I'm looking at my system's (backtest) PNL from that period and VIX futures lost a mere 1.8%, which was recovered over the next 8 days. Being one of many instruments, the effect is limited. UPDATE: If I traded only VIX and at the same vol, my system would have busted. But since VIX at the time was getting just about 1% of my risk, it had a negligible effect.
That also depends on what your signal for VIX future was. I'm talking about worst-case scenario where you have strong short signal and we have a subsequent spike. Anyway, I tried to crunch some numbers out of curiosity and calculated for each instrument I trade the average of the worst 5% rolling ratios between 5-day future absolute log returns and starting blended volatilities. Against my expectations, the value for front VIX future (and back contracts too) is not so far from SPX future (0.42 vs 0.35) and we also have several front commodities contracts and some FX/bonds (USCNH, BTPs) with similar values. The worst 5-day spike remains CHF in 2015.
Most of the term structure in vol is expectation of mean reversion. That's why when index is low, the term structure is very steep and when index is high, the term structure is very inverted. That's also why when index spikes, the back end reacts much less (it's more complicated IRL because VIX futures are priced from the SPX surface, but that's not relevant here). But yes, by going short you are harvesting some risk premium too, though it's probably better classified as carry, not trend.
I agree with the risks, and while it's nice to treat every market the same for simplicity, I do set a reduced risk allocation to VIX. There's a number of contracts I'm comfortable with, and I won't go over that.
I have another system that trades this, using measures of the VRP and Contango % as inputs. It has great $ returns in % of what's allocated, but drawdowns happen while waiting for mean reversion and it has to exit if things get bad. There's always a bigger volmageddon in the future, we didn't have VX futures in 1987 to test gap moves. Since trend seems to work OK with VIX integrated, I may just kill that system off and just rely on trend. I'll handle tail events better this way, and can focus more allocation to trend.
I do the same, but only on the short side. We are usually long only after a spike, where trend and carry signals turn positive: in those conditions both the contract nominal value and its volatility are high, so positions are small anyway.
Same here. I don't trade VIX but do trade V2TX. I started off with equal allocation as the other instruments, but have reduced its weight over time. Not just because of volatility, but also because of the high margin requirements compared to other instruments.
Very true. Although this phenomenon exists across most commodity markets. For example when spot oil is high you have backwardation (and vice versa). The expected mean reversion is what creates carry opportunities. I need to check this but I prefer to blend carry with trend on vol like Rob because i assume it reduces some of that nasty -ve skew.
This - diversification is the best risk management tool. Less polite response: I hope to god this isn't a discretionary limit, or you can get off my f****** thread Polite response: You might want to look at a more systematic way of doing this as I discuss in the blog and AFTS; work out an objective way of limiting the number of VIX contracts and then applying it to *all* instruments. Using carry as well as trend obviously incorporates that term structure information. And all trend systems that use total return (=futures backadjusted price) are effectively a mixture of carry and spot trend. (again discussed in AFTS, I seem to be referencing that a lot, definitely not plugging but it is remarkable how much information is contained in that slim volume on futures trading...) Rob
Lol Not discretionary, though maybe my system design had some discretion in it. I size based on ATR, it's not pysystemtrade but a more classic trend style system. Since I don't have correlation management and dynamic position sizing, I have markets in buckets based on static correlations from a recent period. Each bucket uses a different % risk for sizing, basically if I have 45 indices but only 2 coffees, the indices get sized alot smaller than the coffees but I could have more positions on. The number of contracts is systematic based on those bucket parameters, but I just assigned the VIX to the higher margin & higher risk bucket so that the backtest shows the kind of trades that will not make me pee my pants.