Condor Options

This strategy buys the market when participants are overly fearful, and shorts the market when participants are overly complacent. In this study, we use the ratio of one-month to three-month volatility expectations to determine fearful and complacent market conditions.

Bill Luby over at VIXandMore was the first to the notice and make use of the VIX:VXV ratio, which has now become a fairly popular indicator.  We haven’t seen much thorough testing of the ratio, and wondered how well it would perform as a trading strategy.

Just to review, the VIX is an estimate of volatility over the next 30 days as implied by prices in the S&P 500 options.  VXV, which was introduced in late 2007, also tracks the volatility implied by SPX options prices, but over a 93-day time frame.  The key feature of the relationship between these two indexes is that the shorter-term VIX tends to move more quickly and dramatically than VXV, and the ongoing hypothesis is that when the ratio moves to extreme levels, traders can take positions expecting that ratio to revert to a mean.  Since SPX tends to be negatively correlated to VIX, we want to know whether very high (low) VIX:VXV ratios are reliable signals to buy (sell) SPX.  But more importantly:

Are short-term extremes in horizontal implied volatility skew predictive of short-term mean reversion in the underlying?

Parameters

We defined a VIX:VXV ratio of 1.02 as a buy signal, and 0.94 as a sell signal.  The strategy trades 100 SPX shares on the close when the ratio crosses above/below a signal parameter, and exits when the ratio crosses back over/under the signal line.

For example, on March 10, 2008 the VIX:VXV ratio crossed above 1.02 and closed at 1.031, so the strategy bought SPX at 1273.  The ratio closed below that threshold on the next day, signaling an exit of the long SPX trade at 1320.

Results

Results are quite strong, with a 76% win ratio and a profit factor of 4.72.  Over the past 10 months, the strategy generated 17 signals, no consecutive losses, and was in the market less than a third of the time.  Many trades were close-to-close affairs, which supports the intuition that extreme levels of fear or complancy expressed by a stretched VIX:VXV ratio may be predictive at least of a short-term reversal towards the mean.  Results are robust across a range of parameters, but are restricted to underlyings that track the S&P 500.

We chose parameters above that would signal fairly often and generate very short-term trades.  But the VIX:VXV ratio might also prove useful on a longer term basis, so we ran a second test with fewer trades and bigger swings in mind.  Doing a bit of intentional curve-fitting, we selected the 5 most obvious extremes over the period 11/2007 – 08/2008 (we ignored data from 09/2008 forward for reasons we’ll come to shortly) and tested with parameters of 1.08 for long SPX entries and 0.87 for short entries.  Each trade was successful, with an average hold time of 3 days and profits of 156 SPX points.  Using a time-based exit of 15-30 days would have been even more successful.

Finally, for those who don’t mind some slightly absurdist backtesting, the following parameters proved particularly profitable: buy the market when the ratio crosses above 1.08, and exit at ratio=1.02; sell the market when the ratio crosses below 0.95, and exit at ratio=0.92, or after 9 days, whichever is sooner.  These settings generated 10 trades this year, with a profit factor above 18 and win ratio of 90%.  More details available here.

We should also mention – and this applies to all of the strategies and systems that we publish – that the best way to exploit the edge described here isn’t necessarily just to take long and short positions in the underlying.  Especially when it comes to trading volatility, the best implementation will often be a risk-defined options strategy.  One reason is that if you want to express a view on changes in volatility, it’s actually quite helpful if you can minimize the impact of price movement – so while the backtesting above does quite well with a long/short index implementation, you could expect returns to be significantly better using, for instance, iron condors or calendar spreads where appropriate.

Caveats

The biggest reason for caution about this strategy is that, at the time of writing, we have only about one year of VXV data, and one year doesn’t prove anything.  VXV was introduced in early November 2007; however, the CBOE appears to have calculated VXV closing prices back to January 2002.  We have asked for that data but no one from CBOE has replied yet – if anyone knows how to access that data, we would be grateful.

The other problem with this strategy is the existence of September-October 2008.  The 2008 selloff tells us one general thing and one specific thing.  In general, every strategy seeking to exploit small edges or stretched indicators was soundly defeated during that period, for the simple reason that strategies assuming normal market conditions cannot thrive in an abnormal market.  Specifically, we saw that while in many cases the implied volatility in front month options tends to overestimate short-term realized volatility, and is hence “wrong,” when those front month estimates are “right,” they’re really right.  Put differently, the VIX:VXV ratio appears useful as a sensitive oscillator within the confines of some easily discernible volatility channel, but when that channel no longer holds, it makes sense to ignore the ratio until some new channel is in place.  It will be interesting to revisit this ratio in another year’s time, as three outcomes seem possible: a) the elevated volatility enivronment does not persist, and we return to a VIX range of 15-30; b) the elevated volatility environment does persist, and as the volatility term structure flattens, we arrive at a new and tradable volatility channel, using more or less the same VIX:VXV ratios as before; c) the elevated volatility environment does persist, and sustained event risk causes a persistent and dramatic horizontal skew such that VIX:VXV ratios must be adjusted higher.

For strategies seeking to exploit mean reversion tendencies, it’s clearly possible for an underlying to move in such a way that waiting for mean reversion means flirting with insolvency.  That’s why no strategy, no matter how historically successful, can remove the need for sound risk management.  More work would clearly be needed to turn the VIX:VXV ratio into a standalone viable trading strategy, but a good preliminary step would be to add a generous stop loss – say, 5% or so – and after taking such a loss stay out of the market until the VIX:VXV ratio had recovered to within the normal range, or until a new normal range could be discerned.

Conclusion

Preliminary testing suggests that short-term extremes in horizontal volatility skew are predictive of short-term mean reversion in the underlying.