I think that back when we discussed longcoins and shortcoins ideas it turned out that it is possible to not need a data feed.
Basically at any moment when either party saw a price somewhere that made them prefer that price, they could escape the longcoin/shortcoin position to cash in wherever it was that they thought would for them be a better deal.
It thus needed no oracles aka data feeds, since it is no one else's business but your own why it is that you exercised, or even whether you were in fact going to actually be able to get the better deal that had lured you into exercising
You absolutely do need a price feed, centralized or decentralized somehow (eg. proof of stake voting), in order to do any of this. The problem is that the network has no way of "knowing" about the real world; there is no term in cryptographic functions to represent physical US dollars. Thus, we have the following problem. Suppose that you had some kind of instrument that reflected the price of USD without relying on any kind of price feed. Now, suppose that due to a market fluctuation the price of this instrument drops to 0.99 USD. Presumably, people would buy up this instrument seeing it as undervalued, and thus raise the price back up to 1 USD - that's the self-fulfilling prophecy for which the designers hope. However, the problem with self-fulfilling prophecies is that they go both ways. What if the price of this instrument drops to 0.96 USD, and some investors get a different idea: that the price drop is a sign that the link between the instrument and the USD is broken, and it will only drop further? Then, they sell, and the price of the instrument might drop to 0.93 USD, then 0.9 USD, then more and more people will realize this fact, and the price of the instrument will drop further. Unlike Bitcoin, however, it will likely never recover. Bitcoin can restabilize at a lower price target, and many times did. The USD instrument cannot stabilize at anything but 1 USD, so if it's not at 1 USD it goes to zero.
So these "magic"-based price targeting mechanisms will undoubtedly be stable for a while, but the first black swan event will wipe half of them out in an instant. This is why you need external, hard data on prices.
LIBOR-style manipulation
If people think this is a concern I could write a spec for proof of stake voting on price feeds to require larger collusions to significantly manipulate prices.
Also I do not believe what is described in the OP meets the formal definition of a CFD - at least not the one used in the finance industry. I only interned at an ibank and decided it wasn't for me so I am certainly no expert.
How is it not a CFD? It seems to follow the examples
here pretty closely. In any case, it may be different in some technicality, but precise definitions aren't important; what is important is what it lets you easily have an arbitrary position in a market with an arbitrary margin if you can find a counterparty.
BTW you could even have meta features based on it: I could basically entrust a certain amount of mastercoins to "any CFD that meets certain criterias". This way, even if the duration of the contract runs out, and my counterparty can withdraw their funds ... my CFD will automatically pop up to find the next counterparty, and I won't have to manually find another one.
This is interesting; we could definitely do it.