Shopbots and Pricebots: The Strategic Dynamics of Automated Markets
Introduction
The rise of electronic commerce in the late 1990s promised a "friction-free economy" where near-zero information costs would maximize market efficiency. However, this paper argues that the free flow of information does more than just lubricate existing markets; it fundamentally alters the strategic landscape. This study explores the economic impact of autonomous agents—specifically shopbots (buyer agents seeking low prices) and pricebots (seller agents setting adaptive prices)—on market stability and profitability.
Market Model and Methodology
The authors model an economy with $S$ sellers and $B$ buyers. Buyers are categorized into two groups: "Any Seller" consumers who choose randomly, and "Bargain Hunters" who utilize shopbots to identify the lowest available price. To respond to these automated buyers, sellers employ one of three price-setting algorithms:
- Game-Theoretic (GT): Agents calculate a symmetric mixed-strategy Nash equilibrium based on full knowledge of the buyer population.
- Myopically Optimal (MY): Agents assume competitors' prices are static and adjust their own price to maximize immediate expected profit.
- Derivative-Following (DF): A "black-box" strategy where agents experiment with price changes, continuing in a direction as long as profit increases.
The research combines game-theoretic proofs with asynchronous simulations spanning 100 million time steps to observe long-term dynamics.
The Death of Pure Equilibrium
A central finding is the absence of a pure-strategy Nash equilibrium in markets with shopbots. If all sellers charge the same price, an individual seller is incentivized to undercut. Conversely, if prices drop to marginal cost, sellers are incentivized to raise prices to extract higher margins from non-shopbot buyers.
As shown in Figure 1, at high levels of shopbot usage, the Nash equilibrium becomes bimodal. Sellers tend to either price near the buyer's maximum valuation to capture high margins from random buyers or price just above the competitive floor to capture the high volume of shopbot-driven "Bargain Hunters."
Price Wars and Accidental Collusion
The dynamic behavior of these markets varies significantly depending on the algorithms used. Myopically optimal (MY) agents inevitably trigger cyclical price wars. They successively undercut one another until the profit margin becomes so thin that it is more lucrative to "reset" the price to the maximum valuation and begin the downward cycle again.
Surprisingly, the least sophisticated agents—the Derivative Followers (DF)—often achieve the highest profits. Because they only react to their own profit trends without modeling competitors, they frequently reach a state of "tacit collusion." As illustrated in Figure 2, DF agents maintain prices near the monopolistic peak without explicit communication, outperforming more "intelligent" algorithms in stable environments.
Comparative Results
In simulations with 5 sellers and 75% shopbot usage, the performance gap was stark:
- GT agents earned a theoretical average profit of 0.0250.
- MY agents earned 0.0515, exploiting the cyclical nature of price wars to outperform the theoretical Nash equilibrium.
- DF agents earned 0.0841, nearly reaching the absolute collusive limit of 0.10.
However, this collusion is fragile. When a single MY agent is introduced into a population of DF agents, it "substantially outplays" them by aggressively undercutting the high-price equilibrium.
Conclusion
While shopbots drive market efficiency and lower average consumer prices, they also catalyze complex agent behaviors. The study suggests that information itself may become a commodity, with shopbots eventually charging for the price data they collect. Future research should explore more sophisticated reinforcement learning, such as Q-learning, to model how agents might adapt their strategies in increasingly complex, multi-agent environments.
Summary updated