Knowledge of future traffic conditions is important for travellers - to improve their journey planning (with or without congestion) - as well as for traffic managers - to improve the effectiveness of their control actions, and their planning. In recent years, a lot of effort has been put into investigation towards models that can predict, online, the short-term development of traffic conditions. The results obtained to date have been somewhat disappointing, especially considering the effort and the complexity of the model systems. We have not come across convincing evidence that even advanced models have been capable of consistently outperforming much simpler methods. This paper reconsiders the use of a simple historic database to provide short-term forecasts. Results show that only slight improvements can be achieved by using online measurements. The 'historic forecasts' are however very useful, especially when information can be provided about the certainty of the forecasts.
Abstract