Modeling traffic volatility dynamics in an urban network

This article discusses the application of generalized autoregressive conditional heteroscedasticity (GARCH) time series models for representing the dynamics of traffic flow volatility. The methods encountered in the literature focus on the levels of traffic flows and assume that variance is constant through time. The approach adopted in this paper concentrates primarily on the autoregressive properties of traffic variability, with the aim to provide better confidence intervals for traffic flow forecasts. The model-building procedure is illustrated with 7.5-min average traffic flow data for a set of 11 loop detectors located at major arterials that direct to the center of the city of Athens, Greece. A sensitivity analysis for coefficient estimates is undertaken with respect to both time and space.