Essays on reputation in online marketplaces
- Autores: Michelangelo Rossi
- Directores de la Tesis: Natalia Fabra Portela (dir. tes.), Matilde Pinto Machado (codir. tes.)
- Lectura: En la Universidad Carlos III de Madrid ( España ) en 2020
- Idioma: español
- Tribunal Calificador de la Tesis: Juan José Ganuza Fernández (presid.), Alan Crawford (secret.), Chiara Farronato (voc.)
- Programa de doctorado: Programa de Doctorado en Economía por la Universidad Carlos III de Madrid
- Materias:
- Enlaces
- Tesis en acceso abierto en: e-Archivo
- Resumen
Would you ever rent a house you have never seen; whose owner you have never met; in a city you have never visited? And again: would you ever pay upfront to a person you do not know, and you will never meet, who promises to deliver an object you have never seen? A few decades ago the answers to these questions would have been negative for most customers. Conversely, nowadays millions of users rely on digital platforms, such as Airbnb and eBay, to get services with the same characteristics as those described by the previous two questions. How was that possible? How did users around the world start to trust each other after millennia of skepticism and malevolence? An answer to such questions relies on the innovative way digital marketplaces use to reduce the asymmetry of information between parties: review systems. In almost all digital platforms, users can review the services they have experienced providing new pieces of information to prospective users. Accordingly, reviews reduce the uncertainty about sellers' quality since each feedback increases the precision of buyers' estimates. Besides, reviews also discipline sellers' on-going behavior with the potential punishment of negative feedback. Still, signaling quality and monitoring sellers' behavior are two separate tasks. From a microeconomic perspective, reviews reduce adverse selection effects by signaling sellers' quality, whereas monitoring behavior affects moral hazard issues.
In this dissertation, I study the power, and the limits, of review systems to reduce these two types of asymmetry of information: adverse selection and moral hazard. In the two chapters of this thesis, I examine both signaling and monitoring tasks.
In the first chapter, "How Does Competition Affect Reputation Concerns? Theory and Evidence from Airbnb", I show how changes in the number of close competitors affect the power of reputation to induce sellers to exert effort. The impact of competition on sellers' incentives is theoretically ambiguous. More competition disciplines sellers, but, at the same time, it erodes reputational premia. This paper identifies empirically whether one effect dominates the other using data from Airbnb.
I develop a model of reputation building with congested marketplaces and frictional matching to accommodate the feature of digital platforms and inform my empirical analysis. In this framework, the number of hosts and guests on the two sides of the market (the market tightness) impacts the reputation returns of hosts' effort. The model predicts that, when the number of competitors decreases, hosts' profits increase and hosts exert more effort. With fewer competitors, hosts have a higher probability to be matched with a guest and can charge higher prices. However, the price elasticity of the hosts' demand depends on their reputation. In particular, the probability to be matched with a guest is less elastic to variations in prices for hosts with a good reputation. Accordingly, the premium of exerting effort (and having a good reputation) increases when the number of competitors is lower: hosts with a good reputation can post higher prices with a lower reduction in their demand. Thus, hosts exert more effort when the number of competitors decreases.
In order to test the model's prediction, I analyze empirically the relationship between the effort exerted by Airbnb hosts and the number of their competitors. I measure hosts' effort by studying ratings such as "communication" and "check-in" that are specifically related to hosts' actions. Moreover, to measure the number of competitors for each host, I create host-specific consideration sets by counting the number of listings surrounding each host within a radius of $0.5$, $1$, and $2$ kilometers. Doing so, I assume that Airbnb hosts compete more strongly with listings that are closely located to them, relative to those further away.
My identification strategy exploits a unique quasi-experiment to isolate the effect of changes in the number of competitors from other confounders. In particular, I take advantage of regulatory enforcement on short-term rentals that occurred in San Francisco in 2017: the Settlement Agreement between the San Francisco City Council and Airbnb.
I exploit this regulatory enforcement as an exogenous shift in the number of listings surrounding each host. I focus on hosts renting short-term that were present on the platform both before and after the Settlement Agreement. By such selection, I abstract from hosts' decision to enter or exit due to the regulation enforcement. All hosts renting short-term in San Francisco are affected by the Settlement Agreement. On the other hand, the exposure to this ``shock'' differs since the variation in the number of competitors is heterogeneous across hosts. I take advantage of this heterogeneity in the treatment. To measure the exposure of each host, I use the percentage of listings surrounding each host that were already registered in September 2017. For higher values of this percentage, fewer listings are likely to exit after the Settlement Agreement since they were already complying with the regulation. I employ this measure as a predictor for the variation in the number of listings surrounding each host after the Settlement Agreement.
Therefore, I identify the effect of variations in the number of competitors by using the differential changes in the exposure across listings over time. The core identifying assumption of this design shares the intuition of a difference in differences estimator with a continuous treatment. However, the identification is based on an instrumental variable regression where the excluded instrument is given by the interaction between the measure of exposure (the percentage of listings already registered in September 2017) and time.
The results show a statistically significant negative relationship between the number of competitors and hosts' effort. I corroborate this result studying variations in hosts' profits and in the monetary value of reputation. With the same identification strategy, I find that less competition increases profits so that hosts have higher incentives to exert effort. Moreover, since fewer hosts are going to have a good reputation, I show evidence that the signaling effect of reputation is stronger in more competitive frameworks.
In the second chapter, "Quality Disclosures and Disappointment: Evidence from the Academy Awards", I study the impact of quality disclosures on buyers' rating behavior using data from an online recommender system. Certifications, and in general, third-party quality disclosures are often used in markets with asymmetries of information. In these circumstances, buyers are not perfectly informed about the seller's quality. Thus, a third-party certification may help to reduce the uncertainty on the buyers' side and increase their willingness to pay: used cars' sellers may show the most recent inspections by the car's manufacturer to assure prospective buyers about the good state of the car. When a certifier is credible, a third-party disclosure can effectively increase buyers' expectations and attract high-quality sellers to trade. Still, altering buyers' expectations could have unexpected side-effects when buyers' utility depends on reference points induced by their expectations.
This paper empirically identifies the disappointment effect due to quality disclosures estimating the causal impact of the nominations for the Academy of Motion Picture Arts and Sciences (AMPAS) awards on movie ratings. Nominations constitute the "shifter'' of reference point about the movies' quality; whereas the variations of ratings displayed on an online movie recommender system (MovieLens) provide a measure for the disappointment effect. My findings show that, after nominations, ratings for nominated movies significantly drop relative to ratings for not nominated movies with similar characteristics. In particular, the drop in ratings due to disappointment accounts for more than 5 percent of the rating premium for nominated movies.
This empirical exercise may help to shed light on the welfare impact of quality disclosures in a setting of asymmetry of information and expectation-dependent preferences. With this work, I do not question the positive impact of certifications and awards on sellers' performance. Still, here I document that quality disclosures also produce depressing effects on buyers' satisfaction (ratings) due to disappointment. This latter effect may reduce the positive effects of certifications.
