Conversation

3) Each approach has its own issues. Right now, frankly, something like 'median of CEX orderbooks' is probably the best oracle. But there are issues with it, which points out: a) sometimes CEX APIs crap out b) this is only good assuming most liquidity is on CEXes
2
8
4) c) you have to trust the reporting agents a & b don't apply to DEXes, they get those for free! But (b) does, and right now it's an issue; most liquidity is not on DEXes. In addition, ETH-based DEXes are slow, which means their prices are stale.
1
5
5) Hedging on them can also have issues. If 50% of orders whiff because of race conditions and blockchain capacity issues, you can't trust that the current DEX marketdata is really tradable exactly. Also, for better or for worse, CEXes have more circuit breakers than DEXes.
1
6
6) This can be good (stops crazy prints!) or bad (stops real price movement!) You also have to worry about withdrawal blocked CEXes.
1
5
7) In the end it's tough and there are pros and cons. Right now I think 'median of CEXes' is about as good as you can get, though you prob want to add 'and bound within 5% of a 5-min EWMA' or something like that.
1
7
8) One issue, though: in the end, if the underlying asset isn't liquid enough, no oracles can save you. DAI just isn't very liquid. Not on CEXes, not on DEXes, not anywhere. And partially it could use more market makers. But partially it's way worse than that.
1
7
11) TL;DR: a) LINK is pretty good as far as DeFi oracles go right now b) there are pros and cons to each, and maybe over time direct on-chain price feeds will get more competitive c) whatever you do, DAI isn't very liquid and can blow out
1
11
12) And above all: Oracles aren't magic. They're only as good as their price feeds. Oracle systems solve the problem of "ok here's how I want to price this asset, now get it on chain in a reasonably robust way". They can't get prices when no good prices exist.
3
25