Machine Learning from Big GPS Data About the Heterogeneous Costs of Congestion

The paper presented in this ZEW Research Seminar provides a novel approach to estimate the heterogeneous costs of road congestion and the welfare foregone when failing to differentiate corrective taxes based on 34 million GPS-coded trips from moving vehicles in Berlin. Using unsupervised learning, the authors assign anonymous trips to individual drivers and track their repeated travel behavior throughout a full year. They infer traffic density along the route taken and all potential alternatives to account for the equilibrium responses to rerouting. The identification of the causal effect of traffic density on the time cost of travel relies on a new instrumental variable strategy exploiting intra-weekday traffic patterns. The paper presented in this ZEW Research Seminar finds significant temporal heterogeneity in the daytime marginal external costs of congestion between 2.8 and 34 € cents per vehicle kilometer. Large welfare gains materialize with time-specific congestion taxes. A uniform tax achieves only 32% of the optimal gains.