Integrated Heteroscedasticity Test for Vehicular Traffic Condition Series

Abstract

Because of the growing awareness of the importance of the traffic condition uncertainty-related studies, traffic condition uncertainty modeling is gaining increasing attention from the transportation research community. In this field, traffic condition uncertainty, gauged mainly by the conditional variance of traffic characteristics, has been investigated primarily with two major approaches, generalized autoregressive conditional heteroscedasticity approach and stochastic volatility approach; however, both lack a thorough and sound test on the applicability of these approaches. To complete this modeling gap and hence lay the theoretical basis for traffic uncertainty-related studies, an integrated heteroscedasticity test, including an optimal transformation search and four statistical tests, is proposed in this study. By using real world data collected from 36 stations across four regions in both the United Kingdom and the United States and aggregated at 15-min interval as a typical representative, the proposed integrated heteroscedasticity test is demonstrated, validating the heteroscedastic nature of the traffic conditional series. In addition, the effects of transformations are illustrated together with an online short-term traffic condition forecasting algorithm as an additional validation of this heteroscedastic nature. On firmly establishing the heteroscedastic nature of the traffic conditions, future studies are recommended to further the modeling of traffic condition uncertainties over a spectrum of time intervals and apply the uncertainty models in various applications such as travel time reliability or the proactive traffic control systems.