Manipulation and obsolescence of online information – MOII

We combine web scraping methods allowing the large-scale collection of Internet data and structural models of the behavior of the actors present on the various websites.
Regarding obsolete information, we have assembled a panel of US job ads from Craigslist, Monster and Indeed. We estimate listing mortality rates and renewal rates by age. These rates will then be integrated into the structural models to deduce the proportion of phantom vacancies, the distribution of search efforts by listing age, and the distribution of listing renewal ages.
We study in parallel the macroeconomic consequences linked to the birth and propagation of obsolete information. To this end, we use a theoretical model generating endogenous fluctuations. We calibrate the deterministic cycle of this model on macroeconomic data from OECD countries.
In terms of information manipulation, we have retrieved reviews of a large number of hotels and restaurants from TripAdvisor, Expedia, Booking, TheFork and OpenTable. We will compare the empirical distribution of such evaluations to theoretical counterfactual distributions to detect manipulation behaviors. We will then show how the local competitive context affects these behaviors. We will focus on manipulation wars between providers on this type of site. We will take into account the selection of pairs of consumers and producers in the process of forming evaluations.
In a different but complementary approach, we use data from Pôle Emploi. Some listings advertise for a specific salary, whereas others do not. We connect the decision of providing this information to job types and the extent of local competition between employers.

Our work on obsolete information has given birth to two main articles.
(i) In «Directed search with phantom vacancies« we assume job seekers direct their search based on listing age. Forming a match with an age-a vacancy creates an age-a phantom with probability ß and generates an externality affecting vacancies
aged a and older. Thus, the externality decreases with the match’s listing age. Relative to efficient behavior,
job seekers overapply to younger listings. We calibrate the model using U.S. data. The contribution of phantoms to inefficiency is large.
(ii) The article «Phantom cycles« describes the macroeconomic implications of obsolete information. Phantom vacancies, or phantoms for short, are jobs that everyone can see but that none can get because they are already filled. Phantom cycles are deterministic fluctuations caused by the dynamics of phantoms and its interaction with vacancy supply. Phantom cycles are more relevant for sclerotic labor markets (e.g., France) than for countries with high worker turnover (e.g., the US). We use a model of equilibrium search unemployment with phantom vacancies. This model generates limit cycles associated to a Hopf bifurcation. We calibrate phantom cycles on aggregate data for 6 OECD countries. The expected duration of phantoms increases with the steady-state job-finding rate, reaching implausibly low values in the US where unemployment spells are notoriously short.
Our work on information manipulation has given birth to two main articles.
(iii) In «Can Information Reduce Ethnic Discrimination? Evidence from Airbnb« we use data from Airbnb to identify the mechanisms underlying discrimination against ethnic minority hosts. Within the same neighborhood, hosts from minority groups charge 3.2 percent less for comparable listings. Since ratings provide guests with increasingly rich information about a listing's quality, we can measure the contribution of statistical discrimination. This form of discrimination can account for the whole ethnic price gap. Also, three-quarters (2.5 points) of the initial ethnic gap can be attributed to inaccurate beliefs of potential guests about hosts' average group quality.
(iv) In «Looking for the `Best and Brightest': Hiring difficulties and high-skilled foreign workers« we show that US employers are more likely to seek foreign skilled workers for positions where finding domestic workers takes time. We use a new dataset matching online job posting duration to administrative data on Labor Condition Applications submitted as the first step in applying for H-1B temporary skilled worker visas. We investigate the mechanisms by exploring the heterogeneity of the results across firms and labor markets. This relationship is not driven by firms manipulating their job postings’ duration to demonstrate their good faith efforts in their search for domestic workers. On the contrary, the results are due to the insufficient domestic labor supply in tight occupations.

The main scientific objective is to finalize existing articles. There are thus 8 academic articles in progress, of which two have been accepted for publication, three submitted for publication and three still need to be modified before being submitted for publication.
The second scientific objective is the organization of a scientific event in Aix-en-Provence.
This is a symposium devoted to the economic study of digital platforms, the one we should have already organized in 2020 or 2021. This will take place in Aix-en-Provence and will last two days. The team will present two current articles there, the article that founds the empirical side of the project devoted to obsolete information, and the article devoted to the manipulation of hotel and restaurant reviews.

Albrecht, J., Decreuse, B., Vroman, S., 2022. Directed search with phantom vacancies. International Economic Review
Chéron, A., Decreuse, B., 2021. Phantom cycles. Mimeo, à soumettre pour publication
Decreuse, B., Wilemme G., 2021. Age discontinuity and nonemployment benefit policy evaluation through the lens of job search theory. Mimeo, soumis pour publication
Decreuse, B., Flachaire, E., Hachème, G., 2021. Why employers post wages in job ads: evidence from French online data. Mimeo, à soumettre pour publication
Decreuse, B., Raux, M., Sangnier, M., 2021. Ghosbusters. Mimeo, à finaliser.
Decreuse, B., Laouenan, M., 2021. Death and resurrection on the AirBnB platform. Mimeo, à finaliser
Laouenan, M., Rathelot, R., 2022. Ethnic Discrimination on an Online Marketplace of Vacation Rentals. American Economic Journal.
Raux, M., 2021. Looking for the `Best and Brightest': Hiring difficulties and high-skilled foreign workers. Mimeo, soumis pour publication.

The project focuses on the quality of information on digital marketplaces and other web platforms. It is composed of two axes: information obsolescence and information manipulation. New technologies of communication have facilitated the emergence of two-sided markets where platforms collect and diffuse information about market participants. We question information efficiency in light of two negative externalities: employers have little incentive to withdraw ads for already filled jobs and hotels and restaurants can manipulate customers' evaluations. We plan to measure the magnitude of these externalities. We will combine web scraping methods to obtain data from various websites with structural models of agents’ behaviors and their interactions.
On the one hand we will build a panel dataset of ads from three US job boards - Craigslist, Monster and Indeed - and the French website of Pôle Emploi. We will estimate the mortality rate and renewal rate of job listings by listing age. These hazard rates will be introduced in structural models to compute phantom job proportions, the distribution of applications by listing age, the distribution of listing renewal age and the overall impact of obsolete information on unemployment and matching efficiency. On the other hand, we will use the evaluations received by a large number of hotels and restaurants on TripAdvisor, Expedia, Booking, TheFork and OpenTable. We will confront the empirical distribution of evaluations of each hotel or restaurant with theory-grounded counterfactual distributions. This will allow us to detect information manipulation behaviors. We will show how the local intensity of competition affects manipulation. We will also highlight wars of information manipulation. The research will pay attention to the sorting of consumers and producers in the evaluation-making process.
The consortium is composed of five senior researchers and one doctoral student. Three empirical researchers have extensive knowledge of web scraping methods; two theoretical researchers are very familiar with structural models based on search and matching frictions. The project will also involve a network of coauthors based in the US and the UK, Canada and Australia. The results will be publicized on a dedicated website. It will also provide (anonymized) datasets as well as statistical and numerical simulation programs to facilitate replication and stimulate outside research. The project will last 36 months and cost 208,527 euros. This amount corresponds to hardware expenses, and in particular workstations, useful for web scraping activities and numerical simulations of the structural models, mission expenses in France and in foreign countries, expertise costs in computer science and costs of a two-day conference.