Bloomberg's Odd Lots podcast from a few days ago discusses "How one of the Most Profitable Trades of the Last Few Years Blew Up in a Single Day." Specifically, how did short volatility trades perform so well for so long, and then unravel so dramatically in a short period of time?
In fact, these two things are directly related. This trade performed well for a long time precisely because it was effectively selling insurance against an infrequent, severe event - in this case, a volatility spike. In essence, those who shorted volatility (primarily by selling VIX futures either directly or indirectly through exchange-traded products like the XIV note) were providing insurance against a volatility spike, collected premiums for a long time, and then ended up paying out large amounts when a spike actually occurred. It is analogous to a company insuring against earthquakes: it's rolling in the dough collecting premiums until a big earthquake actually happens, at which time the company has to pay out big time.
If you look at a graph of the VIX, you'll see that the VIX can be well-described as a mean-reverting process (i.e., it doesn't behave like a random walk or a geometric random walk, but tends to return to a base level after it diverges from that level) subject to large upward shocks. After the spikes, mean reversion kicks in, and the index returns roughly to its previous level.
So if you are short the VIX, you pay out during those spikes.
And that's not all. The VIX is strongly negatively correlated with the overall market. That is, it tends to increase when the market goes down:
This means that providing insurance against volatility spikes is costly: the volatility short-seller commits to making payouts in bad states of the world. Thus, risk-averse suppliers of volatility insurance will demand a premium to bear the risk inherent in that position. Put crudely, a short VIX position has a large positive beta, meaning that the expected return (risk premium) on this position will be positive, and large.
The flip side of this is that those with a natural short volatility exposure incur a large cost to bear this risk and might be willing to hedge (insure) against it. Indeed, given the fact that such natural short exposures incur losses in bad states of the world, those facing them are willing to pay a premium to hedge them.
In equilibrium, this means that short volatility positions will earn a risk premium. Since short-sellers of volatility futures will have to earn a return to compensate them for the associated risks, the VIX futures price will exceed the expected future value of VIX at futures expiration. Thus, VIX futures will be in a Keynesian contango (with the futures above the expected future spot). Given that VIX itself is a non-traded risk (one cannot buy or sell the actual VIX in the same way one can buy or sell a stock index), this means that the forward curve will also be in contango.* Further, one would expect that long VIX futures positions lose money on average, and given the spikiness of realized VIX, lose money most of the time, with the gains occurring infrequently and being relatively large when they do occur.
And of course, short positions have the exact opposite performance. Shorts sell VIX futures at a premium over the price at which they expect to cover, and hence make money on average. Furthermore, losses tend to be relatively infrequent, but when they occur they tend to be large.
And that's exactly what happened in the period leading up to February 5. During most of that period, VIX shorts were making money. When the spike occurred on 2/5/18, however, they were hammered.
But this was not an indication of a badly performing market, or irrational trading. Given the behavior of volatility and the existence of individuals and firms with a natural short volatility position that some wanted to hedge, this is exactly what you'd expect. Participants (mainly institutional investors, including university endowments) were willing to take the opposite side of those hedges and receive a risk premium in return. Those short positions would earn positive returns most of the time, but when the returns go negative, they tend to do so in a big way. Again, just like earthquake insurance.
One of the inventors of VIX claims that he doesn't understand why products such as VIX futures or ETPs that have long or short volatility exposures exist. Really? They exist because they facilitate the transfer of risk from those who bear it at a higher cost to those who bear it at a lower cost. Absent these markets, the short volatility exposures wouldn't go away: those with such natural exposures would continue to bear it, and would periodically incur large losses. Those losses would not be as obvious as when volatility products are traded, but they would actually be more costly. The pain that volatility short-sellers incurred earlier this month might be bad, but it was less than the pain that would have existed if they weren't there to absorb that risk.
One interesting question is whether technical factors actually exacerbated the size of the volatility spike. Some sellers of volatility short ETPs (like the XIV exchange-traded note that is basically a short position on the front two-month VIX futures) hedge that exposure by going short VIX futures. To the extent that the delta of the ETPs remains constant (i.e., the sensitivity of the value of the product to changes in forward volatility remains constant), that's not an issue: the hedge positions are static. However, the XIV, in particular, had a knock-out feature: payment of the note is accelerated when the value of the position falls to 20 percent of face amount. The XIV experienced such an acceleration event on the 5th, and to the extent the issuer (UBS) had hedged its volatility exposure, this could have caused it to buy a large number of futures - because as soon as the note was paid off, the short VIX position was unnecessary as a hedge, and UBS would have bought futures to close that hedge. This would have been a discontinuous move in its position, moreover - oh, the joys of hedging barrier options (which is essentially what the acceleration feature created). This buying into a spike could have exacerbated the spike. Whether UBS actually did this or whether liquidating its hedge position was big enough to have an appreciable knock-on effect on prices is not known. But it could have made the volatility event more severe than it would have been otherwise.
These markets exist for a reason - to transfer risk. Moreover, they behaved exactly as expected, and those who participated got - and paid - in the expected way. Insurance sellers (those short volatility futures) collected premiums to compensate for the risk incurred. Most of the time, the risk was not realized because of its "spikey" nature, and those sellers realized positive returns. When the spike happened, they paid out. There is never a free lunch. Yes, the insurance sellers dined out on somebody else most of the time, but when they had to pick up the tab, it was a big one.
*Keynes caused untold confusion by using "normal backwardation" to describe a situation where the futures price is below the expected spot price. In market parlance, backwardation occurs when the futures price is below the actual spot price. Keynesian backwardation and contango refer to a risk premium, which is not directly observable in the market, whereas actual contango and backwardation are. It is possible for a market to be in contango but in a Keynesian backwardation. Similarly, it is possible for a market to be in backwardation but in a Keynesian contango. If interest rates exceed dividend yields, stock index futures are an example of the former situation. No arbitrage forces the market into a contango, but long positions earn a risk premium (a normal backwardation).