Shorting VXX riskier than appears

VXX has been going down since the day it listed, losing on average 8% per month (using data since inception until friday close), so it's not really a very original idea to short it. I did not short the ETF outright, but put on a few small bearish option positions. However the other day I got around to do some digging and found out a few things that I did not know.

EDIT: earlier I wrote that VXX lost about 10% per month. That figure was based on calculations I did few months ago - in the first year of trading VXX declined from 104.58 to 31.64 at a rate of 9.9628%, which I rounded to 10%.

Why did VXX decline so much? As Bill Luby and many others explain there are two reasons: the term structure of VIX futures, and trading costs. Most of the time VIX chain is sloped upwards (contango), that is front month futures are cheaper than the second month, making the roll costly. I collected data from the beginning of VIX futures trading until the end of last year. While I did not simulate the exact rolling algorithm used by VXX, I used rolling just once a month, on the expiration day. What I found is that the average difference between the front and second month futures to be 0.8 VIX points (t=2.47), or 6.7% of VIX value (t=6.03), and second month future is more expensive than the front in 82% of cases (53 out of 64). These costs do not include the bid-ask spread; they are statistically and economically significant.

However this is not a complete picture. Since VXX has about year and half worth of data, I needed to look for more. VXX is based on SPVIXSTR - S&P 500 VIX Short-Term Futures™ Index (overview here). I was able to find historical data for the index, and reproduced the chart below. It is clear from the chart that while the index has been in a steady decline since volatility peak in late 2008 (VXX was launched shortly thereafter in January 2009) going short the index would not have been without significant risk.

In four and a half years (54 months) since the index inception it declined on average by 1.65% per month, much less than 8.3% for VXX. While 1.65% is still significant, it pales in comparison to the maximum drawup (maximum drawdown for short-seller) of 523% from the low of 37,098 in Feb 2007 to the high of 231,276 in Nov 2008. For the life of the index the average drawup is 94%! I used simple returns for drawup calculations.

Now to put things into perspective, during the last 17 months for which VXX existed, VIX declined from 44.84 to 24.98, at a rate of 3.4% per month vs, 8.3% for VXX, and 8.0% for the SPVIXSTR. In the past 54 months VIX rose from 11.19 to 24.98, at a rate of 1.5% per month, while SPVIXSTR declined at 1.6% per month.

To measure the risks I conducted two studies - Monte-Carlo on index returns, and a parametric model like the ones I described last month. I used the two methods to generate forecasts for VXX and VXX options, and while I won't share the details, I realized that what I saw as a relatively low-risk opportunity in VXX, was rather moderate risk, low-edge trade.