Working Paper: CEPR ID: DP16393
Authors: Emilio Calvano; Giacomo Calzolari; Vincenzo Denicolo; Sergio Pastorello
Abstract: We clarify the difference between the asynchronous pricing algorithms analyzed by Asker, Fershtman and Pakes (2021) and those considered in the previous literature. The difference lies in the way the algorithms explore: randomly or mechanically. We reaffirm that with random exploration, asynchronous pricing algorithms learn genuinely collusive strategies.
Keywords: Artificial Intelligence; Reinforcement Learning; Collusion; Exploration
JEL Codes: L41; L13; D43; D83
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| random exploration (C99) | learning collusive strategies (C72) |
| random exploration (C99) | competitive prices (D41) |
| myopic algorithms (C60) | competitive pricing strategies (L11) |
| exploration methods (C90) | ability to sustain high prices (Q31) |
| asynchronous algorithms (C69) | learning outcomes (A21) |