Joe,
Terminating a GRE tunnel on loopback interfaces is usually preferred, as it gives the tunnel stable endpoints (almost) independent of the physical connectivity of the terminating routers. If the loopback is advertised in a routing protocol and there is at least one path towards the router, the routing protocol will find it and make sure that the loopback address is reachable.
Often, however, it might be precisely the problem whether an address on a loopback can be advertised in a routing protocol. If there is no routing protocol between the hub and spokes which is often the case, then it is hardly possible to use loopbacks for tunnel termination. It really depends very much what the internetwork between the hub and spokes can do for you. Very often, it is just because it can't do much that you decide to deploy tunnels.
Best regards,
Peter