To put some flesh on the bones of this idea, imagine a hypothetical security named BLK. Demand for bigger blocks should buy up BLK and demand for smaller blocks should short BLK. The price of BLK in BTC is the ideal block size.
Now imagine that there are futures contracts for the security BLK. On the settlement date of those futures the current BLK/BTC price of those futures is taken to be the new Bitcoin block size for the next 3 months.
For instance, if I predict or want the block size to be higher on September than it currently is, I would buy up September BLK futures. My actions would nudge the price up, and if come September I am right I get what I want and have a floating profit on the futures market.
The nice thing about a futures market is that it allows capacity planning for the months ahead. Also there is no need for an underlying BLK security for a futures market in BLKs to exist.
If the market is efficient and correctly sets the block size, BTC/USD will rise and the BTC profits of the market participants will go up in USD terms as a result.

On 5 August 2015 at 12:35, Hector Chu <hectorchu@gmail.com> wrote:
On 5 August 2015 at 12:07, Adam Back <adam@cypherspace.org> wrote:
This prediction market in block-size seems like something extremely
complex to operate and keep secure in a decentralised fashion.

Why would it need to be decentralised? Bitcoin.org could run the exchange, and the profits from the exchange could be used to fund Core development.

We also have no particular reason to suppose other than
meta-incentive, that it should result in a secure parameter set.

Security is a continuous variable, trading off against others. If security gradually begins to be threatened as a result of block size gradually increasing, the concerns of users will be enough that the bears will gain control over the bulls on the block size market.

I suspect that, while it is interesting in the abstract, it risks
converting a complex security problem into an even more complex one,
rather than constituting an incremental security improvement which is
more the context of day to day discussions here.

Hard problems call for complex solutions.