Estimating Search with Learning

Abstract

In this paper we estimate a structural model of search for differentiated products, using a unique dataset of consumer online search for hotels. We propose and implement an identification strategy that allows us to separately estimate consumer's beliefs, search costs and preferences. Learning plays an essential role in this strategy. It creates variation of posterior beliefs across consumers that is orthogonal to the variation in search costs. We show that ignoring endogeneity of choice sets due to search may lead to significant biases in estimates of consumer demand: from 50 to more than 200 percent depending on informational assumptions. Second, th median search cost is about 25 dollars per 15 hotels; there is also a significant heterogeneity of search costs among the population. We perform a statistical test between models of search from known (Stigler 1967) and from unknown (Rothschild 1974) distribution and find that our data favors the latter: we find a statistically significant amount of Bayesian learning.