InfoBid: A Simulation Framework for Studying Information Disclosure in Auctions with Large Language Model-based Agents

TL;DR

InfoBid introduces a flexible simulation framework that uses GPT-4o-based agents to study information disclosure in auction environments, specifically examining signaling strategies in second-price auctions. By simulating private signaling schemes such as Full Disclosure, Tiered Pooling, and Randomized Pooling under Known-Valuation with 10 bidders, the framework reveals how signaling affects bidder rationality, revenue, and social welfare. The findings show that carefully designed signaling can increase revenue and welfare, with richer information (including tier-average values) mitigating deviations from truthful bidding and enhancing market efficiency. This work bridges theoretical market design and practical simulation, providing a scalable tool for analyzing information design and agent-based reasoning in digital economies.

Abstract

In online advertising systems, publishers often face a trade-off in information disclosure strategies: while disclosing more information can enhance efficiency by enabling optimal allocation of ad impressions, it may lose revenue potential by decreasing uncertainty among competing advertisers. Similar to other challenges in market design, understanding this trade-off is constrained by limited access to real-world data, leading researchers and practitioners to turn to simulation frameworks. The recent emergence of large language models (LLMs) offers a novel approach to simulations, providing human-like reasoning and adaptability without necessarily relying on explicit assumptions about agent behavior modeling. Despite their potential, existing frameworks have yet to integrate LLM-based agents for studying information asymmetry and signaling strategies, particularly in the context of auctions. To address this gap, we introduce InfoBid, a flexible simulation framework that leverages LLM agents to examine the effects of information disclosure strategies in multi-agent auction settings. Using GPT-4o, we implemented simulations of second-price auctions with diverse information schemas. The results reveal key insights into how signaling influences strategic behavior and auction outcomes, which align with both economic and social learning theories. Through InfoBid, we hope to foster the use of LLMs as proxies for human economic and social agents in empirical studies, enhancing our understanding of their capabilities and limitations. This work bridges the gap between theoretical market designs and practical applications, advancing research in market simulations, information design, and agent-based reasoning while offering a valuable tool for exploring the dynamics of digital economies.

Figures (4)

• Figure 1: Information flow between auctioneer and bidders in an example auction. Gray-colored messages represent private information, such as Your value is 0.7 provided individually to bidders by the auctioneer. Green-colored text represents sealed bids submitted privately by bidders to the auctioneer, such as I bid 0.7 or I bid 0.8. Blue-colored messages signify public information, such as the auctioneer announcing, Bidder 2 wins and pays 0.7, visible to all participants. This example showcases a specific disclosure strategy, including conditions for revealing true values and withholding total bidder count.
• Figure 2: Impact of Signaling Strategies on Bid Deviation Behavior The chart illustrates the percentage of deviated bidding behavior across various signaling strategies, as represented on the x-axis. The y-axis denotes the proportion of bids categorized into three groups: truthful bidding (where the bid value equals the estimated value), overbidding (where the bid value exceeds the estimated value), and underbidding (where the bid value is lower than the estimated value). Each bar is derived from 4000 bid records.
• Figure 3: Revenue Comparison Across Disclosure Strategies and Scenarios. The x-axis represents the average probability of full disclosure under various scenarios. For instance, pooling bidders with values above the 20th percentile corresponds to a full disclosure probability of 0.2. The y-axis shows the average revenue generated per round for each configuration. The chart compares four disclosure strategies: Full Disclosure (depicted as a horizontal line); Pool-High and Pool-Low with each strategy evaluated across different cut-off values; Random, where bidders are randomly selected for full disclosure without providing additional information to the remaining bidders.
• Figure 4: Social welfare Comparison Across Disclosure Strategies and Scenarios. The x-axis represents the average probability of full disclosure under various scenarios. For instance, pooling bidders with values above the 20th percentile corresponds to a full disclosure probability of 0.2. The y-axis shows the average revenue generated per round for each configuration. The chart compares four disclosure strategies: Full Disclosure (depicted as a horizontal line); Pool-High and Pool-Low with each strategy evaluated across different cut-off values; Random, where bidders are randomly selected for full disclosure without providing additional information to the remaining bidders.