Research

There has been increasing attention to privacy in the media and in regulatory discussions. This is a consequence of the increased usefulness of digital data. The literature has emphasized the benefits and costs of digital data flows to consumers and firms. The benefits arise in the form of data-driven innovation, higher-quality products and services that match consumer needs, and increased profits. The costs relate to the intrinsic and instrumental values of privacy. Under standard economic assumptions, this framing of a cost-benefit trade-off might suggest little role for regulation beyond ensuring consumers are appropriately informed in a robust competitive environment. The empirical literature thus far has focused on this direct cost-benefit assessment, examining how privacy regulations have affected various market outcomes. However, an increasing body of theory work emphasizes externalities related to data flows. These externalities, both positive and negative, suggest benefits to the targeted regulation of digital privacy.

Constrained Markov Decision Process (CMDP) is a natural framework for reinforcement learning tasks with safety constraints, where agents learn a policy that maximizes the long-term reward while satisfying the constraints on the long-term cost. A canonical approach for solving CMDPs is the primal-dual method which updates parameters in primal and dual spaces in turn. Existing methods for CMDPs only use on-policy data for dual updates, which results in sample inefficiency and slow convergence. In this paper, we propose a policy search method for CMDPs called Accelerated Primal-Dual Optimization (APDO), which incorporates an off-policy trained dual variable in the dual update procedure while updating the policy in primal space with on-policy likelihood ratio gradient. Experimental results on a simulated robot locomotion task show that APDO achieves better sample efficiency and faster convergence than state-of-the-art approaches for CMDPs.

Understanding consumer privacy choices is important because firms trade off the competing goals of data access and privacy protection. This paper demonstrates that past privacy choices affect consumers’ current privacy choices. Such state-dependent choices suggest that privacy choices can have externalities within a platform in which one app's data requests can affect the ability of other apps to collect data. Specifically, I use a consumer-level panel to investigate data consent decisions by consumers on Alipay, a major digital platform that connects users and third-party apps. Taking advantage of a natural experiment that encourages users to accept data requests, I find that the probability of rejecting the next request declines 15%. This effect decays over time, is larger for users who did not previously look over the privacy terms, and is larger when the next request is in a weak preference service context, categorized by a large language model (LLM). The effect does not differ by whether the specific data requested in consecutive data consent decisions is the same nor by the overall rate of data consent given by the consumer. Overall, I interpret these results to suggest that the externalities arising from state-dependent data consent choices are temporary. Nevertheless, they can positively or negatively impact the ability of apps to collect data, and therefore platforms have incentives to encourage apps to provide consumer-friendly data request designs.

Privacy regulation relies on informed consent, delegating consumer protection to consumer choice. But the payoff to sharing depends on how firms use data and on who else shares, so disclosure is a strategic decision, and consumers differ in their ability to reason through it. This paper studies how bounded rationality affects the welfare consequences of privacy policies that improve information or individual sophistication, such as national campaign. We model data disclosure as a simultaneous strategic game in which a consumer’s payoff depends on both firm use of the data and the composition of the pool that agrees to share. We estimate a cognitive-hierarchy model using a data-sharing study fielded through the Dutch LISS panel, in which 3,598 respondents rated their trust in nine organizations and made a real decision whether to share their data with academic researchers. The estimates imply substantial heterogeneity in strategic sophistication: most consumers do not fully internalize the strategic consequences of disclosure, and beliefs about downstream data use differ sharply across organizations. We then evaluate a transparency policy that improves consumers’ understanding of disclosure and shifts the population toward higher cognitive sophistication. The policy generates a direct welfare gain for previously naive consumers by correcting their disclosure mistakes, but it also induces adverse selection in the disclosure pool, lowering realized payoffs through an equilibrium channel. This sorting creates a privacy externality whose extent is bounded by the cognitive distribution. Whether transparency increases average welfare, therefore, depends on the joint distribution of cognitive types and data values in the population.

Platforms increasingly offer privacy controls, but does being transparent about these controls help or hurt the platform? We conduct a large-scale field experiment on one of the largest global platforms, with over 8,000 users, in which the treatment group receives information highlighting centralized data authorization tools, while the control group does not. Using field experiment and difference-in-differences specifications, we find that the treated group exhibits both higher direct engagement with the privacy tools and higher downstream platform engagement and trust. First, direct engagement with the privacy tools substantially increases. Second, downstream platform engagement and trust increase, with page views, active duration, and asset balances all rising, while contextual app-level data consent decisions remain unchanged at users' own privacy preference levels. Third, the direct and downstream effects are driven by distinct user segments: tool usage increases are concentrated among low-knowledge users (closing an awareness gap), while engagement and trust increases are concentrated among intrinsically privacy-sensitive users (consistent with enhanced perceived control). Consumers bear real privacy costs on platforms. Transparency decreases these costs by improving the decision environment through greater perceived control and clarity, while leaving consent-level data sharing with third-party apps unaffected. Transparency pays.

Online platforms enable firms, creators, and communities to grow by initiating lightweight interactions (e.g., following and resharing) that may lead users to reciprocate. Because reciprocation depends not only on a target user’s attributes but also on endogenous social proof created by previously acquired connections, outreach is a sequential decision problem on a directed network. We study the follow-back problem: given a set of target users and a budget of outreach actions, choose an interaction sequence to maximize the expected number of targets who follow back. Using a large set of real platform interactions, we find that the degree of the target and the size of the mutual neighborhood of the agent and target in the network affect the probability of a target following an agent. Based on our empirical findings, we then propose a model for targets following an agent. Using this model, we solve the follow-back problem exactly on directed acyclic graphs and derive a closed-form expression for the expected number of follows an agent receives under the optimal policy. We then formulate the follow-back problem on an arbitrary graph as an integer program. To evaluate our integer-program-based policies, we conduct simulations on real social network topologies. The results show that our policies substantially outperform random and centrality-based policies, revealing that effective growth strategies prioritize high-susceptibility “stepping-stone” users who create overlap with high-value targets rather than globally central users.

Privacy law evaluates consent one prompt at a time. On digital platforms that view is incomplete: users meet repeated requests through a shared interface, and each choice can shift the next. This article treats consent as a platform-mediated market process, using panel data from a leading global all-round service platform together with a randomized field experiment in the same environment. What a screen-by-screen focus misses is that consent decisions do not stay local; they spill across the platform through two externalities. The first is a cross-app externality: one app’s data request can shift the user’s later consent rate for another app’s request. The second is a trust externality: poorly designed or hard-to-reverse consent affects not just one authorization but also users’ trust in the platform environment. Better consent infrastructure redirects data toward firms users genuinely trust and away from those that profit on friction, turning privacy into a dimension of quality competition rather than a compliance cost.