There is a article worth reading at the MarketSci Blog about how the VIX may not be as magical as some make it out to be. The blog post is responding to another on the subject written by Mark Hulbert, whose opinion I agree with on this subject. BTW, the article is also discussed at the Daily Options Report, and there is a good exchange of comments with Bill Luby at the VIX and More site (also worth the read, as usual). The article does a pretty good job illustrating how the VIX is like any other indicator - it is just another tool in the toolbox, albeit useful and better than some.
While the VIX is based on a somewhat complex algorithm of implied volatility, it is not a perfect model in the strict sense of the term for how it is being used, in that it has not been fine tuned with the optimal parameters or the best inputs in order to measure something other than, as discussed by Hulbert, "... the volatility that options traders are expecting the stock market to experience over the subsequent 30 days." The fact that we use the market sentiment implications of the VIX for timing market turning points, or by assigning relatively arbitrary benchmarks (like the market will rally when the VIX is above 30), makes it no more or no less a better indicator in my opinion, or more importantly, a specific model for the purpose we are using it for. Assigning a heuristic is helpful in that it starts to get us interested, but it does not make it any more predictive.
Like many indicators, the VIX may be just another way to measure an oversold market. Helpful, yes, and something I look at, but it should not come as any surprise that it is not anymore accurate than 3/4th of the time, and even then, timing is not exact. In fact, if it was nearly 100% accurate, with good timing, then I would not be writing this article, and you would not be reading it. Instead, a few lucky traders who spotted it first would be sitting on a private island somewhere, while a second group sometime later would have made some money and garnered some fame writing about its use (with some nice historical relationships provided as support).
Meanwhile, the indicator's predictive ability would have probably self-destructed, causing the rest of us to end up disappointed as its widespread use has now caused the indicator to simply tell us what we already know ........ while nicely giving us approximately 3/4th direction accuracy with less than perfect market timing.