It is no secret that the BATS:VXX is not an instrument to buy and hold. At a staggering effective annual return of -67% since its inception, holding the VXX would have resulted in a total capital loss in excess of 99%.
Conversely, any long-term based short position would have reaped astronomical gains, and it still will.
Here is why:
Contrary to popular opinion, the terrible performance of the VXX is not due to the "rolling decay" associated with the phenomenon of positively sloping term structures frequently called contango. Instead it is because the VIX term structure has been (and still is) in a perpetual state of true contango since 2009.
"True contango" means that each individual VIX future is overpriced compared to its "fair price" - the mathematical expectation of the future's settlement value. There is no exact science to precisely calculate this theoretical "fair value", but it can be reasonably estimated with a model using empirical analysis of historic VIX and VRO values. (The VRO is the VIX derivatives settlement value which reflects a snapshot of the VIX every third Wednesday of the month at approximately 10am.)
To be clear, "true contango" has nothing to do with the common usage of contango in meaning that the term structure has a positive slope. The "rolling decay" associated with this usually positive slope has only occurred because the VIX has behaved differently than the market expected. If the term structure were fairly priced, it would not matter if the second month's future costs more or less than the first month's. Otherwise there would be a statistical arbitrage opportunity, thus contradicting the idea of "fair price".
Finally this mispricing in the volatility term structure is the result of the VIX not mean-reverting as it is commonly thought to. Few center-reverting stochastic processes are truly mean-reverting (with Ornstein-Uhlenbeck being an example of one that is), but I suppose this is just mathematical semantics.
Volatility still clearly reverts back to some center but this center is not constant. I submit that it decreased significantly after 2008 with the introduction of so much regulatory legislation. The commonly held "mean-reverting" philosophy failed (and still fails) to account for this sharp decrease and has resulted in materially flawed models and true contango in the VIX term structure.
So in a manner which feels almost tautological, true contango is the underlying cause of the terrible performance of the VXX and it also explains why the VXX has a net negative return even in times when the VIX term structure has been sloped negatively. Therefore until VIX futures become substantially cheaper, it seems to make sense to bet against the VXX on a long term basis.
To be clear, I say a "long-term" basis because true contango only results in a kind of statistical arbitrage; obviously not pure arbitrage.
The SVXY and XIV are instruments which are designed to perform opposite to the VXX (on a daily basis). In order to capitalize from this statistical advantage, it may be easiest to own one of these outright although both are subject to the risk of a "termination event". This is the possibility that the VXX will increase over 100% in a day thus resulting in a total loss.
Although a termination event is quite unlikely it is possible, and it would end the potential to continue holding the position to recovery. Since the key to profiting from statistical arbitrage is a long-term position, I suggest a "sell and hang tight" short-sale of the VXX. This can be hedged with long call options whenever it becomes too hard to sleep at night.
Disclosure: I hold a number of positions and options on the VXX, TVIX and SVXY which profit when VIX futures decrease in value.