It's essentially a moving average of historical volatility...an MA with a PhD. Once you figure out which one to use: Magic eight ball says go with Exponential GARCH, you will have to design the even more hairy model that will decide to trade something.
I don't think there are any. That's why you have a whole family of them. So much research has been poured into that. Even more so in the study of forward curves. The best ones that are publicly known probably come from the central banks.
Predicting realized volatility is easier than predicting implied volatility. Unfortunately, we often need the latter. Ernie Chan has a good discussion of that and related topics here: Slides:
Thanks for the post lesserfool. I should have guessed that Ernie would have something on this topic. Great stuff but as you pointed out, not a complete solution out of the box.
Forecasting / trading volatility. Or implied volatility to be specific, is like trying to play archery in a hurricane.
In the index world, it's pretty straight forward. Implied volatility is the best way to forecast of the realized volatility. There are plenty of sophisticated methods (weighted GARCH of various forms), but in general, event risk and seasonality will overpower whatever minor improvements you'd get from it. If you game is to sell risk premium, you will gain nothing from forecasting vol (as it's the one time it's wrong that will kill you). You would, however, gain a fair bit from understanding how rich or cheap vol is. It will help you make trading decisions, size your risk and manage your deltas. If you are trading spreads (e.g. calendars), you can thing of various forms of "uncertain vol model" to gauge relative value as well as manage your deltas. If you are trading IMPLIED vol, then it's a different story - you wanna have an idea of how implied vols (fixed strikes or full surface) move due to shocks, directional moves etc. Anyway, most academic literature will be worthless or near-worthless and you want to do your own legworks with regards to analysis.
There's a lot of mathematical masturbation behind volatility forecasting but in essence it reduces to one question: How much margin should I add to my purchase in order to sell it for a profit? That's when both the purchase and the sell are probabilistic.