Alpha Strategy

Updated 9/30/2011

Overview

The alpha strategy has one objective:

  • High total returns

A modest portion of my overall portfolio will be dedicated to this strategy.  Volatility and risk are crazy high.  Potential returns are also crazy high.  We’re talking annualized returns approaching 100% over long periods.  That comes at a price.  It’s possible for this strategy to lose a ton in a short amount of time.

Details

I spent a lot of time researching the new volatility ETNs on the market.  I was mainly focused on two – VXX and XIV.  Do a google search if you want to learn more about them.  There’s a lot to learn so make sure you take the time to do so.  (Note: I’m greatly indebted to Juan of the Inteligent Investor Blog.  He posted a super helpful spreadsheet of simulated VXX values in this post: http://investing.kuchita.com/2011/08/21/vxx-data-since-vix-futures-avilable-march-2003/  The charts of XIV and related strategies below are based off of values in his spreadsheet.)

From what I read it seemed like XIV could be a big winner over time due to the contango that often occurs in the first two VIX futures.  The problem is it’s subject to huge drawdowns during bear markets and large volatility spikes.  I tried many different strategies to time XIV/VXX entries and exits and in the end found a couple variables that appear to be super helpful in timing these plays.

One valuable key to a  timing strategy appears to be the volatility of volatility.  More specifically, I’m referring to the standard deviation of the VIX index during the recent past.  The chart below compares simulated XIV values to the standard deviation of daily percent changes in the VIX index over the previous 10 trading days.  You’ll notice the standard deviation spiking right as XIV prepares for major plunges.

Using the following rules would have theoretically made one a lot of money in the past:

Rule 1:  Go long VXX when the standard deviation rises above 11 percent.

Rule 2:  After Rule 1 has gone into effect, go long XIV when the standard deviation falls below 10 percent.

Here is a look at the results of such a timing model:

Another valuable piece of information is the relationship of the first and second month futures over the recent past.  The average level of contango tends to be high when XIV is rising.  The opposite has been true when the futures are characterized by strong backwardation.  The chart below compares XIV to the average front month premium over the previous 20 days.  Negative values represent contango and positive values represent backwardation.

Using the following rules would have also theoretically made one a lot of money in the past:

Rule 1:  Go long VXX when the average front month premium over the past 20 days is above 5%.

Rule 2:  Hold cash when the average front month premium over the past 20 days is between -5% and 5%.

Rule 3: Go long XIV when the average front month premium over the past 20 days is below -5%.

Here is a look at the results of this timing model:

Both models provide great returns with much less drawdown than simulated XIV values.  That said, both did have significant drawdowns during the testing period.  The problem with the VIX standard deviation model is that sometimes XIV will gradually drift lower when the standard deviation is low.  This was the case in parts of 2007.  Long periods of backwardation combined with a low VIX standard deviation are a big threat that need to be addressed.  The problem with the futures spread model is that it wont react fast enough when VIX is spiking fast and hard.  This was the case in 2010.

A hybrid approach addresses the weaknesses of each individual model.  It will keep one out of XIV during periods of backwardation/ weak contango even when the VIX isn’t super volatile.  It will also get one out of XIV faster when the VIX is spiking crazy fast.

The Alpha strategy uses the following rules:

Rule 1: Go long VXX when the standard deviation of VIX over the past 10 days rises above 11 percent.  Stay in VXX till the standard deviation falls below 10%.

Rule 2: Go long VXX when the average front month premium over the past 20 days is above 5%.

Rule 3: Hold cash if the average front month premium over the previous 20 days is between -5% and 5% AND Rule 1 is not in play.

Rule 4: Go long XIV when the average front month premium over the previous 20 days is below -5% AND Rule 1 is not in play.

Below are the results of the Alpha strategy compared to the other two models:

Stats related to all three model backtests can be seen below.  Max drawdown for the Alpha model was significantly less than the other two.  Returns weren’t quite as good as the StDev model but the Alpha model should be much more robust for varying market conditions in the future.

One final rule needs to be added to help protect against monster one day jumps in the VIX index (think VIX spikes greater than 50-100% or more in a day).  A stop should be utilized when in a XIV position.  Ideally, I’d like to have a 10% stop for each day that I’m in XIV but I don’t want to set that up each day and don’t know how to do it automatically.  So…I plan to use a 15% trailing stop that i will readjust if the price gets close to the stop value.  This still doesn’t offer 100% protection.  A monster gap down could still theoretically decimate XIV.  Let’s hope that doesn’t happen!

  1. This is very good work. Big fan of your blog (which i just found). We have VERY similar interests, hobbies, and thoughts.

    I will caution you on one thing… the period from 92-04 has a few wrenches to the above system… there are several periods very similar to 07-early 08 but worse.

    I would strongly recommend a much more hedged approach to protect from brutal (albeit gradual erosion) draw down from long periods of backwardation. My findings show that the years 00-jan03 the market spent more days backwardated than in contango and 96-98 was almost the case as well… i know its hard to believe but think about how biased you are now by limited data (i was too). Less we also not forget the other risk, a real meltdown, ie 1987. I assume their was no warning for that draw-down unless you had perpetual hedges in place. Estimates for that event’s VIX (from VXO data) are in the 180plus area in hours…. scary.

    i have founded tactically balancing the ratio of short and long volatiliy along the curve works best. Staying closer to an XVIX type ratio as a core works best and is robust for the unseen environments in which i mentioned. XVIX is actually a great product which will prevail in time.

    Such a new frontier with this stuff… enjoy the epiphany every 3 days :) i have been having mine for almost 6 months now on this specific topic. (i can tell you have as well)

    Best,

    D

    • Damian,

      Thanks for your words and for sharing some of your findings and advice. I really appreciate it. It’s nice to hear from someone else who’s been researching this stuff. The epiphanies keep coming and I imagine this strategy will be different in a week. :-) Among other things, I view this blog as a journal of findings. Ideas will continue to evolve…

      I haven’t heard anyone talk about contango/backwardation pre-2004. Where did you find this info? Is there futures data out there for the old VIX index (assuming futures were traded on it)? I plotted out my timing signal to 1990 and it hardly signals any warning in the 2000-2003 period. That makes me nervous – especially if futures spent as much time in backwardation as you suggest. And yes, there is always 1987… I assume XIV would have had a termination event that day…very scary indeed!

      Thanks again for the feedback and best of luck in your investments!

      -Mike

  2. well funny enough… I got access to some research I was not supposed to from Barclay capital via an emailed link from a friend who worked there. I called the firm and tried to get more information and they black listed everything… literally within a day or two every trace online that had links and all email links were password protected. (i might even have a hit out on me lol)

    In it was a model portfolio for the XVZ… which ended up coming out as a product 3 months later… which is what i assume all the panic was all about on their part.

    however in that doc they also needed to create a synthetic basket of futures… which dont exist to further test this concept. I assume they only published results till 04 because of liability over hypothetical of hypothetical returns, but i assure you that from what i found product like XVIX and XVZ killed it over the past 20years.

    they imported sp500 options, to recreate vix, then they used their models between sp500 skew relationships and vix future relationships to create a proxy vix futures market. next they calculated the term structure and then created data for vxx and xvz going back to 1991.

    I wish i still had the doc… as i have not been able to come across it again… I know better now to save everything as a pdf!!!

    -D

  3. Nice, I wish I could have seen that data! Seems like a lot of work to construct a synthetic futures basket going back that far. Too much work for those of us with day jobs. :-)

    XVIX and XVZ intrigue me. I need to take more time to learn about them. A much safer strategy would be to go long XVIX (or some combination of XIV/XVIX) instead of XIV in my current strategy. This would hopefully still make some money when the VIX standard deviation is low. I’m not as worried about holding VXX when the volatility is high. Short term risks are much less than XIV and the typical holding period (excluding late 2008) was fairly short. The above strategy would have still made very good gains by just going long VXX when the volatility of VIX was high and staying in cash at other times.

    I’m working on a more hedged approach but have been too busy to work on it lately. Thanks again for sharing your findings!

  4. I am doing something similiar to what you just described… however directly in futures market not etp space… I wish to avoid management fee, plus path dependancy of XIV can be a risk of its own (at the very least it is a variable i cant hedge easily with a long vol position).

    moreover, i have been using vix options in addition to futures to execute a more diversified hedging approach.

    Side thought-

    knowing the inverse relationship of vix to equities(spx namely)…. have you thought of using your stdv model as a flight to cash indicator for your other holdings? Just curious… on the surface seems like it could work.

  5. That’s a good thought about using a stdv model for other holdings (e.g. equities). I’ve tested a couple ideas for the SP 500. I haven’t found anything great yet – that is anything that doesn’t get whip-sawed to death. I think there’s still promise with the idea though.

  6. Hey guys, thanks for the great post here.

    I am doing almost exactly the same thing as you are, i.e. switching between long VXX (sometimes short VXX put) and long XIV (sometimes short VXX call). I use a hysteresis algorithm to time the switch, and right now, it’s entirely based on the spread between 1st and 2nd VIX futures. I just switched all my long VXX to long XIV Friday because the spread dropped to -0.2, so we will see what happens next.

    As for Damian’s comments on volatility pre-2004, I have found that when volatility is choppy for a long period of time, both XIV and VXX can lose value, so during these times, it is best to short VXX put (to replace long VXX) or short VXX call (to replace long XIV). However, I have yet to find a meaningful indicator to switch into this strategy.

    Actually, the current environment may be a good candidate for this type of strategy, if VIX doesn’t drop below 30 for another month or so.

  7. Kim, thanks for your comment! I had to look up what hysteresis means. :-)

    I think both the futures spread and volatility have a lot to offer when timing these plays. I’m working on a modification to the model that will either be in XIV, cash, or VXX depending on both the spread and volatility. Per Damian’s comments, I’m afraid to make my model rely solely on VIX volatility.

    Here’s a look at what I’m thinking for the new system:

    -Long XIV when average second month premium over past 20 days is greater than 5% AND the StdDev of VIX over the past 10 days remains low.
    -Cash when average second month premium is less than 5% AND StdDev of VIX remains low.
    -Long VXX when the StdDev of VIX is high.

    The results still seem great with smaller drawdowns and less volatility and implicit risk. I plan to update this page when I get a little more time to think it over.

    Best,
    Mike

  8. Thanks for the great post and blog. Similar interests here too…
    Just to make sure I understand your strategy well since your last post and following comments, do you now have 3 or 4 rules on ? Pretty clear on XIV but would you go long VXX on two different events (stdev of Vix high AND/OR? -5% disc on front month futures?)
    Also practically where do you get the data on futures prices ? On the CBOE website each day and you calculate the 20 day average discount using Excel, or -like me- you can access a Reuters/Bloomberg terminal?

    Charles

  9. Hey Charles, glad you like the post and blog.

    Yes, there are now 4 rules (5 if you count the loose trailing stop for XIV). I plan to switch to VXX if either the stdev rises enough or backwardation is strong enough (rules 1 and 2). The hopping into VXX on the backwardation condition is the one I’m least confident in. Except for 2008 and 2011 there just haven’t been many days where we’ve seen that kind of backwardation. I’m leaving it for now though.

    I don’t currently have access to any fancy data services. I have a google docs spreadsheet that I update the futures and vix data on. I use either the CBOE website or barchart.com for the futures prices. I can get (near) realtime data for VIX from the CNBC website. Using delayed data, sometimes I won’t know for sure if a signal happens until after the markets close. This was the case yesterday when the average contango turned out to be something like 4.996%. Sheesh.

  10. Excuse my ignorance but how do you calculate the vix standard deviation ? Where do you take yours numbers ? Thanks for sharing yours inputs.

    • There are several ways to do it. I calculate the standard deviation based on daily percent changes in the VIX. For example, last November the VIX experienced the following percent changes over a 10 day period: -5.84%, -6.84%, -1.11%, -1.03%, -7.94%, 31.59%, -9.26%, -8.44%, 3.63%, 0.29% I then use a google docs formula to calculate the standard deviation. Here we could get the answer by typing the formula =STDEV(-5.84%, -6.84%, -1.11%, -1.03%, -7.94%, 31.59%, -9.26%, -8.44%, 3.63%, 0.29%). This would give us a standard deviation of 12.07%. The formula is basically identical if you use excel.

      VIX closing values can be found multiple places including the CBOE website and yahoo finance. Note that a closing value doesn’t occur til about 15 minutes after US equity markets close. Official signal changes for this strategy will occur AFTER the market is closed for the day. My backtesting is based on changing allocations the day a signal change occurs. This is unrealistic since it’s impossible to tell where the VIX and VIX futures will settle before the equity markets close. This doesn’t bother me much though since positions are generally held so long. In most cases the allocation change would have to take place the following morning.

  11. First off, love the site, nice work! Have you ever thought of just using a moving day average cross over as you timing signal of when to get in and out of XIV/VIXX? It seems by just taking a quick glance that you might hold on to more gains and avoid big draw downs. I’ve been playing around using a 20EMA/100SMA crossover with a 50SMA as a trailing stop. I’m not a savvy as you and your followers, a novice at best, but would love to hear your feedback. Also any thoughts of using SVXY/UVXY?
    Thanks,
    Eric

    • Hey Eric, thanks for the comment! I’m glad you enjoy the site.

      I’ve tested a few moving average strategies in the past on XIV/VXX. None of them seemed to work as well during my back-test period. There probably are variations of MA strategies that would work better than the strategy on this page. I just haven’t seen any yet. Quick question, when would your crossover system re-enter after the 50 SMA stop gets hit?

      Yeah, I am a fan of the ETF versions of XIV and VXX (i.e. SVXY and VIXY). I would personally avoid the leveraged products like UVXY. There’s potential for huge gains but you need to have great timing. VXX is volatile enough for me. :-)

  12. I would get back in when or if the price topped the last high. If you don’t mind, when did you enter this last XIV trade? Yeah, I hear ya on the extremes of UVXY/TVIX, but it’s fun to dream “what if”.

    • Initially got into XIV on December 13. Went to cash on January 18th and then got back into XIV on January 23rd and have been in since. It’s been a wild ride. Strategy isn’t for the faint of heart!

      You can check out when any of the strategies changed allocations by clicking on the “transactions” category on the right side.

  13. Cool, I’ll check that out. Looking forward to seeing the revised Stock/Bond Timing charts. Good stuff!

  14. Hi Mike,

    I am studying some products related with volatility like VXX, XIV, XVZ and so I found your blog. I saw your “Alpha” strategy and it seems very interesting. I would like to check this strategy with my own data (I don´t have the database of Investing Kuchita), but I don´t know how to calculate the “Spread model”. I´m Spanish and I don´t understand what you want to say when you say “The chart below compares XIV to the average front month premium over the previous 20 days”. I think that it is a problem with the language, but I would be grateful you if you could explain me with an example with numbers like you made in the above message of DBLT.

    If you want I can contact with you by mail and I can send my results when I finish it. My results in “Stdv Model” were a little different than yours, because the database that I used was different. I bought it to Vance of “sixfigureinvesting.com”.

    Thanks for your help

  15. Hey Fernando, no worries. I’m basically referring to the price ratio of the first and second month vix futures.

    The “front month premium” I use is calculated as follows:
    First month vix future / Second month vix future – 1

    For example, today the July future settled at $18.80 and the August future settled at $20.87. The front month spread is 18.80/20.87-1 = -9.92%

    The average front month premium is just a 20 day simple moving average of that spread.

  16. It seems to me that using a rising standard deviation in the VIX to “time” a drop in XIV price is flawed, simply because they both happen at the same time. One does not precede the other in any repeatable or measurable way. Don’t forget, the only reason the 10 day standard deviation of the VIX is rising in the first place is because the VIX is experiencing the larger price changes. That means the daily changes in the XIV are also happening at the same time as the rising 10 day standard deviation. You can’t use an effect to predict a cause.

    All you have to do is look at the chart that you posted above. In 2008, the XIV began tanking at the same moment that the VIX’s 10 day standard deviation began to rise. By the time the VIX standard deviation would have signaled anything to you, the XIV was already near it’s bottom.

    Of course there are a few examples of where the 10 day would have helped you, but there are about an equal number of instances where it wouldn’t have. I think you’d be better off trying to determine the “trend” of the 10 day VIX standard deviation, rather then using the actual SD itself.

    • It’s not a perfect indicator but it’s not as bad as you make it out to be. It’s true there are times when it does work well and times it doesn’t work well. I also agree that XIV will be experiencing a similar increase in volatility as the vix does.

      2008 is the best example of when it worked really well. Between the time the std dev rose to high levels in Sept 2008 and the time it fell back below these levels in November 2008, XIV would have lost about 75% of its value.

      2010 was another time when it would have paid off big to, at the very least, be out of XIV when the std dev was high.

      Again, it won’t always pay off or work for every scenario. It won’t get you out at the top. I feel there is value to it nonetheless.

  17. I agree there is some value to it and I definitely give you credit for outside the box thinking because tracking the SD of the VIX is not something that many people are playing around with, but my point was it is about as reliable as any of the other half dozen or so indicators people try to use to “time” when to get out of XIV. That is to say, basically hit and miss.

    The truth of the matter is, it’s nearly impossible to not make nice gains in the long run trading XIV systems. As for the “holy grail” so to speak about when to exit trades before you give back a good chunk of your run up gains, I don’t believe there is an effective way out there. If anyone does solve the mystery, they will be very rich indeed.

    I really enjoyed reading your thoughts on trading the XIV though and will continue to follow along with interest. I have tried for more then a year now to improve on my initial XIV system I came up with and have had little to no success. I’m all but convinced that no amount of geeky research on the subject will improve gains beyond a simple, go long XIV during contango, and add in some VXX during extended periods of backwardation.

    I’ll keep trying though, and I know you will too. The potential gains are there. We just have to figure out the right set of system signals…

    • I appreciate the comments. I need critical feedback to find weak spots in my thinking. I think most of the gains from these systems in the long run will be from the roll yield and not great timing of the vix futures. So far contango/backwardation has been a great indicator.

      Take care

  18. Two things:

    1. Is there an online source for live data on when VIX is in contango and when in backwardation?
    2. What do you think about using option puts as insurance against severe losses (as opposed to stops, which don’t protect one against gaps)?

    And, thanks for all the great ideas on this site.

    • Check out the site http://vixcentral.com/ for one source of futures quotes. It’s not real time but close to it. I don’t know of any sites that offer free real time VIX futures quotes.

      Conceptually, I like the idea of using options as insurance. I also like the idea of using a completely different approach that is less volatile (for instance always having a mix of short and long volatility, something more similar to XVIX or XVZ). I would expect performance of such systems to be much lower, depending on how much risk you want the system to be exposed to.

  19. I wonder if VVIX is a better indicator than the 10 day historical std of the VIX. It wouldn’t have the 10 day lag for a start.

  20. OK I tested the VVIX and it has no predictive ability on its own for VXX from August 2006 to today.

  21. Have you been trading the Alpha strategy since posting it, and if so, is it possible to see an updated chart for it’s results? Thanks

    • I have been trading it, though I didn’t start doing so right away. I spent time debating whether to get into VXX mid-signal last year. The strategy had a big drawdown in October, 2011 and I lost a bit once I did get in. The gains since then have been great.

      You can see 2011 performance here: http://dontfearthebear.com/2012/01/02/2011-strategy-performance/ I’ve been updating 2012 performance periodically on the bottom of the charts page. You can also track the strategy indicators at the top of the same page.

      Individual transactions for this and other strategies can be seen by clicking on the “Transactions” category.

  22. Ah found it, thanks. Looks similar to several other XIV system returns, in that the gains are huge, but drawdowns are painful. I’ve yet to find a way to “set and forget” a system for trading it, and not have to suffer through those drawdowns. The person who does figure out how to capture most of those gains during upswings and not give too much back during drawdowns will be a happy trader. Good luck

  23. Mike, have you had any further thoughts on using stops?

    Also, here’s a very simple system that seems to do very well. Buy XIV when it’s above its 2 month monthly average on a monthly basis (i.e. at the close of each month buy XIV if it’s above its 2 month MA). Otherwise, stay in bonds (TLT, for example).

    I’m playing around with this portfolio (and others) on ETFreplay.com. Unfortunately, I can’t use simulated XIV values on this website, so the results only go back two years or so.

    • Hey Alex, I ran a test of your XIV/TLT system. Results were great going back to 2004. About 71% CAGR with max DD of 50% in 2008. Very low drawdowns 2009-2012.

      Would you be willing to let me do a post on it (giving you credit for the idea)? May not get to it right away though as I’ll be pretty wrapped up for the next few weeks (wedding is next Saturday!).

      Mike

      • Mike,

        By all means do a post.  The whole point of all this is to share ideas.

        And, I wish you a wonderful, life-affirming wedding.

        Best, Alex

        ________________________________

  24. One more thought: it occurred to me that TLT might not be a good cash substitute as far as future performance is concerned. It’s unlikely that bonds will perform as well in the future as they have for the past ten years or so. Perhaps a short term bond fund (BSV) will give a more accurate picture of possible future performance of this system.

  25. The contrarian in me tends to agree that Bonds may have reached the end of their run, but the truth is people have been saying bonds are at the end of the bull run for many many years now. I trade TLT quite a bit, and the longer this run goes on the harder it is to take long positions (or at the very least neutral positions which I mostly do through Iron Condors) However, it’s a trade that has been paying off for so long now, it’s hard to justify saying that it’s over. I think the best thing to do as a trader is to just trade what’s in front of us, and wait for confirmation about whether a run is actually over or not.

    Remember, as low as rates are right now, there is nothing saying they can’t continue down and actually go negative. There are several real world examples of that going on right now. Why not in the US? Just a thought…..

  26. Mike,

    I was wondering if, when you first developed the Alpha strategy, you did a lot of optimizing? In other words, did you come up with the basic idea and then optimized the settings, or did you just come up with the idea and then test the results?

    I ask because I wonder if your stellar backtests are due in any significant measure to curve-fitting.

    Alex

    • Hi Alex, some parameters were more optimized than others. For the spread component, I didn’t test periods other than 20 days or a -5%/5% spread for the cut-off. For the standard dev component, I didn’t test periods other than 10 days. The 11%/10% cut-off was optimized to a degree.

      Not all optimization is bad. Good optimization will seek to find parameters that are clustered together and work consistently well. For example, if using a 10, 15, 20, 25, and 30 day look-back period all provide consistently good results for the spread component, I can be more confident that 20 is a reliable number to use. If the results swing wildly when I switch from 15 to 20 or 20 to 25, I know that the system is not stable. This is when curve fitting becomes a problem.

      If anything, I did not optimize the system enough. It might just be that I got lucky picking these parameters. I’m trying to be more mindful of good optimization as I move forward. A system with an awesome backtest whose returns vary greatly with small tweaks to the parameters should not inspire confidence.

      Mike

  27. Your response inspires confidence. It is absolutely true that a system in which backtested results vary significantly with small changes in basic parameters is likely to be the product of curve-fitting. On the other hand, if these changes produce smooth variations, then the system is likely to be robust. Of course, the real test is live trading, but your response suggests that that we have a decent chance to pass that test.

  28. Thanks for sharing your hard work. With regards to DD have you ever thought trading your equity curve. For example you could add a 20 sma on your equity curve and increase decrease risk (or simply stop trading and go cash) when the sma is above/below the curve. This kind of technique does not work with all systems, you could miss some run ups when the equity is below the sma, in addition it’s one more way to curve fit your system, nevertheless you could give it a try and see how it impacts on your profits vs DD’s. I mentioned using sma on your equity but you could try any indicator like RSI or Bollinger Bands.

    Al

    • Hi Al,

      I’ve tested trading the equity curve with moving averages. I haven’t come accross anything great yet. What appears promising while looking at a chart becomes dissapointing when actually tested.

      • Hi Mike,

        Thanks for your prompt reply. In an attempt to replicate your alpha indicators on my platform (ninjatrader) and datafeed (DTN IQ feed) I found some discrepancies with the 20 sma. You can view some pics and check data to the following links:

        Thx

        Al

      • Hi Al,

        Are you using Oct/Nov futures to calculate the ratio back in August? The b*#$ with the SMA I use is that a new future is introduced every month when the front month expires. I recommend downloading market sci’s spreadsheet for historical data. His differs a bit from the one I originally downloaded (from the kuchita link). The market sci one doesn’t switch futures till the day after expiration while the one I originally used changes the day of expiration. This causes some differences but nothing major in the grand scheme of things.

        Mike

  29. This may be of interest regarding the use of equity curves to turn systems on and off: http://www.breakoutfutures.com/Newsletters/Newsletter0205.htm

    • Thanks Alex, I’m incorporating position sizing into a monthly model I continue to refine. Instead of using an sma, I’m using a rate of change to continually increase/ decrease position size. I think the same principles could be applied to the alpha strategy.

  30. The futures are back adjusted (current->front month) as for http://cfe.cboe.com/Products/historicalVIX.aspx, I haven’t checked the whole data set but the initial part was matching. I will check sci’s nevertheless differences compared to my data should be small. I really don’t understand the gap bewteen your data and mine. In order to experiment and start backtesting the system with my platform I need to sync the alpha indicators. I will also check kuchita’s data and come back asap.

  31. Hi Mike,

    I am now in line with the Kuticha excel 20 sma http://tinyurl.com/95ydbwu , my VX futures were not back adjusted correctly, I need to figure out why, meanwhile I will export the Kuticha data into my platform and start playing a bit, hope to contribute with new ideas. Can I ask what kind of platform do you use for your backtesting.

    Al

  32. I started to play with my platform, for example you can notice (see chart on the link below) how the supertrend indicator (Atr=3 Period=3) optmized the VXX exit signal letting run some extra profits. Basically I use same entry point as alpha (above 10% StdDev) and exit on the supertrend signal (red arrow). This is just one operation I need to backtest with a consistent amount of data but it’s a good start. I have the feeling during strong and long trending periods the supertrend indicator can outperform the StdDev signal. You can also notice that the signals from Supertrend (Green and Yellow arrows) of the VIX and XIV are always matching.

    http://tinyurl.com/8wl7xna

  33. Nice ideas Mike. I played around with the StDev model and tried back testing using following day prices to open/close positions since you may not know a signal until after close. The profit was cut by 60%. This seems extreme and I may have done something wrong. I was wondering if you tried the backtest that way.

    • Hi Roy,

      Using the next day’s price (especially the open) is definitely a more realistic assumption for the model. As you say, the signal may not be evident till after the close.

      During my backtest period of 2004-late sept 2011 I found using the next days closing price cut the annualized return on both the stdev model and overall model by about 7%. A 7% annualized return difference can make a big impact on total return over time.

      • Thanks. Will check my calculations. One other question: There doesn’t seem to be a standard deviation spike above 11 until 4/2005 as shown in your first graph, but your stats show 165% for 2004. Are you starting the backtest with a position in XIV since the system must always be either long or short?

      • Yeah, I start the standard deviation model in XIV. I didn’t wait for the first spike above 11 to come and go.

  34. Interesting. Do you happen to know how using the following day’s opening price would have affected returns?

  35. These are my Open/Close stats according to yahoo’s data. I counted 19 signals for VXX and 16 Signals for XIV (Kuticha 2004-Uptodate) http://tinyurl.com/98nv29j

    • Interesting. Are those stats for only signal days or for all days. I would expect fast moving prices to happen around std dev spikes.

      • Stats are for all days. In order to avoid the Close/Open issue we could:
        1) Set a limit order on the closing value, order could not be filled in case market opens & stays up
        2) Calculate what would be the closing value to reach STD DEV =11 and set a limit order accordingly but the market could close to a lower value and stay down

        in both cases it is very hard to replicate an existing strat on the closing value.

        From my point of view the number of signals generated by the alpha strat are too small to be reliable, I would like to make some walk forward testing to check results on a more consistent set of out of sample data. My issue with the historical data are:
        1) Missing open high low close Before 2009
        2) Vix 1st and Vix 2nd futures are not accessible in a way that works with my platform

  36. Of course, it’s possible to trade after hours. XIV appears to remain pretty liquid for quite a while after market close.

  37. I was curious about what the performance of the Alpha system has been ytd. Thanks.

  38. Mike, back on August 23, 2012 you answered a post from Alex regarding a XIV/TLT system that seems to have a lot of promise. Do you have any more information on it since then? Thank you.

  39. Mike, since vxx split about 1.5 months ago, vxx dropped 15% while xiv only gained 9%. This is of course due to it being an “inverse”. I don’t recall if you have looked at shorting vxx, gradually increasing the position as vxx falls, rather than long xiv? Obviously, the risks are greater but the inverse mechanism of xiv really hurt it.

    • Hi Fisher, I’ve spent a little time in the past on this. If I recall correctly, shorting vxx isn’t always a slam dunk compared to going long xiv. Sometimes a long xiv position will do better over a given period (say, a month). It depends on the type of path the two take. I haven’t researched it extensively and would be interested in any future research you do on the subject.

  40. If you bought XIV on Nov. 1, 2010 and held it until now, you’d have a profit of around 102%. If you shorted VXX on the same date, your profit would be around 86%.

    This is a crude way of looking at Fisher’s point, but it is instructive.

  41. The key is to systematically increase your short position. Say every 10%, you short more to get back to your original exposure. Otherwise you’re limited to no more than 100% even in the best case scenario.

    You start by shorting $1,000 worth of vxx on Nov 30th 2010. It drops 10% from $49 to $44, so you go and short $100 more. For every 10% drop in the price of vxx you increase your position and you still make $100 at the next 10% drop rather than $90, $81… if your position was never changed.

    Starting with $49, how many 10% decreases were needed to get to the split adjusted price of $7.4 ($29 current price)? vxx had to decrease by 10% 18 times to .

    10%*18 = 180% profit for vxx short vs 100% for xiv long.

  42. I was wondering what you think about QUSMA’s hedging strategy (for example, when long XIV short equal amount SPY), especially how it would look with a longer backtest than QUSMA’s.

    • Looking over the spreadsheet there, I’m not sure if the hedged return formulas are correct. I could just be misunderstanding what’s being tested. For an example, assume on a given day that the system is long XIV and XIV is up 5% and SPY is up 2%. The hedged returns on the spreadsheet for that day are calculated as 5% – 2% or 3%. I think it should be .5 * 5% + .5 * -2% or 1.5%. Again, I could just be misinterpreting what’s being tested.

      • Another way to put it is the hedged formulas give a % return on the capital going toward XIV. They don’t give a % return for the sum of capital being used for XIV and SPY.

      • Well, it all depends on your account type (reg T vs portfolio margin) and your other positions. If, for example, you have a long US equities position in your “core” holdings, decreasing it as a hedge vs the XIV strat will only free up capital.

        Similarly, in the case of portfolio margin, where the offsetting short SPY position will decrease your margin requirements. The calculation of expected returns on the capital/margin in use is a case-by-case thing when it comes to this type of hedge.

  43. Thanks, Mike. What you say makes sense.

    I was wondering what your latest thoughts on hedging the Alpha System are. Are you still an advocate of using stops? I ask because most systems I’ve analyzed are, in the long run, hurt by stops. Of course, that doesn’t take into account catastrophic events (but, even in such a case, a stop would be useless if the catacysm were to occur overnight).

    • I don’t have any hard stops in place. Think it’s too risky if things get flashy.

      I have some options data for vxx but haven’t done much with it. It’s a pain to use due to the number of strike/expiration date options. One of these days I’ll get serious about testing hedging strategies via options. Hasn’t happened yet.

      One reason I haven’t put more work into the alpha model is I’m transitioning to other models to trade volatility. Haven’t been content with the drawdown characteristics of this one. I’m still a believer in its long term performance prospects. Just think there are better options out there.

  44. Mike, a suggestion: add the S&P to the core performance chart and the vix level to the alpha performance chart.

  45. Thanks for the reply, Mike. I know that you have some proprietary volatility systems, and I urge to keep them proprietary, but when you say that there are better options to trade volatility, are you referring to any specific systems?

    I ask because there are several respected analysts who are selling access to their systems’ signals, but, as far as I can see, these systems have drawdown characteristics very similar to the ones that characterize your system.

    • I was just referring to my own research. The original backtest for the Alpha model has an average drawdown around 11%. I think it’s possible to cut this by at least 50% without sacrificing returns over the long haul.

  46. All right, you’ve stoked my interest. How can the average drawdown of the Alpha system be reduced by 50%. Clearly, this would be a significant improvement to the system.

  47. Excellent blog and good discussion

    Re. “The key is to systematically increase your short position. Say every 10%, you short more to get back to your original exposure. Otherwise you’re limited to no more than 100% even in the best case scenario.” I hadn’t thought of this, but then I don’t hold forever. I considered it an advantage that the position shrinks, as this reduces the risk, though it can also reduce the return. I don’t worry so much because I also trade a short term strategy that takes advantage of compounding.

    Re. “The truth of the matter is, it’s nearly impossible to not make nice gains in the long run trading XIV systems.” This is true, many approaches work very well. All good things must end sooner or later but I am enjoying the profits while this opportunity lasts.

  48. I know you are not trading this anymore, but the vix spike a few days ago made the vix 10-day std dev go over > 11% for a few days. But then the VXX got absolutely hammered. I wonder if all of this interest in shorting vix futures has changed the dynamic a bit (quick short squeezes).

  49. Mike.
    Have you tested recently Argyros ‘ strategy with XIV/TLT bi montly moving average (August 2012). ”Also, here’s a very simple system that seems to do very well. Buy XIV when it’s above its 2 month monthly average on a monthly basis (i.e. at the close of each month buy XIV if it’s above its 2 month MA). Otherwise, stay in bonds (TLT, for example). ”

  1. Pingback: Hedging VIX ETP Strategies Using SPY : QUSMA

  2. Pingback: Backtest: Don’t Fear the Bear’s StDev Model | Volatility Made Simple

  3. Pingback: Backtest: Don’t Fear the Bear’s Spread Model | Volatility Made Simple

Leave a Reply

Fill in your details below or click an icon to log in:

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

Follow

Get every new post delivered to your Inbox.

Join 174 other followers