Modeling vehicle kilometers traveled: A cross-comparison of estimation methods

The recently observed “peak car” phenomenon in several countries has motivated numerous studies in the literature to model Vehicle Kilometers Traveled (VKT) or Vehicle Miles Traveled (VMT) using different methodological approaches. However, there has been little focus on evaluating statistical methodologies being used in modeling VKT. Applying different approaches without considering the methodological strengths and limitations of each approach may create inconsistency in the observed patterns, proposed policies, and forecasting results. In this paper, we present a cross-comparison of different estimation methods to model VKT. We provide a comparative analysis and discussion of Linear Least Squares Regression, ARIMA, and ARIMAX models. For validation purposes, we re-estimate the models using different time windows in the data. We show that all three models overestimate the VKT per capita and are unable to predict the peak car phenomenon. Therefore, we argue that economic and socio-demographic variables are insufficient to capture changes in VKT in Australia.