Are Ratings Informative Signals? The Analysis of the Netflix Data

Abstract

This paper analyzes ratings as informative signals about the quality of movies. A structural Bayesian learning model links the revealed experience utilities of raters, who are prior consumers, to the product choice of the future consumers of the same good. I postulate that movies are chosen based on the consumers prior beliefs and the precision of the signals provided by the ratings. Consumers use the ratings signals more when more consumers have revealed their preferences in the ratings. I specify and estimate a simulated maximum likelihood model using the Netflix data on rental choices and ratings. The very rich data set allows me to identify the effect of ratings on demand while controlling for the inherent popularity of each specific DVD using fixed effects. The results demonstrate that the ratings provide signals of quality to consumers. If the signal is based on only one rating, it is very noisy, and the consumer might ignore it. As more consumers rate the DVD, the signal becomes more informative, and the results show that the consumers surplus increases. The estimation shows that the ratings system has economically significant value. As 100 more people rate a DVD, the quantity demanded for the newly released DVD can rise by as much as 35 percent. If Netflix were to offer DVDs without providing the rating service, 88 percent more consumers would choose movies elsewhere. Finally, the absence of the Netflix system with a high volume of ratings for each DVD would cause consumers to rent a narrower selection of DVDs than what they currently rent from Netflix. Specifically, the share of newly released DVDs would go up from 5 to 17 percent.