LEARNING-BY-DOING IN SINGAPORE’S TAXI MARKET

Abstract

We study taxi drivers’ location choices in Singapore’s taxi market. Using a unique dataset consisting of taxi drivers’ trip records and GPS trajectories, we estimate a dynamic discrete choice model with a Dirichlet learning framework incorporated, in which taxi drivers choose where to search for passengers when they are vacant and update their belief regarding the demand condition by accumulating multiple types of experiences. Our results demonstrate that the experience of finding street-hail trips is the most impactful one in updating the probability of finding passengers, followed by the experience of finding booking trips and not finding trips. We further document that the effect sizes of the different types of experiences are affected by (1) the extent of gain or loss of the corresponding experience, (2) the rarity of the corresponding experience, and (3) the amount of the same type of experience accumulated in history.