The IP address market is sort of like banking on a planet where there is only a single gold mine containing 4 billion ounces of gold, run by the central bank, and the planet experiences severe deflation as the bank runs out of gold and can't issue any more coins. I suppose one could try to analogize NAT (using private addresses on small networks like the one in your house) to fractional banking. We'd apply Gresham's law to private RFC 1918 addresses ("bad addresses drive out good"), but that'd be a stretch.

What the bank analogy doesn't capture is the routing costs. The reason that pure IP address markets won't work is the routing externalities. Every time someone breaks a block of IP addresses into multiple pieces, that imposes costs on all the operators of the large routers that need to keep a route to every chunk of active IP addresses. But there is no workable way to impose per-route charges on end networks, nor to arrange to pay the money to the router operators to subsidize the costs that all the routes impose.

I suppose it might be possible for network peering arrangements to take into account the number of routes used by each peer as well as what they currently do, the amount of traffic in each direction. I'm not aware of anyone doing this, but I don't claim to be totally au courant on peering negotiations.