Research

We exploit the differential responses of Airbnb hosts to two distinct policy interventions in Berlin to shed light on the optimal design of policies targeting short-term rental platforms to mitigate rental market inflation. The first intervention, which affected commercial listings, significantly impacted long-term rental markets, unlike the second intervention, which mainly affected non-commercial listings. Leveraging these policy variations, we estimate the marginal impact of Airbnb on rental supply and rents. Each additional commercial Airbnb listing displaces 0.23 to 0.37 rental units and increases rent per square meter by 1.3 to 2.4 percent. This underscores the importance of targeting commercial listings when regulating short-term rental markets.

The ability to make accurate predictions relating to consumer preferences is a key factor of a digital firm's success. Examples include targeted advertisements and, more broadly, business models relying on capturing consumers' attention. The prediction technologies used to learn consumer preferences rely on consumer generated data. Despite the importance of data-driven technologies, there is a lack of knowledge about the precise role that data-scale plays for prediction accuracy. From a policy perspective, a better understanding about the role of data is needed to assess the risks that “big data” might pose for competition. This article highlights potential complementarities between different data dimensions in algorithmic learning. We analyze our hypothesis using search engine data from Yahoo! and provide evidence that more data in the within-userdimension enhances the efficiency of algorithmic learning in the across-user dimension. Our findings suggest that ignoring these complementarities might lead to underestimating scale advantages from data.

It is now generally understood that personal data––that is, data that relate to individual consumers––drive digital markets. Personal data underlie targeted advertising, which draws billions of dollars into ad-supported markets. Personal data are useful for other purposes as well. Firms in digital markets rely on personal data to deliver their core products and services––we refer to these collectively as “web services”1––to hone and improve them, and to recommend related products and services. These data facilitate innovation, allowing yet more services and “smart” products with increasingly personalized functionalities. Personal data can allow governments to deliver better public services, such as transportation systems, or can help researchers better understand how humans interact with algorithms and which policies might best serve society. And data can also facilitate competition, by improving quality and providing insight into consumer conduct that encourages entry. In these various ways, the massive quantity of personal data currently collected undoubtedly contributes to consumer welfare. But there also are downsides to the collection and use of personal data on such a grand scale. “Surveillance capitalism,” as Professor Shoshana Zuboff has termed it, has blurred the line between the personal and the public, and has commodified our habits, interests, and beliefs in ways that can feel distasteful and invasive. Massive data collection also has made information about us more accessible to government and commercial actors who often face little to no accountability for its misuse.

Algorithmic collusion among pricing algorithms has raised concerns about
sustained supra-competitive prices and their implications for social welfare.
Existing work has largely focused on the probability that reinforcement-learning algorithms converge to cooperative strategies, typically under the assumption that exploration vanishes over time. Motivated by the observation that algorithms deployed in practice are likely to continue exploring in order to remain adaptive to changing environments, we study learning dynamics under constant exploration. In this setting, the relevant question is no longer whether an algorithm converges to a particular strategy profile, but rather what fraction of time the algorithms spend playing cooperative strategies. Even in the benchmark case of the repeated Prisoner's Dilemma with one-period memory, this yields high-dimensional stochastic learning dynamics, for which a complete analytic treatment is intractable. We show that cooperative strategies can be dominant in this time-averaged sense and derive a boundary predicting when such dominance arises, based on the expected dynamics of the Q-learning process. Extensive simulations show that this boundary is a strong predictor for non-defection-dominated behaviour under epsilon-greedy Q-learning.

Using data from Paris in 2017, we estimate demand for short-term accommodations, explicitly accounting for localized variation in demand across the city. Our counterfactual simulations show that Airbnb increases total consumer surplus by 924 million euros, affords Airbnb hosts a surplus of 21 million euros, while reducing total hotel profits by 778 million euros, resulting in an overall welfare gain of 167 million euros. Airbnb’s value to consumers is highest when demand is high and hotels operate close to capacity constraints. The impact of Airbnb on consumers and hotels is heterogeneous across the city: Hotels in outer districts would gain most from a ban of Airbnb. Conversely, consumer surplus would be reduced the most from a ban of Airbnb in these outer districts.

Digital platforms increasingly observe individuals' browsing behavior beyond the boundaries of their own services. This paper studies whether such off-platform tracking generates data externalities by enabling platforms to infer personal characteristics of individuals who do not disclose them. We examine this mechanism in the context of Facebook's tracking technologies embedded on third-party websites. Using clickstream data on about 40,000 individuals, we document that Facebook can observe a substantial share of browsing activity both for its own users and for individuals outside its user base. We then train prediction models on Facebook users, for whom demographic characteristics are available to the platform, and apply these models to the trackable browsing behavior of non-users. The results show that demographic characteristics of non-users can be inferred above a zero-information benchmark, implying economically meaningful data externalities. We then study how privacy regulation affects these externalities, in particular, the introduction of the General Data Protection Regulation (GDPR). Although the GDPR sharply reduced Facebook's off-platform tracking ability, its effect on prediction accuracy for non-users was much more limited. The findings suggest that privacy regulation focused on limiting data collection may leave important inference-based privacy risks unresolved when platforms can use data from some individuals to learn about others.

We investigate the impact of prices on ratings using Airbnb data. We theoretically illustrate two opposing channels: higher prices reduce the value for money, worsening ratings, but they increase the taste-based valuation of the average traveler, improving ratings. Results from panel regressions and a regression discontinuity design suggest a dominant value-for-money effect. In line with our model, hosts strategically complement lower prices with higher effort more when ratings are relatively low. Finally, we provide evidence that, upon entry, strategic hosts exploit the dominant value-for-money effect. The median entry discount of seven percent improves medium-run monthly revenues by three percent.

Using simulations between pairs of ϵ-greedy q-learners with one-period memory, this article demonstrates that the potential function of the stochastic replicator dynamics (Foster and Young, 1990) allows it to predict the emergence of error-proof cooperative strategies from the underlying parameters of the repeated prisoner's dilemma. The observed cooperation rates between q-learners are related to the ratio between the kinetic energy exerted by the polar attractors of the replicator dynamics under the grim trigger strategy. The frontier separating the parameter space conducive to cooperation from the parameter space dominated by defection can be found by setting the kinetic energy ratio equal to a critical value, which is a function of the discount factor, f(δ)=δ/(1−δ), multiplied by a correction term to account for the effect of the algorithms' exploration probability. The gradient at the frontier increases with the distance between the game parameters and the hyperplane that characterizes the incentive compatibility constraint for cooperation under grim trigger.
Building on literature from the neurosciences, which suggests that reinforcement learning is useful to understanding human behavior in risky environments, the article further explores the extent to which the frontier derived for q-learners also explains the emergence of cooperation between humans. Using metadata from laboratory experiments that analyze human choices in the infinitely repeated prisoner's dilemma, the cooperation rates between humans are compared to those observed between q-learners under similar conditions. The correlation coefficients between the cooperation rates observed for humans and those observed for q-learners are consistently above 0.8. The frontier derived from the simulations between q-learners is also found to predict the emergence of cooperation between humans.

Data is often at the core of digital products and services, especially when related to online advertising. This has made data protection and privacy a major policy concern. When surfing the web, consumers leave digital traces that can be used to build user profiles and infer preferences. We quantify the extent to which Facebook can track web behavior outside of their own platform. The network of engagement buttons, placed on third-party websites, lets Facebook follow users as they browse the web. Tracking users outside its core platform enables Facebook to build shadow profiles. For a representative sample of US internet users, 52 percent of websites visited, accounting for 40 percent of browsing time, employ Facebook's tracking technology. Small differences between Facebook users and non-users are largely explained by differing user activity. The extent of shadow profiling Facebook may engage in is similar on privacy-sensitive domains and across user demographics, documenting the possibility for indiscriminate tracking.