Data Privacy in Pricing: Estimation Bias and Implications

Abstract

Problem definition: Motivated by emerging regulations, we study how two privacy mechanisms, limited retention (deleting historical data) and self-protection (a subset of customers hiding their private data), bias demand estimation in data-driven pricing, and how these biases affect prices, consumer surplus, and firm revenue. Methodology/results: Using linear demand models, we show that the direction of price changes depends on product type (e.g., normal vs. inferior goods), while the magnitude increases with the degree of historical personalization but is nonmonotonic in the number of protected features. Self-protection also causes prices to move in opposite directions for protective vs. non-protective customers, potentially harming low-income privacy-aware consumers. We then develop a unified analytical framework demonstrating that privacy induces and amplifies fundamental epistemic uncertainty that persists even under nonlinear models. Firms may quantify this uncertainty, but the accuracy is statistically limited, and attempts to correct the bias according to the quantification can increase the risk of large pricing deviations. We also identify diminishing returns to improved quantification accuracy when privacy-induced uncertainty is large, a challenge especially acute under self-protection. Managerial implications: Comprehensive privacy laws can unintentionally amplify pricing uncertainty and create divergent outcomes across customer groups. Policy-makers should weigh the benefits of data confidentiality against the risk of volatile and inequitable market outcomes, particularly in industries with strong personalization.

Keywords: Privacy, Personalized Pricing, Model Estimation, Revenue Management