Abstract
The feasibility of using disaggregate models for aggregate prediction is demonstrated. It is shown that acceptably small aggregation errors can be obtained using normally available data, in a computationally inexpensive procedure. In particular, the importance of information on the availability of alternatives is established.