Periodic Reporting for period 1 - DMPDE (Data and Market Power in the Digital Economy)
Période du rapport: 2023-09-01 au 2026-02-28
Improvements in technology are enabling firms to collect and process large amounts of data, about both their customers and their competitors. Customer data includes personal information, browsing history, and even social media activity; competitor data includes real-time price information. Firms’ access to such data fundamentally changes how they compete---in terms of the sophistication of their strategy (via data-driven AI), their ability to tailor offers to individual consumers (via data-driven personalisation), and their incentives to offer bundles of products (via data-driven ecosystems). Hence there is considerable policy interest surrounding the collection and use of data, including the European Union’s General Data Protection Regulation (GDPR), Data Act, and Digital Markets Act (DMA).
The overarching goal of this project is to understand the relationship between firms’ data collection and market power, and to explore how policy should be adapted to prevent harm to consumers. There are four interrelated workstreams.
1) Big tech companies increasingly operate as ecosystems, offering many different services. These services collect data and share it with each other. This workstream explores whether this “cross-market data usage” is harmful to consumers, and how it should be regulated. It also explores firms’ incentives to become ecosystems, either through organic growth or through mergers and acquisitions, and considers the role of data in mergers.
2) Companies increasingly use data to make personalised offers, such as tailored prices and recommendations. This workstream explores novel ways in which personalisation can harm consumers, and studies appropriate policy responses.
3) Consumers spend a lot of time browsing the web, and this search activity can reveal information about their preferences. This workstream views search as a data-generating process, which firms can manipulate to learn more about consumers. It models the complex interaction between search behaviour and data-driven recommendations.
4) Firms increasingly delegate pricing and recommendations to algorithms. There is growing concern that algorithms can learn anti-competitive behaviours, as illustrated by a recent literature on algorithmic collusion. This workstream explores new ways that algorithms can relax competition. It also investigates the effectiveness of policies that restrict what data can be used to train and run the algorithms.
The overarching goal of this project is to understand the relationship between firms’ data collection and market power, and to explore how policy should be adapted to prevent harm to consumers. There are four interrelated workstreams.
1) Big tech companies increasingly operate as ecosystems, offering many different services. These services collect data and share it with each other. This workstream explores whether this “cross-market data usage” is harmful to consumers, and how it should be regulated. It also explores firms’ incentives to become ecosystems, either through organic growth or through mergers and acquisitions, and considers the role of data in mergers.
2) Companies increasingly use data to make personalised offers, such as tailored prices and recommendations. This workstream explores novel ways in which personalisation can harm consumers, and studies appropriate policy responses.
3) Consumers spend a lot of time browsing the web, and this search activity can reveal information about their preferences. This workstream views search as a data-generating process, which firms can manipulate to learn more about consumers. It models the complex interaction between search behaviour and data-driven recommendations.
4) Firms increasingly delegate pricing and recommendations to algorithms. There is growing concern that algorithms can learn anti-competitive behaviours, as illustrated by a recent literature on algorithmic collusion. This workstream explores new ways that algorithms can relax competition. It also investigates the effectiveness of policies that restrict what data can be used to train and run the algorithms.
1) Ecosystems. Along with Jidong Zhou (Yale) and Junjie Zhou (Tsinghua) I have developed a framework of cross-product data usage by an ecosystem. In our working paper “Digital Ecosystems and Data Regulation” we consider an ecosystem that is present in many different markets, and which competes in each market with a different specialised firm; competition is in both price and quality. Data collected by the ecosystem in one market enables it to improve its offering in other markets. We use our framework to evaluate policies that limit data collection by the ecosystem (e.g. the GDPR), or restrict its ability to leverage data from one market into another, or force it to share data with specialised firms (e.g. the DMA). We show these policies are beneficial to consumers if and only if the specialised firms are relatively good at improving their quality; we also show that mandating data sharing is better than restricting the ecosystem’s data usage. Here we fix the set of firms, but in “Optimal Merger Remedies” with Volker Nocke (Mannheim) we explore the role of data sharing in remedying otherwise anti-competitive mergers. We show that the non-rivalrous nature of data makes it an ideal part of a merger remedies package.
2) Personalisation. In “Personalized Pricing and Competition” (AER, 2024), joint with Jidong Zhou, we consider a discrete choice model where firms sell differentiated products. We show that personalised pricing---meaning firms observe each consumer’s tastes and tailor prices accordingly---is anti-competitive when market coverage is relatively low (e.g. when products are niche). We also demonstrate that a situation where only some firms can personalise prices (e.g. because only some firms have access to the relevant data) can be worse for consumers than when either all or no firms can do personalised pricing. Meanwhile in “Platform Disintermediation with Repeated Transactions” (Management Science, forthcoming), joint with Andreea Enache (Stockholm), we consider a model where buyers and sellers use a platform to meet. They can then transact on the platform or disintermediate and transact off it; they differ in how convenient they find transacting on the platform. We show that if the platform can learn users’ convenience benefits and offer personalised fees, users neither gain nor lose from the ability to disintermediate.
3) Search. The paper “Personalization and Privacy Choice” (RAND, forthcoming), joint with Jidong Zhou, bridges the gap between this and the previous workstream. We consider a setting where a buyer needs to search for a suitable seller, but if she shares her data a platform can offer a personalized recommendation about which seller is most suitable for her. Even when the recommendation is unbiased, buyers can be harmed due to a novel “privacy-choice externality” whereby sellers raise their prices. Along the same lines, in “Mobile Payments and Interoperability: Insights from the Academic Literature” (Information Economics and Policy, 2023), with several coauthors we discuss, amongst other things, how access to financial data can change the products offered to consumers. Meanwhile, in a working paper “Dynamic Consumer Search”, joint with Alexei Parakhonyak (Oxford), we consider a model in which consumers wish to buy products repeatedly over time, but need to search initially to find a good match. We show that when firms can observe a consumer’s search and purchase history, and tailor prices accordingly, consumers can be better or worse off depending on how forward-looking they are.
4) Algorithms. A preliminary project entitled “Algorithmic Steering and Advertising on Platforms”, joint with Justin Johnson (Cornell) and Matthijs Wildenbeest (Arizona), considers how a platform’s ranking algorithm uses data on a seller’s price to decide how prominently to display it to consumers. The plan is to have both a theoretical part and a simulation part where sellers and the platform use Q-learning algorithms to decide their actions.
2) Personalisation. In “Personalized Pricing and Competition” (AER, 2024), joint with Jidong Zhou, we consider a discrete choice model where firms sell differentiated products. We show that personalised pricing---meaning firms observe each consumer’s tastes and tailor prices accordingly---is anti-competitive when market coverage is relatively low (e.g. when products are niche). We also demonstrate that a situation where only some firms can personalise prices (e.g. because only some firms have access to the relevant data) can be worse for consumers than when either all or no firms can do personalised pricing. Meanwhile in “Platform Disintermediation with Repeated Transactions” (Management Science, forthcoming), joint with Andreea Enache (Stockholm), we consider a model where buyers and sellers use a platform to meet. They can then transact on the platform or disintermediate and transact off it; they differ in how convenient they find transacting on the platform. We show that if the platform can learn users’ convenience benefits and offer personalised fees, users neither gain nor lose from the ability to disintermediate.
3) Search. The paper “Personalization and Privacy Choice” (RAND, forthcoming), joint with Jidong Zhou, bridges the gap between this and the previous workstream. We consider a setting where a buyer needs to search for a suitable seller, but if she shares her data a platform can offer a personalized recommendation about which seller is most suitable for her. Even when the recommendation is unbiased, buyers can be harmed due to a novel “privacy-choice externality” whereby sellers raise their prices. Along the same lines, in “Mobile Payments and Interoperability: Insights from the Academic Literature” (Information Economics and Policy, 2023), with several coauthors we discuss, amongst other things, how access to financial data can change the products offered to consumers. Meanwhile, in a working paper “Dynamic Consumer Search”, joint with Alexei Parakhonyak (Oxford), we consider a model in which consumers wish to buy products repeatedly over time, but need to search initially to find a good match. We show that when firms can observe a consumer’s search and purchase history, and tailor prices accordingly, consumers can be better or worse off depending on how forward-looking they are.
4) Algorithms. A preliminary project entitled “Algorithmic Steering and Advertising on Platforms”, joint with Justin Johnson (Cornell) and Matthijs Wildenbeest (Arizona), considers how a platform’s ranking algorithm uses data on a seller’s price to decide how prominently to display it to consumers. The plan is to have both a theoretical part and a simulation part where sellers and the platform use Q-learning algorithms to decide their actions.
The work on ecosystems provides one of the first frameworks to model data linkages between a big tech company’s business units, and highlights different trade-offs faced by regulators; as such it speaks to several different policies currently being considered by policymakers in Europe and further afield. The work on personalisation challenges a common view in the existing literature that personalised pricing in competitive environments is pro-competitive; our paper goes significantly beyond one of the classic and very highly cited papers on this topic. The work on search provides a new framework to study the interplay between privacy policies and general forms of data-driven personalisation; this provides a conceptual framework to think about the good (and potentially bad) effects of privacy-enhancing policies.