I propose the following rule of thumb for VIX interpretation:
If you think some VIX movement entails a proposition p and movement in the other volatility indexes VXN, RVX, and VXD doesn’t entail p, you shouldn’t believe p.
Why accept this rule? Because equity indexes are highly correlated, especially over the very short term, and volatility indexes are calculated using the same methodology, such that in the case of a divergence of one volatility index from the others, the majority rule. Here’s a stab at some empirical evidence. I applied a less optimized variation of the mean reversion strategy first profiled here in September 2008 on each of the following pairs: SPX/VIX, RUT/RVX, NDX/VXN, and DJIA/VXD. Equity curves are provided below in that order.
The y axis is logarithmically scaled and is deliberately obscured because the performance of the strategy isn’t really the point here. The point is that the four volatility indexes perform similarly relative to their underlying indexes as far as this strategy is concerned. That’s not a surprising result, given that the calculations used to derive each volatility index are the same.
Put another way: imagine that Standard and Poor’s disappears from the Earth tomorrow, and that spontaneous mass amnesia causes us all to forget which companies belonged to the S&P 500. There is no more SPX, and hence no more VIX. If VIX is special, then we should all be concerned, for we will be data-poorer than we were; traders who use VIX data in any strategy should panic. But since equity indexes are highly correlated, and since the methodology used to calculate VIX is also applied to the other indexes, volatility indexes are fungible.
So contrarian traders who might otherwise be inclined to divine significance from a strange result in VIX vs. SPX should wait for confirmation of that relationship from the other indexes. A worthwhile idea for future research might be to compare returns after these sorts of divergences among the volatility indexes, but I expect the data set would be rather small.