Please rotate your device to landscape mode to view the charts.

Artificial Intelligence, Algorithmic Pricing, and Collusion

Journal: American Economic Review

Date: 20201001

Author: Calvano, Emilio; Calzolari, Giacomo; Denicolo, Vincenzo; Pastorello, Sergio

Abstract:
Increasingly, algorithms are supplanting human decision-makers in pricing goods and services. To analyze the possible consequences, we study experimentally the behavior of algorithms powered by Artificial Intelligence (Q-learning) in a workhorse oligopoly model of repeated price competition. We find that the algorithms consistently learn to charge supracompetitive prices, without communicating with one another. The high prices are sustained by collusive strategies with a finite phase of punishment followed by a gradual return to cooperation. This finding is robust to asymmetries in cost or demand, changes in the number of players, and various forms of uncertainty.

Link: Google Scholar

Key Findings

Autonomous Price Collusion

Q-learning pricing algorithms systematically learn to collude and charge supracompetitive prices, achieving 70-90% of monopoly profits without any explicit communication or instruction.

Punishment Strategies

Algorithms learn sophisticated punishment strategies with finite duration and gradual return to pre-deviation prices, making price cuts unprofitable in over 95% of cases.

Robust Collusion

Collusive behavior persists across different market conditions including asymmetric costs, stochastic demand, and varying numbers of competitors.

Profit Gains Across Market Conditions

  • Baseline duopoly achieves 85% of monopoly profits
  • Three-firm markets still maintain 64% profit gains
  • Four-firm markets achieve 56% of monopoly profits

Cost Asymmetry Effects

  • Collusion persists even with significant cost differences between firms
  • Profit gains remain above 70% even with 75% cost differential
  • Less efficient firms receive disproportionately higher profit shares

Price Response After Deviation

  • Initial punishment phase with price war
  • Gradual return to pre-deviation prices over 5-7 periods
  • Deviation reduces cheater's profits by 3-4% on average

Contribution and Implications

  • First demonstration that AI pricing algorithms can autonomously learn to collude without communication
  • Suggests need to reconsider antitrust policy approaches as algorithmic pricing becomes more prevalent
  • Opens possibility for new forms of antitrust intervention through direct testing of pricing algorithms

Data Sources

  • Profit gains chart based on results reported in Section V.A on number of players
  • Cost asymmetry effects visualized from Table 4 data on varying cost differentials
  • Price deviation responses constructed from data in Tables 2-3 and Figure 4