VIX-based Pairs Trading (Market Neutral) Strategy

20Jan10

This post was inspired by CSS’s VIX as Moderator of Small vs Large Cap Performance.

In a nutshell, when the VIX (CBOE volatility index) is “relatively” high/low, large caps tend to outperform/underperform small caps. I’m going to spin CSS’s observation into a market neutral pairs trading strategy between the S&P 500 (large caps) and the Russell 2000 (small caps).


[logarithmically-scaled, growth of $10,000]

The graph above shows the results of this pairs trading strategy (red) versus the S&P 500 (grey) from 1989.

Because there’s so little day-to-day deviation between the S&P 500 and Russell 2000, we’re using leverage to juice returns. Specifically, I’m assuming the strategy was applied to ProFunds/Rydex 2X leveraged mutual funds (my weapon of choice), so we can ignore transaction costs and slippage.

Strategy rules: if the VIX will close at least 5% higher than the 63-day (1 quarter) moving average of the VIX at today’s close, go long the S&P 500 and short the Russell 2000 at the close. If the VIX will close at least 5% lower than the 63-day average, reverse the pair and go long the Russell 2000 and short the S&P 500 at the close. If the VIX will close within this 5% band, the pair is unpredictable so move to cash.

One last bit of slickness: the S&P 500 and Russell 2000 are not equally volatile. At this moment in history for example, the Russell 2000 is about a third more volatile than the S&P 500. If we allocated the portfolio 50/50%, we’d lose market neutrality. So instead, we’ve weighted each leg in the pair to balance expected volatility.

Numbers for the number lovers…

The strategy has not been particularly potent generating return, but most market neutral approaches aren’t.

The benefit of this type of strategy is extremely low volatility (about 40% less than the S&P 500) and high returns relative to that volatility. I especially like that this strategy trades very frequently yet performance has stayed consistent through very different market regimes.

A Word on Execution

The success or failure of all strategies depends on the specific vehicle being traded. A strategy may work very differently on an asset that tracks an index closely (ex. leveraged mutual funds) than on an asset with significant tracking error (ex. ETFs).

This is less of an issue with longer-term strategies (ex. the Golden Cross) than with shorter-term ones (ex. RSI(2)) because they tend not to rely on such small trading advantages. But this is a GINORMOUS issue for this pairs trading strategy because it is so sensitive to minute differences in price.

Because leveraged mutual funds haven’t been around as long as this test, I simulated daily returns based on the underlying indexes. We can do this because these funds track their benchmarks so tightly, but note that I’ve also tested the strategy directly on the fund data that was available and (surprisingly) achieved slightly better results.

So I’m confident we could achieve these returns on this specific asset, but (and this is a huge BUT), I would not assume these results hold true for any other trading vehicle.

Note: (a) click for a follow up to this post, and (b) daily updates for this strategy are available on the new Addendum to the State of the Market report.

Happy Trading,
ms

. . . . .

To stay up to date with what’s happening at the MarketSci Blog, we recommend subscribing to our RSS Feed or Email Feed.



29 Responses to “VIX-based Pairs Trading (Market Neutral) Strategy”

  1. 1 JB

    Do you vary the relative position weight as a function of relative vol over some lookback period or is the relative position weight fixed to some constant?

  2. 2 MarketSci

    RE to JB: good question. It’s over a lookback (1-year in this test), so it will constantly drift. You can see the %’s for this strategy in real-time on the addendum:

    http://marketsci.wordpress.com/state-of-the-market/the-addendum/

    michael

    • 3 JB

      Thanks, I can’t get the 1989-1999 performance to match up using my implementation. Do you treat that period any differently?

      • 4 MarketSci

        RE to JB: Nope. I did forget to mention in the post that pre-1990 I used the VXO in place of the VIX, but besides that it’s the same. If it’s still giving you trouble, feel free to send it over and i’ll take a look. michael

    • 5 JB

      VIX data had a couple of holes as compared to the index data. I’m able to replicate now. Thanks!

  3. 6 Kevin

    Michael – on the VIX Pairs Trading strategy, did you assume trades were placed the next day or did you anticipate the close same day (like YK)?

    Thanks for all your sharing of ideas.

    Kevin

    P.S. – I’ve loved the last several days’ YK returns!

    • 7 MarketSci

      RE to Kevin: you and us both.

      I assumed the trader anticipated the VIX close (like YK).

      michael

  4. 8 Ben

    Hi Michael,
    Does the back test assume daily rebalancing to account for profit/losses that would lead to portfolio directionality?

    Cheers, Ben.

    • 9 MarketSci

      RE to Ben: very good question. It actually rebalances daily (remember, I come from the leveraged mutual fund world where this isn’t a big deal to do). michael

  5. 10 Jim P.

    Great post Michael. A couple of questions for you:

    Re. the relative position weight (% allocation)… will this stay the same for the short term – next trade or two,..and change only infrequently,..or will the % allocation change substantially from one trade to the next?

    Is there any way for an individual trading the system to simply calculate the % allocation?

    And lastly,…how would the VIX-Based Pairs Trading program’s returns be different if one used a simple 50%/50% FIXED Long/Short allocation (still using the same 2X funds, etc)?

    Thanks in advance for making your work available. Best Regards, Jim

    • 11 MarketSci

      RE to JimP: hello sir – always good to hear from you. The % allocation (SP vs R2K) will change VERY slowly, so shouldn’t be a problem just say looking at the Addendum once a week or month and going with that (it’s the same split we do w/ YK). The 50/50% mix performs a bit worse in risk-adjusted terms (if I remember correctly) b/c it adds unnecessary beta risk (sometimes long and sometimes short).

      michael

  6. 12 Kevin

    Michael – Can you provide some insight as to the number of trades per year the strategy generates assuming one is not rebalancing daily?

    Kevin

    • 13 MarketSci

      RE to Kevin: over the last 5 years, about 50/year. I’m counting each side of the trade as 1 trade, including moving to/from cash or changing the direction of the pair. I’d have to do a little work to go beyond 5 years (I took that number right off the addendum). michael

  7. 14 John French

    Very interesting Michael, I will code myself when I get the chance.

  8. 15 MarketSci

    FROM MARKETSCI…

    I suspect based on the comments here and the flurry of emails that I’ve received that we have quite a few very smart folks crunching numbers on this strategy. I have to confess that this one has my goat at the moment as well.

    I’ll be posting a follow up to this post in the next 24 hours or so w/ some additional thoughts and research.

    michael

  9. My first reaction was also Fantastic! But looking at the graph, it appears that most of the gain came from 1991 – 1992 and 1998 – 2003. (These are rough eyeball estimates.) Most of the rest of the time, the curve is relatively flat. Of course that’s better than losing money, but it’s not a steady significant winner.

    It was interesting to trace the history of the public postings about this. (It’s 3 1/2 years old.)

    * Leistikow, Dean and Susana Yu (July 2006) “VIX Signaled Switching for Style-Differential and Size-Differential Short-term Stock Investing”. (http://www.fordham.edu/workingpapers/images/VIX%20September26.pdf) Link no longer good.)

    * LeCompte, Steve (July 2006) A Short-term VIX Trading Strategy That Works? CXO Advisory (http://www.cxoadvisory.com/blog/external/blog7-12-06/)

    * Leistikow, Dean and Susana Yu (2009) Vix Signaled Switching for Style-Differential and Size-Differential Short-Term Stock Investing, SSRN. (http://papers.ssrn.com/sol3/papers.cfm?abstract_id=912786)

    * Varadi, David (2010) Research Synthesis: The VIX as a Moderator of Small vs Large Cap Performance and the “Generals Lead the Troops” CSS Analytics (http://cssanalytics.wordpress.com/2010/01/17/research-synthesis-the-vix-as-a-moderator-of-small-vs-large-cap-performance-and-the-generals-lead-the-troops/)

    * Stokes, Michael (2010) VIX-based Pairs Trading (Market Neutral) Strategy, MarketSci (http://marketsci.wordpress.com/2010/01/20/vix-based-pairs-trading-market-neutral-strategy/)

    One of your great skills, Michael, is that you get to the essence of things and explain them clearly and simply.

    • 17 MarketSci

      RE to Blue: you stole my thunder sir =)

      I’m going to do a follow up today making both the points you made above: showing some previous work done on the subject, and talking about how the strategy appears to be losing effectiveness.

      Thanks for the kind words,
      michael

  10. Is there any notable characteristic about the periods when the strategy did work well? That is, besides the fact that the strategy worked well, is there anything else about those periods that distinguish them from other times?

    • 19 MarketSci

      RE to Blue: that’s a good question.

      One interesting characteristic I found was that the strategy performed considerably better in the beginning and middle periods you mentioned when the average daily divergence between the S&P and Russell was also particularly high.

      Normally that wouldn’t be such an impressive observation, but during the 2008/09 crises that average divergence also became very, very high, yet there was no corresponding increase in strategy effectiveness (a strong indicator that strategy effectiveness is waning).

      michael

    • Here’s a hypothesis. It seems to work best (again on raw eyeball estimates) when the VIX is rising. But the VIX tends to be rising when the market is falling. In those conditions, small caps tend to fall faster than large caps. So what this seems to be saying is: be long the S&P 500 and Short the Russell 2000 in bear markets.

  11. 23 JB

    If implemented using funds, the capital required is 2X because of the long and short legs… right? Doesn’t seem like that is being taken into account in the numbers despite the 2X leverage? thx.

    • 24 MarketSci

      RE to JB: i’m not sure I understand your question, but the short answer is NO, twice the capital wouldn’t be required.

      The same capital split between two sets of 2X funds means twice the capital is “as risk” (b/c of leverage) but twice the capital is not required to put on the trade. michael

  12. 27 pairtrading

    Interesting article thanks!

    A step further would be to look for stocks within each group that have certain valuation and momentum characteristics, for example when going long stocks you should stick to stocks with the highest 12 mo % gain and the lowest PEG reading, and vice versa for going short stocks. I haven’t tested this, but Im sure using the theory presented in this article in tandem with my above filters would result in one powerful stock trading system.

  13. 28 John French

    Michael, could you detail the methodolgy you use for “So instead, we’ve weighted each leg in the pair to balance expected volatility”? The only further information I saw said ” It’s over a lookback (1-year in this test), so it will constantly drift.”

    Thanks

    John


  1. 1 Perfect Alignment and S&P500 Returns « CSS Analytics

Leave a Reply

Please log in using one of these methods to post your comment:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s