There are trading systems which switch between owning XIV (which is “short” “volatility”) and VXX (which is long volatility). Some of these trading systems have backtested results going back to 2004 of ~100%/year! VXX has lost money over every twelve month period. Lots of money. What it has going for it is that on rare occasions it spikes up and that spike is negatively correlated with the market. XIV on the other hand is roughly the inverse of VXX. It makes lots of money almost all the time except for rare occasions. If the VIX spikes suddenly by 80% or more then XIV can go to zero. As far as I know this has only happened once in the 25 or so years since volatility has started to be measured in the mid 1980’s. There is also a risk that Barclays can go under but I consider that relatively less likely as the ETN is a senior debt security (meaning that the ETN holders will be first in line for any assets remaining after bankruptcy) and Barclays may get bailed out in any case.
What is the optimal position size? Position sizing is a very important part of investing that doesn’t get enough attention. Having a position that’s too large will cause higher volatility than necessary, lower returns and eventually may even lead to gamblers ruin. Having a position that’s too small means leaving money on the table. (Most sane people would probably rather the risk of making too little over the risk of losing it all.) I am not confident enough in these systems that they can predict with absolute certainty when a spike in the VIX will happen that would wipe out XIV.
One strategy to estimate optimal position sizing is to optimize the ratio of stocks/bonds/volatility/cash for the optimal Sharpe ratio using historical backtesting data. Some calculations have pegged the optimal allocation to such a volatility strategy using this system in a bond portfolio to be 10% and in a stock portfolio up to 50%. I am not sure sure if this strategy is sound because the backtests only go back until 2004 and the tail risk (surprising enough) has not come up in the backtest period.
Another strategy to estimate optimal position sizing is with the Kelly criterion. See [url=http://www.quantwolf.com/calculators/stockbondcalc.html]http://www.quantwolf.com/calculators/stockbondcalc.html[/url].
Of course, the accuracy of these results depends on estimating the odds correctly and the payoffs correctly. But since we know that if you invest half the Kelly amount, you get about three-quarters of the return with half the volatility, and since it is better to take the risk of leaving some money on the table rather than the risk of gamblers ruin, we use conservative estimates.
What numbers should we plug into the Kelly calculator?
Because this calculator assumes that there are only two possible returns for the stock (or in our case ETN) in a given time period I chose to divide the periods into the average backtested year (2004-2013 with 100% upside) and 100% downside for the possibility of a VIX spike. (It is possible that the with proper position sizing the downside would actually be less despite the spike because of the fact that if you would have the guts to stick with the strategy it would probably recover some of it’s losses assuming that you don’t get wiped out)
I also punched in 0.96 as the success rate because 24/25 years = 0.96. I also chose 0.025 as the risk free return as that is basically what you can get these days risk free. The result: 92% allocation to the volatility strategy recommended; so a 1/2 Kelly = 46%.
This implementation of the Kelly criterion is limited because in practice you would have some money in stocks and stocks is not risk free, and may in fact be correlated with XIV. (Besides for the other problems of accurately estimating the future).
Is there a better way to go about this?