I was wondering, what if the FRR used a Delta (the math definition, not the Bitfinex)? Bitfinex could take the last 15 min Bitcoin price range and matched it to the price change and volume of the historical Bitcoin price data. The time frame would then be cross referenced to the historical swap data. The historical rate change would then be used as the basis for modifying, not determining the FRR rate. If the Bitcoin price is changing, the swap rate would change in a historically predictable way. If done properly, the rate change could only be gamed by the traders and only to the existent that it agrees with history.
The FRR would rise and fall much more quickly than now but it would stabilize much faster. It would be a trailing indicator but also be based on past market performance. I think everyone knows the perils of using the past to predict the future but this method would isolate itself slightly since it is using the near instantaneous (15 minute) price change to determine the corresponding historical swap rate change from a point in time disregarding the surrounding historical market tread. It couldn't be worse than the current FRR calculation which is based on the most simplistic (and statistically meaningless ) 'average' available.
USD swap demand goes up if the price goes down ("BTC are getting cheaper, let's buy more, TO DA MOON!") and it goes up if the price goes up ("It's_happening.gif"), it only seems to go down if the price doesn't change a lot over several days. Basing a rate on the price delta of the last 15 minutes is not going to model demand easily I'm afraid.
I'm still more in favour of keeping FRR as is and rather changing defaults or introducing more strict rules (e.g. FRR can only be set for 3 day durations max. so fixed rate swaps have more space to discover prices?). Also it might be interesting to know how FRR would look like if 30 day FRR is based on 30 day fixed rate swaps, 10 day FRR on 10 day fixed rate etc.