Learning from the Offline Trace: A Case Study of the Taxi Industry

The growth of mobile and sensor technologies today leads to the digitization of individual's offline behavior. Such large-scale and fine-grained information can help better understand individual decision making. We instantiate our research by analyzing the digitized taxi trails to study the impact of information on driver behavior and economic outcome. We propose homogeneous and heterogeneous Bayesian learning models and validate them using a unique data set containing complete information on 10.6M trip records from 11,196 taxis in a large Asian city in 2009. We find strong heterogeneity in individual learning behavior and driving decisions, which significantly associate with individual economic outcome. Interestingly, our policy simulations indicate information that is noisy at individual level can become most valuable after being aggregated across various spatial and temporal dimensions. Overall, our work demonstrates the potential of analyzing the digitized offline behavioral trace to infer demand as well as to improve individual decision efficiency.