Accounting and Finance Department Research Seminar
zoom
Hongkang Xu
5089999187
hxu5@umassd.edu
https://umassd.zoom.us/j/4387551509?pwd=bVB2QzdtRmFCdkk1WTVQSUxHMS9iQT09&omn=98524562508
The Department of Accounting & Finance announces the following research seminar.
Speaker: Professor of Business Administration, Charles C.Y. Wang (Harvard Business School)
Title: The Governance of Algorithmic Disclosures: A Black-Box Audit for Incentive-Aligned Distortion
Date: Friday, October 16, 2026
Time: 9:30 -10:45 AM
Location: via Zoom Meeting
Meeting ID: 438 755 1509
Abstract:
Algorithmic disclosures, the signals that algorithms generate to inform consumer decisions, increasingly shape how people spend their time and money. Yet their accuracy is rarely verified by anyone outside the platform that produces them. We propose a framework for auditing them without access to the underlying algorithm or its data. In a simple model, a platform generates a signal about a product attribute to influence a consumer’s transaction decision. The model predicts that distortion increases with the platform’s conversion value and the strength of the consumer’s outside option, and decreases with detection cost. In contrast, disclosures with little influence on conversion exhibit little incentive-aligned distortion. This suggests a black-box audit: identify the disclosures likely tied to the platform’s incentives, and, where available, lower-stakes comparison disclosures; compare those disclosures with independently measured outcome benchmarks across conditions that capture variation in distortion incentives. We apply this approach to one of the largest ride-hailing platforms in Hong Kong through a black-box field audit in which seven trained mystery riders, each using a dedicated account, generated 515 completed rides under a common protocol. Platform-recorded pick-up durations exceed the displayed estimates by 55.4% on average, compared to 6.8% for travel time. Consistent with incentive-aligned distortion, we find that pick-up underestimation is more severe for routes near mass transit substitutes and during peak hours and less severe for attentive riders; travel-time estimates show little corresponding variation. We discuss other applications and what an audit regime for algorithmic disclosures would require.
For additional information, please contact Prof. Hongkang Xu at hxu5@umassd.edu.