Alone, Together: A Model of Social (Mis)Learning from Reviews [Paper]
We develop a model of social learning from consumer ratings in a horizontally differentiated market. At the core of our model lies the dynamic feedback loop between choices, ratings and beliefs, of which we characterize the long-run properties. We show that long-run, equilibrium ratings are biased in systematic and sizable ways. They relatively advantage lower quality and more polarizing products. Thus, in stark contrast with the winner-takes-all dynamics of classic observational learning models, learning from consumer opinions (as opposed to actions) generates excessive choice fragmentation compared to the normative optimum. Our results are robust to different assumptions about consumer learning and rating behavior. We provide corroborative evidence for our results using data obtained from Goodreads, a popular consumer book ratings platform, and Book Marks, a professional book critics ratings aggregator. Our findings have implications for the optimal design of crowdsourced exploration, cast apparent cognitive biases such as the “love for large numbers” in a new light, and inform the debate on the impact of fake reviews.
Amazon and the Evolution of Retail (with Luís Cabral) [Paper]
The growth of Amazon and other online retailers questions the survival of bricks- and-mortar retail. We show that, in response to the online trend, offline retailers – especially smaller ones – optimally follow a specialization strategy, in particular specialization in nar- row niches. This may lead to an offline long tail that is thicker than the online long tail, contrary to existing research. Offline specialization benefits consumers; in fact, consumers would benefit from more specialization than it results in equilibrium. We discuss this and other relevant comparative statics based on a simple model of consumer demand and retail design. We complement our theoretical analysis with corroborative empirical evidence. To do so, we employ a large proprietary dataset obtained from a major US publisher detailing all sales to book retailers (both online and offline) over the 2016-2019 period.
When to Talk Politics in Business: Theory and Experimental Evidence of Stakeholder Responses to CEO Political Activism (with Vanessa Burbano and Fabrizio Dell'Acqua) [Paper]
CEO political activism, wherein firm leaders communicate public stances on overtly political issues unrelated to their core business, is a nascent and emerging, and thus understudied, phenomenon. We first propose a parsimonious model of firm political communication. In our model, stakeholders value both ideological proximity to firm’s political stances, and the perceived sincerity of such stances. The model generates a variety of predictions regarding when, and why, communication is beneficial to firms. We test these predictions by conducting two survey-based experiments to examine individuals’ responses to CEO political activism. Theoretically and empirically, we find that, on average, CEO political activism elicits a negative response amongst stakeholders, and that donation-backed communications can elicit stronger responses. However, because communication polarizes perceptions, the percentage of stakeholders holding extremely positive views increases as a result of political communication. We provide evidence of both a congruence benefit and an incongruence benefit (rather than the more commonly identified incongruence penalty) in the context of CEO political activism. Finally, we discuss some extensions to our model, as well as heterogeneous effects discovered in our experimental data, and conclude with additional strategic con- siderations. We thus shed light on circumstances under which it is more or less beneficial to talk politics in business.
The Good, the Bad and the Picky: Reference Dependence and the Reversal of Movie Ratings (with Michelangelo Rossi and Ryan Stevens) [Paper]
We explore the consequences of referent-dependent preferences on the nature of online reviews. Consumers differ in their experience, which has two effects. First, experience is instrumental to choice: experts purchase better products than non-experts. Second, because of their superior choices, experts endogenously form higher reference points, and post harsher ratings for given quality. Combined, these two facts imply a bias against higher quality products. When this bias gets large, ratings are non-monotonic in quality: higher-quality products can obtain lower ratings than their inferior alternatives. We test our theory using two large datasets obtained from well known movie rating websites and find strong support for it. We proxy users’ expertise with the total number of ratings posted on the platforms. Using external measures of quality, such as the Academy Awards, we show that experts rate movies of higher quality compared to non-experts. Moreover, experts post more stringent ratings for the same movies. Finally, we debias the ratings exploiting the full history of users’ ratings to level up their stringency levels. This approach leads to normalized aggregate ratings that reduce the bias against higher-quality products and are more in line with external measures of quality