On Contango-Based XIV Trading Strategies
Summary In July 2014, Seeking Alpha author Nathan Buehler discussed a strategy where you short VXX when VIX goes from backwardation to contango, and cover when VIX re-enters backwardation. Buying XIV rather than shorting VXX is a very similar idea. The XIV version of Mr. Buehler’s strategy can be viewed as making a 1-day bet on XIV whenever VIX is in contango. VIX contango is a useful predictor of 1-day XIV growth. But historically a contango cut-point around 5% rather than 0% generates better raw and risk-adjusted returns. XIV is extremely risky (beta > 4), but trading strategies based on VIX contango appear promising. Background The VelocityShares Daily Inverse VIX Short-Term ETN (NASDAQ: XIV ) has had tremendous growth since it was introduced in late 2010, but has suffered major losses recently. (click to enlarge) The recent 11.9% dip in the SPDR S&P 500 Trust ETF (NYSEARCA: SPY ) coincided with XIV losses of 55.7%. XIV is still ahead of SPY since inception by a fair amount ($26.2k vs. $18.0k), but the extreme volatility of XIV makes it arguably an inferior investment (Sharpe ratio = 0.040 for XIV, 0.055 for SPY). In my view, XIV is a rather dubious fund to buy and hold long-term. It amplifies returns, but seems to amplify volatility even more, resulting in worse risk-adjusted returns than SPY. But trading XIV based on VIX contango – that is, the percent difference between the first and second month VIX futures prices (available at vixcentral.com ) – appears very promising. The purpose of this article is to assess the predictive value of VIX contango, and to assess and attempt to improve a strategy proposed by Seeking Alpha author Nathan Buehler. Data Source and Methods I obtained daily VIX contango/backwardation data and historical XIV and SPY prices from The Intelligent Investor Blog . Daily contango/backwardation is defined as the percent difference between the first and second month VIX futures. While the Intelligent Investor dataset includes simulated XIV data going back to 2004, for this article I only use the actual daily closing prices for XIV since its inception in Nov. 2010. I used R (“quantmod” and “stocks” packages) to analyze data and generate figures for this article. A Look at Nathan Buehler’s Strategy In the Seeking Alpha article Contango and Backwardation Strategy for VIX ETFs , Mr. Buehler suggests shorting VXX when VIX goes from backwardation to contango, and closing the position when VIX re-enters backwardation. The exact time frame for back-testing is a little unclear to me, but Mr. Buehler reported 221.09% total growth from ten VXX trades between May 21, 2012, and April 14, 2014. That is impressive growth. Then again, VXX fell 86.1% over this time period, and XIV gained 213.9%. So it’s a bit unclear how much of the strong performance was due to VXX tanking over the entire time period, and how much was due to the contango strategy providing good entry and exit points. I am not a short seller so I’m more interested in the “buy XIV” version of Mr. Buehler’s strategy. Let’s consider an approach where you look at VIX contango at the end of each trading day. If VIX has entered contango, you buy XIV; if it has entered backwardation, you sell XIV. If we backtest this strategy since XIV’s inception, ignoring trading costs, we get the following performance: (click to enlarge) The contango-based XIV strategy performs well relative to buying and holding XIV for the entire period, achieving a higher final balance ($57.0k vs. $26.2k), smaller maximum drawdown (56.3% vs. 74.4%), and a better Sharpe ratio (0.061 vs. 0.040). Looking at the graph, we see a major divergence in mid-2011 when selling XIV avoided a huge loss. However, there were many times where the contango strategy failed to prevent big losses. Note that buying XIV when VIX enters contango, and selling when it enters backwardation, is equivalent to holding XIV for 1 day whenever VIX is in contango. So this strategy is entirely dependent on VIX contango predicting 1-day XIV growth. VIX Contango and 1-Day XIV Growth For Mr. Buehler’s strategy to have worked so well over the past 5 years, there must have been positive correlation between VIX contango and subsequent 1-day XIV growth. There was indeed some correlation, but not very much. (click to enlarge) The Pearson correlation was 0.059 (p = 0.04), and the Spearman correlation 0.027 (p = 0.35). Note that VIX contango explained only 0.3% of the variability in subsequent 1-day XIV growth. But there does appear to be some predictive value in VIX contango. It’s a little easier to see when you filter out some of the noise and look at mean 1-day XIV growth across quartiles of VIX contango. (click to enlarge) Naturally, we’d hope that VIX contango has enough predictive power to pull the distribution of XIV gains a little bit in our favor. The next figure compares the distribution of XIV gains on days after VIX ended in contango to days after it ended in backwardation. (click to enlarge) The mean was higher for contango vs. backwardation, but the difference was not statistically significant (0.22% vs. -0.26%, t-test p = 0.37). Surprisingly the median was a bit higher for backwardation (0.50% vs. 0.86%, Wilcoxon signed-rank p = 0.62). Towards A Better Cut-Point Holding XIV whenever VIX is in contango is somewhat natural, but there’s no reason we have to use 0% as our cut-point. We might do better if we hold XIV when VIX is in contango of at least 5%, or at least 10%, or some other cut-point. Actually if you look at the regression line in the third figure, you can work out that the expected 1-day XIV growth is only positive for VIX contango of 1.65% or greater. Based on that, we actually wouldn’t want to hold XIV when contango is betwen 0% and 1.65%. Let’s compare 0%, 5%, and 10% VIX contango cut-points. (click to enlarge) The higher cut-point you use, the less frequent your opportunities to trade XIV, but the better the trades tend to be. Notice how the 10% cut-point rarely allows for trades, but tends to climb really nicely when it does. Performance metrics for XIV and the three contango-based XIV strategies are summarized below. Performance metrics for XIV and XIV trading strategies with various VIX contango cut-points. Fund Growth of $10k MDD Overall Sharpe Ratio Sharpe Ratio for Trades XIV $26.2k 74.4% 0.040 0.040 Contango > 0% $57.0k 56.3% 0.061 0.065 Contango > 5% $65.1k 37.3% 0.072 0.090 Contango > 10% $49.3k 14.9% 0.110 0.293 Total growth was best for a contango cut-point of 5%, while maximum drawdown decreased and Sharpe Ratio increased with increasing contango cut-point. (Note that “overall Sharpe ratio” includes the 0% gains on non-trading days, while “Sharpe ratio for trades” does not.) Of course we aren’t restricted to cut-points in 5% intervals here. Let’s play a maximization game and see what VIX contango cut-point would have been optimal for total growth and for overall Sharpe ratio. (click to enlarge) Final balance peaks at VIX contango in the 5-6% range, and is maximized at $100.4k for VIX contango of 5.42%. Overall Sharpe ratio is maximized at 0.115 for VIX contango of 9.95%. Sharpe ratio for trades is maximized at 4.231 for VIX contango at the highest possible value, 21.6%. Of course it wouldn’t make much sense to use a cut-point of 21.6%, as that number is hardly ever reached. Play Both Sides of the Trade? If sufficient VIX contango favors holding XIV, it seems that sufficient VIX backwardation would favor holding VXX. That brings to mind a trading strategy where you buy XIV when VIX contango reaches a certain value, and buy VXX when VIX backwardation reaches a certain value. Trading both XIV and VXX would provide more opportunities for growth. Indeed many of the analyses presented so far are similar when you look at holding VXX based on VIX backwardation. In particular: VIX backwardation is positively correlated with 1-day VXX growth. Regression analysis suggests that VXX on average grows when VIX backwardation is at least 0.38% (equivalently, VIX contango is -0.38% or more negative). Growth of $10k for a backwardation-based VXX strategy is maximized at $13.3k, when you hold VXX when VIX backwardation is at least 5.67%. Unfortunately, 33% growth over 5 years with VXX is nothing compared to 900+% growth with XIV. I experimented with strategies that use both XIV and VXX, but was unable to improve upon XIV-only strategies. Concerns One of my concerns with these strategies is that we’re working with a very weak signal. VIX contango explains about one-third of one percent of XIV’s growth the next day. Contango-based volatility trading strategies do appear to have potential, but keep in mind that VIX contango just isn’t a strong predictor of XIV growth. Another concern is that the excellent historical performance of these strategies may be driven by the bull market of the past 5 years. I think it is very possible that in a bear market these strategies might work poorly for XIV, and perhaps well for VXX. Each strategy involves holding XIV/VXX at certain time intervals, so of course they will be affected by the underlying drift of XIV/VXX. After all, the absolute best you can do with either version of the trade is the total upswing in the fund you are trading over a period of time. Finally, I have noticed in the past that XIV seems to have positive alpha when markets are strong, and negative alpha when markets are weak. This makes it really hard to do portfolio optimization, as the net alpha of a weighted combination of funds including XIV actually depends on what sort of market you’re in. I think an analogous problem could arise for contango-based XIV strategies. For example, holding XIV when VIX contango is at least 5% may only be prudent in periods when XIV itself is rapidly growing, which would typically occur in a strong market. And a strategy that only works during bull markets isn’t very exciting. Conclusions A variant of a strategy discussed by Nathan Buehler, where you hold XIV whenever VIX is in contango, appears promising based on backtested data since Nov. 2010. But increasing the contango cut-point from 0% to 5% increases total returns while also improving Sharpe ratio and reducing MDD. Going to 10% further improves the Sharpe ratio and reduces MDD, but sacrifices total growth as there are fewer trading opportunities. Since Mr. Buehler’s strategy is based on the idea that VIX contango favors XIV, increasing the contango cut-point above 0% makes a lot of sense. It allows us to trade XIV only when we have a substantial advantage due to contango, which reduces trading frequency and therefore trading costs. Strategies based on backtested data are almost always overly optimistic, and I suspect that this analysis is no exception. I am particularly concerned that much of the excellent historical performance is due to XIV’s positive alpha during the past 5 years, which itself was due to a strong market. Therefore, I probably wouldn’t recommend implementing these strategies just yet, at least not with much of your portfolio. Personally, I would consider freeing up a small portion of my portfolio for occasional high-conviction XIV trades based on VIX contango. For example, I might buy XIV on the relatively rare occasion that VIX contango reaches 10%.
