算法决策、合谋与反垄断规制

• 作者： 喻言 ¹ 周悦 ¹
• 作者单位：
  1.  湖南农业大学
• 通讯作者： 周悦 
• 提交时间：2024-11-26

摘要: 随着人工智能的发展，算法能够自主学习并实现目标行为规则，会显著影响市场行为和竞争模式。针对此问题，本文聚焦于Q-learning算法在市场合谋中的应用，通过对伯特兰德双寡头竞争市场进行模拟，研究了Q-learning算法在不同参数设置下的合谋能力和稳定性。结果表明，虽然提高学习率和实验参数会导致合谋结果，但可能需要更长的周期来达到合谋程度的上升。该结论在扩大动作网格维度的情况下依然稳健，但在扩大边际成本设置时则会导致无效合谋，且折现系数δ不宜过低，否则无法实现合谋。最后基于研究结果提出相应的政策建议。

Abstract: With the development of artificial intelligence, algorithms can autonomously learn and implement target behavioral rules, which will significantly affect market behavior and competition patterns. To solve this problem, this paper focuses on the application of Q-learning algorithm in market collusion, and studies the collusion ability and stability of Q-learning algorithm under different parameter Settings by simulating Bertrand duopoly competition market. The results suggest that while increasing the learning rate and experimental parameters will lead to collusion outcomes, it may take a longer period to reach an uptick in collusion levels. This conclusion is still robust when the dimension of the action mesh is enlarged, but it will lead to ineffective collusion when the marginal cost setting is enlarged, and the discount coefficient δ should not be too low, otherwise collusion cannot be realized. Finally, the corresponding policy recommendations are put forward based on the research results.

• 来自： 周悦
• 分类： 应用经济学 >> 产业经济学
• 稿件状态： 已投稿期刊
• 引用： PSSXiv:202411.02350 (或此版本 PSSXiv:202411.02350V1)
  DOI:10.12451/202411.02350
  CSTR:32012.36.PSSXiv.202411.02350
• 推荐引用方式： 喻言,周悦.算法决策、合谋与反垄断规制.哲学社会科学预印本平台:https://zsyyb.cn/abs/202411.02350.[PSSXiv:202411.02350V1] (点此复制)