Working papers

This figure shows the dependence of search result quality on the amount of data that a search engine uses to produce search results. There are 5 lines, for 5 levels of popularity of the query.

How important are user-generated data for search result quality? Experimental evidence

joint with Madina Kurmangaliyeva, Jens Prüfer, and Patricia Prüfer

September 2022

TILEC Discussion Paper No. 2022-16, CCP Working Paper No. 22-07, and CCP Policy Brief


Do some search engines produce better search results because their algorithm is better, or because they have access to more data from past searches? In the latter case, mandatory data sharing, a policy that is currently discussed, could trigger innovation and would benefit all users of search engines. We document that the algorithm of a small search engine can produce non-personalized results that are of similar quality than Google’s, if it has enough data, and that overall differences in the quality of search results are explained by searches for less popular search terms. This is confirmed by results from an experiment, in which we keep the algorithm of the search engine fixed and vary the amount of data it uses as an input.


This figure shows the evolution of beliefs about seller type (the probability that it's a good type) for one simulated seller (blue) and the simulated percentage positive feedback over time (dashed red). Obtained after calibrating the model to our data.

When and Why Do Buyers Rate in Online Markets?

joint with Xiang Hui and Konrad O. Stahl

February 2022

C.E.P.R. Discussion Paper 17006 and CRC TR 224 Discussion Paper No. 267


Online ratings play an important role in many markets. We study the often disputed information content of these ratings, by proposing a reduced-form Bayesian model of the typical buyer's rating decision. Our empirical evidence based on eBay raw data is in line with even intricate predictions from it. We thus have good reasons to calibrate the model to moments of the data. Our simulations suggest that the rating record reveals the seller's type after about 100 transactions, or 65-70 ratings.


This figure shows the evolution of the price patients paid for a dermatological procedure. The treatment group had transparent prices from week 31 onward, while the control group did not.

Increasing price transparency in the Dutch health care market does not affect provider choice

joint with Maciej Husiatyński and Misja Mikkers

March 2021

C.E.P.R. Discussion Paper 15981


Price transparency is often viewed as an effective way to encourage price shopping and thereby lower health care expenditure. Using individual claims data for 6 frequent, non-emergency dermatological procedures, we estimate the short-run effect of unexpected publication of prices by a major Dutch health insurer on spending and provider choice. Visits to the price transparency website surged, but spending, the likelihood to visit a new provider, distance traveled, and type of provider visited remained unaffected.


Price Competition in Two-Sided Markets with Heterogeneous Consumers and Network Effects

joint with Lapo Filistrucchi

NET Institute Working Paper #13-20

We model a two-sided market with heterogeneous customers and two heterogeneous network effects. In our model, customers on each market side care differently about both the number and the type of customers on the other side. Examples of two-sided markets are online platforms or daily newspapers. In the latter case, for instance, readership demand depends on the amount and the type of advertisements. Also, advertising demand depends on the number of readers and the distribution of readers across demographic groups. There are feedback loops because advertising demand depends on the numbers of readers, which again depends on the amount of advertising, and so on. Due to the difficulty in dealing with such feedback loops when publishers set prices on both sides of the market, most of the literature has avoided models with Bertrand competition on both sides or has resorted to simplifying assumptions such as linear demands or the presence of only one network effect. We address this issue by first presenting intuitive sufficient conditions for demand on each side to be unique given prices on both sides. We then derive sufficient conditions for the existence and uniqueness of an equilibrium in prices. For merger analysis, or any other policy simulation in the context of competition policy, it is important that equilibria exist and are unique. Otherwise, one cannot predict prices or welfare effects after a merger or a policy change. The conditions are related to the own- and cross-price effects, as well as the strength of the own and cross network effects. We show that most functional forms used in empirical work, such as logit type demand functions, tend to satisfy these conditions for realistic values of the respective parameters. Finally, using data on the Dutch daily newspaper industry, we estimate a flexible model of demand which satisfies the above conditions and evaluate the effects of a hypothetical merger and study the effects of a shrinking market for offline newspapers.