From Poon and Granger's extensive survey (2003) it seems that implied vola was seen as the most proven, if biased, predictor of future realised volatility back in 2003.Is that still the accepted tenor or did the last 7 years yield models that are now preferred or have shown the predictive power of implied vola to be inferior than originally thought? (I am more interested in daily vola, not necessarily in high-frequency, minute by minute, measurements)I have come across several other studies that deal with "model-independent" implied volas, spiels on GARCH models etc.Do you know of any seminal papers I should read? Thank you very much for your insights!

That's a hard question.nevertheless I found quickly this very recent paper (end 2009) :Forecasting Realized Volatility with Linear and Nonlinear Models But I think that the implied volatility is still the most used one. After that it's always the big debate between implied volatility deterministic function, stochastic volatility model, model with jumps or path dependant GARCH like model. in any case you have the choice to extrapolate the historical volatility (quite dangerous I think) or to foreseen implied volatility with a model. The Relationship Between Implied and Realized Volatility of SP500 Index: Wilmott Magazine Article Anyway it's a hot topic

Check 'Paul' the Octopus