Conference on Principles and Practice of Multi-Agent Systems (PRIMA 2020)
In this paper, we address the problem of unintentional price collusion, which happens due to auto pricing, such as systems using reinforcement learning. Firstly, Q-learning, sarsa, and deep Q-Learning models were used for auto pricing to test whether they cause collusion. To test them, we performed multi-agent simulations of a competitive market with a pre-defined demand function. In each simulation, the agents learn their pricing strategies using reinforcement learning. And we defined and calculated the new collusion metric representing how agents collude. Secondly, we tested cases with open and shield bidding with multiple numbers of agents. In our result, we observe that deep Q-Learning demonstrates the highest collusion metric. Also, contrary to expectations, we found that shield bidding has no significant effect on collusion levels when agents employ outperforming reinforcement learning, such as deep Q-learning. Moreover, the number of agents also contribute to less collusion levels.
Auto pricing; Unintentional collusion; Reinforcement learning; Deep learning; Market;
@inproceedings{Hirano2020-prima-scm,
title={{Simulation of Unintentional Collusion Caused by Auto Pricing in Supply Chain Markets}},
author={Masanori HIRANO and Hiroyasu MATSUSHIMA and Kiyoshi IZUMI and Taisei MUKAI},
booktitle={The 23rd International Conference on Principles and Practice of Multi-Agent Systems},
isbn={978-3-030-69322-0},
pages={352-359},
publisher={Springer},
doi={10.1007/978-3-030-69322-0_24},
year={2020}
}