Temporal aggregation in traffic data: implications for statistical characteristics and model choice

Abstract

Time series techniques are useful for analyzing transportation data, uncovering past trends and providing projections. Such analyses are sensitive to the temporal aggregation of the data, an issue that has been widely ignored in the transportation literature. In traffic engineering, aggregation usually equals to the average of a variable across large regular time intervals, such as 15 minutes, hours, days, months and so on. We investigate the effects of temporal aggregation on time series of traffic volume and occupancy in urban signalized arterials. Results indicate that aggregation eliminates long memory characteristics and variance heterogeneity; this leads to smoothing traffic variation and creating a time series structure that has reduced sensitivity to changes in traffic. Moreover, aggregation was found to directly affect volatility as captured by the parameters of the Generalized Auto-Regressive Conditionally Heteroskedastic (GARCH) models.

Keywords:

• data aggregation
• traffic flow
• stationarity
• long memory
• volatility