Supply curves are essential for demand prediction in aggregate networks, an
d need to relate cost of use to flow demanded rather than to flow performed
. The paper defines the parameters of relevance and the relationships requi
red. A microsimulation model is used to generate such relationships for hyp
othetical networks. Network performance curves can exhibit backward-bending
relationships between speed and flow performed, but these curves are inapp
ropriatefor demand prediction. Relationships between speed and flow demande
d are monotonic, but not necessarily linear. They can exhibit spatial and t
emporal dependencies, both of which requires them to be reestimated if dema
nd patterns change.