Do Capacity Constrained Bots Collude?
Jan 10, 2025·
·
1 min read
John Sæten Lilletvedt
Ole Kristian Dyskeland
Abstract
Reinforcement learning algorithms have been shown able to collude on supracompetitive prices when playing pricing games. However, the literature suggests that collusion may not be robust to alternative setups of the game. We simulate models where the algorithms are constrained in production capacity. Our findings suggest that capacity constraints may lead to cycling behavior and high prices as a consequence. We do not find evidence that the cycling behavior is due to tacit collusion.
Type
Publication
Work in progress
Reinforcement learning algorithms have been shown able to collude on supracompetitive prices when playing pricing games. However, the literature suggests that collusion may not be robust to alternative setups of the game. We simulate models where the algorithms are constrained in production capacity. Our findings suggest that capacity constraints may lead to cycling behavior and high prices as a consequence. We do not find evidence that the cycling behavior is due to tacit collusion
I am a contributing author on this paper. For inquires, contact John.
This paper won the “Prize for the best PhD paper 2024” at the 46th Annual Meeting of the Norwgeian Association of Economists.
Algorithmic Pricing
Capacity Contraints
Artificial Intelligence
Computational Economics
Industrial Organization
Competition
Authors
John Sæten Lilletvedt
PhD Research Scholar
Authors
PhD Research Scholar
PhD Research Scholar in ‘Business Economics’ at NHH Norwegian School of Economics. Research in applied economic theory for industrial organisation and media economics.