On incentives to manipulate online ratings

With the proliferation of online rating platforms, there has been an increasing concern over the authenticity of reviews posted online. While much effort has been dedicated to improving fake review detection algorithms, little attention has been spent on understanding the incentives that drives some sellers to solicit fake reviews. To fill this gap, this paper develops a theoretical model in which sellers dynamically choose their effort spent on review manipulation. Among other things, the model predicts that sellers' optimal investment in fake reviews is not a monotone function of their reputation. More precisely, sellers that currently possess a very good or very bad history of past reviews have less incentives to solicit fake reviews praising their own products, the intuition being that, for sellers with very bad reputation, it is too costly to pretend that they are high quality sellers; while sellers that have already accumulated a very good reputation do not need to spend much effort in convincing buyers that they are high quality sellers. Another prediction from the model is that, in order to maximize the impact from each fake review, sellers tend to concentrate review manipulation at the initial stages after they have entered the market. Using data collected from Amazon, I was able to observe those two features from the model on the empirical level by estimating a Logit regression that predicts the probability of a review being fake as a function of the product's reputation and the time it took for the review to be posted since the seller entered the market.