A methodology is presented for forecasting traffic volatility in urban arterial networks with real-time traffic flow information. This methodology provides a generalization of the standard modeling approach, in which both the mean, modeled by an autoregressive moving average process, and the variance, modeled by an autoregressive conditional heteroscedastic process, are time-varying. The statistical analysis and forecasting performance of the proposed model are investigated with real-time traffic detector data from a real urban arterial network. The results indicate the potential of the proposed model to improve the accuracy of predicted traffic volatility across different lengths of forecasting horizon in comparison with the standard generalized autoregressive conditional heteroscedastic methodology.