Volatility Pair Trading Based On Contango

| About: VelocityShares Daily (XIV)

Summary

SPY price movement explains 69.5% of the variability in daily XIV gains; contango explains 0.3%. Thus contango-based volatility trades are very unpredictable.

The key idea here is to trade the difference in growth between XIV and a beta-matched S&P 500 fund, rather than XIV growth itself.

The correlation between contango and the 1-day difference in gains between XIV and UPRO is reasonable (Pearson's r = 0.16, Spearman's rho = 0.13).

Hold XIV and SPXU when contango > 6.77%, UPRO and VXX when contango < 2.90%, and SPY otherwise. No short-selling is required.

Back-tested performance indicates smaller raw returns compared to Nathan Buehler's "Hold XIV during contango" approach, but better Sharpe ratio, smaller maximum drawdown, and much lower beta.

Contango-based Volatility Trading

There is some great material on volatility trading here on Seeking Alpha. For the purposes of this article, I'm going to assume a basic understanding of volatility products. For readers with less experience, I would suggest reading Nathan Buehler's three-part series:

The Beginner's Guide to Volatility: VXX

The Beginner's Guide to Volatility: XIV

The Beginner's Guide to Volatility: ZIV

While some authors advocate buying and holding inverse volatility products (e.g. the VelocityShares Daily Inverse VIX Short-Term ETN (XIV) or the VelocityShares Daily Inverse VIX Medium-Term ETN (ZIV)) long-term, I think the consensus view is that this is a bad idea. In my opinion, there are 3 major problems:

  1. These funds are extremely high beta (XIV ~ 3.5, ZIV ~ 1.5), so they can suffer huge drawdowns. XIV lost about 70% of its value in the past 6 months.
  2. XIV and ZIV both have lower historical Sharpe ratios than the SPDR S&P 500 Trust ETF (NYSEARCA:SPY).
  3. They both have positive alpha sometimes, negative alpha other times.

Numerous SA authors are working on developing contango-based volatility trading strategies. The aforementioned Mr. Buehler wrote an interesting article in July 2014 exploring the idea of shorting the iPath S&P 500 VIX Short-Term Futures ETN (VXX) whenever VIX futures entered contango. David Easter examined the "buy XIV" version of the strategy in an Oct. 2014 article, and presented an alternative exit strategy. In a Sep. 2015 article, I found that a 5% cut-point for contango seemed to give better results than 0%.

Of course, implicit in the effort to develop contango-based trading strategies is the notion that contango predicts XIV performance. Empirically, I found that it does, but rather weakly. Theoretically, Robin Hewitt recently shared her view that contango is the result rather than the cause of changes in XIV value. This was an excellent article, albeit controversial. I strongly recommend reading both the article and its comments.

The New Idea

In my previous article, contributor Martin Vlcek posted a comment proposing "shorting volatility hedged at certain times by being short the S&P 500 index."

This is a great idea. It solves the problem that even when contango is favorable, XIV price movement is mainly driven by SPY price movement. Rather than using contango to bet on XIV increasing, we can use contango to bet on XIV outperforming (or underperforming) SPY.

Even better, we can implement the strategy without actually short-selling. When contango is high, we want to play (XIV - SPY), which is similar to (XIV + SH). When backwardation is high, we want to play (SPY - XIV), which is similar to (SPY + VXX). No shorting necessary.

Developing the Strategy

First of all, to account for XIV price movement due to market movement, we shouldn't use (XIV - SPY). XIV's beta is roughly 3.5, so when SPY gains 1%, XIV gains about 3.5%. Thus we would ideally play (XIV - 3.5x SPY). But we want to avoid short-selling, so we'll use (XIV + -3.5x SPY) instead. Alas, a 3.5x inverse S&P 500 fund does not exist, so we settle for a -3x fund, say the ProShares UltraPro Short S&P 500 (NYSEARCA:SPXU).

To briefly highlight the importance of trading (XIV + SPXU) rather than XIV alone when contango is high:

Daily SPY movement explains 69.5% of the variability in XIV's price movement (p < 0.001). Contango alone explains 0.3% of variability in XIV's price movement (p = 0.060), but 2.6% of the variability in the average of XIV and SPXU's daily price movement (p < 0.001). So trading (XIV + SPXU) gives you a stronger predictive signal to work with.

Figure 1 shows a scatterplot where the y-axis represents the average daily gain for XIV and SPXU, and the x-axis represents contango. Data for contango and XIV are from The Intelligent Investor Blog, and data for SPXU are from Yahoo! Finance.

Figure 1: Scatterplot of average daily gain for XIV and SPXU vs. contango, using data from Nov. 30, 2010, to Feb. 12, 2016. Click to enlarge

Contango is a highly significant predictor of the average gain for XIV and SPXU. The correlation is small, but not tiny (Pearson's r = 0.16, Spearman's rho = 0.13).

I say we buy XIV and SPXU whenever contango is such that the predicted Sharpe ratio is greater than the historical S&P 500 Sharpe ratio.

The Vanguard 500 Index Fund Investor Shares (MUTF:VFINX) has data going back to 1980, and has a historical Sharpe ratio of 0.0396. From the fitted regression model for XIV and SPXU vs. contango, that Sharpe ratio is exceeded when contango is greater than 6.77%.

Now for the other side of the trade. During backwardation, we expect XIV to underperform a beta-matched SPY, so we'd like to play (3.5x SPY - XIV). We substitute VXX for XIV to avoid short-selling, and use the 3x ProShares UltraPro S&P 500 ETF (NYSEARCA:UPRO) rather than 3.5x SPY. Figure 2 shows the average gain for UPRO and VXX vs. contango.

Figure 2: Scatterplot of average daily gain for VXX and UPRO vs. contango, using data from Nov. 30, 2010, to Feb. 12, 2016. Click to enlarge

Again, we see a fairly weak but highly significant relationship (Pearson's r = -0.14, Spearman's rho = -0.12).

Let's also play this trade whenever the predicted Sharpe ratio is greater than VFINX's historical Sharpe ratio. This occurs when contango is less than 2.90%.

So we hold SPXU and XIV (in equal allocations) when contango > 6.77%, and hold UPRO and VXX when contango < 2.90%. When contango is between these two values, we'll stick with SPY.

Historical Performance

Figure 3 shows historical performance of this strategy, using closing prices for all funds going back to Nov. 30, 2010, and neglecting practical issues like trading costs and bid/ask ratios. It also shows XIV growth and performance of the Buehler strategy with a 0% cut-point and a 5% cut-point. Note that I modified the Buehler strategy slightly, holding SPY rather than cash when contango is not greater than the cut-point. That way there is not a built-in disadvantage relative to the contango-based pair trades.

Figure 3: Growth of $10k for the contango-based pair trading strategy, two variants of Nathan Buehler Click to enlarge

In terms of raw growth, the Buehler methods do better than the contango-based pair trades. But the pair trades have some excellent performance metrics, as you can see in Table 1.

Table 1. Performance metrics for various strategies.

Fund Growth (%) MDD (%) Sharpe Alpha Beta
Pair trades 286.4 18.9 0.101 0.0010 0.14
Buehler 0% 412.4 50.9 0.055 0.0006 2.39
Buehler 5% 693.5 33.1 0.073 0.0011 1.87
XIV 74.5 74.4 0.031 -0.0004 3.49
Click to enlarge

The contango-based pair trading strategy had the best MDD, highest Sharpe ratio, and a very low beta. I think these are very promising results.

Conclusions

To recap the idea presented in this article:

  • Contango is a poor predictor of short-term XIV growth, because XIV growth is primarily dictated by market movement.
  • Contango is a reasonable predictor of the relative performance of XIV vs. a beta-matched S&P 500 fund.
  • To avoid selling short, I suggest a contango-based pair trading approach. Hold XIV and SPXU when contango is greater than 6.77%, VXX and UPRO when contango is less than 2.90%, and SPY otherwise.

As always, backtested performance is likely overoptimistic, and trading costs and bid/ask ratios will detract from actual performance. I personally will not implement this method until it proves to perform well prospectively.

I'd like to acknowledge Nathan Buehler for laying a nice foundation for contango-based volatility trading, Martin Vlcek for commenting on my prior article and planting the seed for the approach used here, and The Intelligent Investor Blog for making historical volatility data available.

Disclosure: I/we have no positions in any stocks mentioned, and no plans to initiate any positions within the next 72 hours.

I wrote this article myself, and it expresses my own opinions. I am not receiving compensation for it (other than from Seeking Alpha). I have no business relationship with any company whose stock is mentioned in this article.

Additional disclosure: Any opinion, findings, and conclusions or recommendations expressed in this material are those of the author and do not necessarily reflect the views of the National Science Foundation.