Everything all the time? Re-evaluating the gains from varieties

Sample zeros---products available that do not register positive revenues due to the finite sample period---introduce several biases in the computation of the import price index developed by Feenstra (1994). We model sample zeros by considering the discrete choice model of Anderson, De Palma and Thisse (1987) with a finite number of consumers/orders. We show that there is a small upward bias when the number of orders is constant across periods. When the number of orders grows over time there is a possibly large downward bias. We illustrate both cases for the import categories and sample period analyzed in Feenstra (1994).