Easy Volatility Investing (paper / trading strategies)

[Marco]

for those who trade VIX - be it though futures, options or ETP's (VXX,XIV,ZIV and the like).

I came across a paper that I found very interesting - well I only glanced over it for now and read a few things in more detail, but it looks very interesting so far.

Trading VIX related strategies is slightly more complicated in my opinion that 'normal' option strategies, so if you are new to options and still get your head around delta,gamma,theta keep this for later and certainly venture into this area slowly and carefully. Having said this I think this is written not overly complicated and you don't need a degree in maths to understand it.

 

It also offers a number (5 to be precise) of actual trading strategies and compares them - so it's not only about theory.

 

At the first glance I particularly like these two (for simplicity and results):

 

 

#3-Roll Yield:

if the 10 day moving average of VXV/VIX > 1 go long XIV else go long VXX

 

#4-VRP: (volatility risk premium)

if the 5 day moving average of (VIX - 10 day historical volatility) > 0 go long XIV else go long VXX

 

I have heard of the first one before here (note this guy has the ratio the other way around). The results looks pretty good however I would caution that these reverse ETN's havent been around for too long yet (~2.5-3 yrs - so after the 2008 financial crisis) and trend in VIX was down over that time frame so a simply buy and hold would have been a pretty good investment as well. So the backtesting has to be seen in that context.

 

A quick backtest since 30-Nov-10 (when XIV was listed) and I start with a long position in XIV on both:

 

roll yield: 7(!) trades since inception, total (non compounded return (just adding % returns so +12%, +13%, -5% = +20%) = 296% (the strategy is long XIV since 12-Oct-11 without any trades since)

 

VRP: 23 trades since inception, total return 153%

 

buy and hold XIV since inception (30-Nov-10) 146%

 

NB: I'm counting a flip from one ETN into the other as one trade. Backtesting has been done with EOD prices.

I urge you to understand how VXX and XIV work and the risks involved before you actually try and trade this. Also read the paper (they mark the bits about risk with a 'grim reaper' )

 

okay here the paper and the link to the back testing sheet (let me know if you come across any errors).

Marco.

 

00R_Easy Volatility Investing + Abstract - Tony Cooper.pdf

https://dl.dropboxusercontent.com/u/26062189/XIV_VXX_strategies.xlsx

 

 

[PaulCao]

Hi Marcos,

 

If you are interested in more VIX related strategies, there's a bunch of other VIX related strategies with backtest results here: 

http://godotfinance.com/workingpapers/

 

One idea I'm exploring is trading VXX option against VIX option: 

 

VIX option on VIX future has an unusual nature that as the option approaches settlement date, IV spikes more and more. This is because VIX is an mathematical formula with lots of jump risks. So the forecasting spot isn't a smooth. 

 

However VXX option, as a constant rolling 30-day out VIX future behave as a normal option. 

 

Case in point the current expiring VIX ATM May future-option has an IV of 88.75 (compared to VIX ATM June future-option IV of 79.86) whereas the VXX ATM May option has an IV of 33. 

 

What's more interesting is that VIX future-option has an settlement date two days after the regular option settlement date, May 22nd vs. May 17th; giving the future-option 2 more days of time value. 

 

Given that VXX has a typical correlation ratio of 0.5 to the VIX spot price. I wonder how effective it'd be to sell VXX option against the equivalent dollar weight of VIX future-option to exploit the rising IV of VIX future-option in the waning week prior to settlement, with the delta being hedged with VXX. 

 

Best,

PC

[Kim]

Marco, thanks for sharing.

 

I'm wondering why you get much higher returns on the roll yield (almost double in fact) while they have much higher returns for VRP.

[Marco]

Marco, thanks for sharing.

 

I'm wondering why you get much higher returns on the roll yield (almost double in fact) while they have much higher returns for VRP.

I only backtest since Nov-10 (when XIV was listed) they calculated 'synthetic' prices for XIV and VXX going back to 2004 using VIX future prices. So they cover the 2008 finincial crises and the aftermath. I didn't have the time and the raw data for VIX futures going back to 2004 to do that. But yes VRP seems to do much better over the whole period - that's why I picked it next to roll yield to check it out myself.

[Marco]

Hi Marcos,

If you are interested in more VIX related strategies, there's a bunch of other VIX related strategies with backtest results here:

http://godotfinance.com/workingpapers/

One idea I'm exploring is trading VXX option against VIX option:

VIX option on VIX future has an unusual nature that as the option approaches settlement date, IV spikes more and more. This is because VIX is an mathematical formula with lots of jump risks. So the forecasting spot isn't a smooth.

However VXX option, as a constant rolling 30-day out VIX future behave as a normal option.

Case in point the current expiring VIX ATM May future-option has an IV of 88.75 (compared to VIX ATM June future-option IV of 79.86) whereas the VXX ATM May option has an IV of 33.

What's more interesting is that VIX future-option has an settlement date two days after the regular option settlement date, May 22nd vs. May 17th; giving the future-option 2 more days of time value.

Given that VXX has a typical correlation ratio of 0.5 to the VIX spot price. I wonder how effective it'd be to sell VXX option against the equivalent dollar weight of VIX future-option to exploit the rising IV of VIX future-option in the waning week prior to settlement, with the delta being hedged with VXX.

Best,

PC

Thanks for the link. Looks interesting.

How is that VIX vs. VXX trade going for you? I told you what problem I see with it in the original thread on this (http://steadyoptions.com/forum/topic/1121-vxx-and-vix-trading/)

Edited May 15, 2013 by Marco

[Guest Peticolas]

One problem I see is that you compare compounded returns with arithmetic summation of returns. So you say buying and hold XIV gives you a 146% return, but that's really a compounded return. When I start with a hypothetical $100,000 portfolio, and compute gains and losses on the whole amount, your comparison is not 153% vs 146% for VRP vs buy and hold. It is 387% vs 146% when both are computed using compounded returns.

 

The other thing I don't quite understand about your spreadsheet is exactly what constitutes a buy/sell signal. Based on the paper, I would have thought it was crossing the zero line. You seem to anticipate that cross in your signals. If it were the zero line, and because these are end-of-day values, your first opportunity to trade would be the day after the signal changed signs. 

 

I'm actually uncomfortable trading this sort of thing unless it has been tested on out-of-sample data. Buying and holding the xiv is certainly viable but adding xiv to a stock portfolio hurts the risk characteristics of that portfolio, although it may boost returns  Maybe ditching the stock portfolio and holding only xiv is the answer.

Edited July 8, 2013 by Peticolas

[Marco]

 

 

One problem I see is that you compare compounded returns with arithmetic summation of returns. So you say buying and hold XIV gives you a 146% return, but that's really a compounded return. When I start with a hypothetical $100,000 portfolio, and compute gains and losses on the whole amount, your comparison is not 153% vs 146% for VRP vs buy and hold. It is 387% vs 146% when both are computed using compounded returns.

agreed, at the time I wanted to have a quick look at whether the method works at all so just adding returns was the quicker way, compounding gives you a better comparison. 

 

 

 

The other thing I don't quite understand about your spreadsheet is exactly what constitutes a buy/sell signal. Based on the paper, I would have thought it was crossing the zero line. You seem to anticipate that cross in your signals. If it were the zero line, and because these are end-of-day values, your first opportunity to trade would be the day after the signal changed signs. 

there were some mistakes for the VRP strategy on the sheet which I discovered a bit later, makes no's a bit worse for the VRP strategy (I think it was around 10% points or so less of the non compounded returns) but doesn't change much of the ball park no's and as the discussion here didn't suggest too much interest I didn't bother posting a new sheet. 

 

 

 

I'm actually uncomfortable trading this sort of thing unless it has been tested on out-of-sample data. Buying and holding the xiv is certainly viable but adding xiv to a stock portfolio hurts the risk characteristics of that portfolio, although it may boost returns  Maybe ditching the stock portfolio and holding only xiv is the answer.

 

well as I said this was a quick and ready attempt to have a look at the strategy that the paper presented, So I also only went back until the time XIV was listed which gives you limited data. The paper calculates XIV and VXX back a few years longer with VIX futures data so you get a bit better sample.

I'm still following both signals and one thing I found about the VRP strategy is that after you have a few days of big moves and HV10 and VIX both go up but you still have a positive spread of VIX - HV10. However when markets calm down and VIX drops quickly (like it happened recently) but HV10 is still elevated the strategy generates a sell signal (=buy VXX) into a dropping VIX and now when HV is catching up with VIX and the spread returns to a positive number you sell VXX and buy back XIV at a loss.

 

update: okay I fixed the above mentioned error and added compounded returns to the spreadsheet (columns T and Z - only updates when there is a new buy or sell to keep the formula simple) https://dl.dropboxusercontent.com/u/26062189/XIV_VXX_strategies.xlsx (same link as before)

 

So with that I get to

VRP 82% compounded return

roll yield 532% return and 

XIV buy and hold 136% return since  Nov'10

 

that spectecular return and also spectecular outperformance of roll yield vs. buy and hold is mainly due to a switch from XIV to VXX between Aug11 and Sep11 where XIV dropped 55% and VXX rose 78%. Without that trade the performance would be much closer to buy and hold XIV as thats what roll yield has been doing most of the time (only 6 switches in the whole period and long XIV since Oct'11 now) Thats one thing I don't like about compounding you get to very high numbers quickly and while the reflect real returns if you really had reinvested all your gains in the trade all the time I find them somewhat misleading...

 

 

Peticolas I'm not sure how you got to 387% compounded return for VRP as of mid may though. I have 100$ turned into ~270$ by then - thats 170% return for me (so are you adding 1 where I'm subtracting 1?)

Edited July 8, 2013 by Marco

[PaulCao]

Hi Guys, 

 

For those who are interested, there's an actually a following of the backtest of the VRP strategy here, with some responses from the author of the original paper, 

 

https://www.quantopian.com/posts/system-based-on-easy-volatility-investing-by-tony-cooper-at-double-digit-numerics

 

Also an implementation that hedges VXX/XIV with SPY since there's correlation there, 

 

Best,

PC

[Marco]

Hi Guys, 

 

For those who are interested, there's an actually a following of the backtest of the VRP strategy here, with some responses from the author of the original paper, 

 

https://www.quantopian.com/posts/system-based-on-easy-volatility-investing-by-tony-cooper-at-double-digit-numerics

 

Also an implementation that hedges VXX/XIV with SPY since there's correlation there, 

 

Best,

PC

thanks for the link! Keep us posted if you see more in that direction and/or from the author of the paper.

thanks.

[Guest Peticolas]

Thanks Marco. I do think it's hugely worthwhile to this sort of study. It was years before economists figured out that Reinhart & Rogoff made some big errors in their spreadsheet supporting their paper on austerity. It's always best to check.

 

I like the new version. Regarding my calculation on your old sheet, I didn't add one instead of subtract one, but I certainly could have made a mistake. 

 

I do think about how you would actually trade this, and I think you're still pulling the trigger one day early. Look at 8-11-11. Your VRP signal flips to negative and you switch etfs; however, that's an end-of-day signal. I was thinking unless you're trading after-hours, your first opportunity to act would be the next day.

 

I think we're on the same page regarding roll yield. Although your result looks fabulous, there are really too few signals to feel very comfortable about it. I do like working on this idea though. There just aren't many instruments so sure to go up over time as xiv (and conversely with vxx).  I'll look into the link PaulCao provided when I have some time.  

 

[cwelsh]

Printing out the paper now -- I would like to note that the paper,and some of the commentators on the site mention Zhang's work on options -- I'd like to recommend those, I've read several of their papers.

[Marco]

 

Thanks Marco. I do think it's hugely worthwhile to this sort of study. It was years before economists figured out that Reinhart & Rogoff made some big errors in their spreadsheet supporting their paper on austerity. It's always best to check.

 

I like the new version. Regarding my calculation on your old sheet, I didn't add one instead of subtract one, but I certainly could have made a mistake. 

 

I do think about how you would actually trade this, and I think you're still pulling the trigger one day early. Look at 8-11-11. Your VRP signal flips to negative and you switch etfs; however, that's an end-of-day signal. I was thinking unless you're trading after-hours, your first opportunity to act would be the next day.

 

I think we're on the same page regarding roll yield. Although your result looks fabulous, there are really too few signals to feel very comfortable about it. I do like working on this idea though. There just aren't many instruments so sure to go up over time as xiv (and conversely with vxx).  I'll look into the link PaulCao provided when I have some time.  

 

 

I think you could trade near the end of the day when you have a signal with HV10/VIX/VXV at current levels but even if you wait until the next day most of the time it shouldn't make a huge difference

[Guest Peticolas]

I think you could trade near the end of the day when you have a signal with HV10/VIX/VXV at current levels but even if you wait until the next day most of the time it shouldn't make a huge difference

 

If you could trade near the end of the day, it might work. Although the next day would not be significant for most prices we might want to predict, the variability of VXX and XIV are so large that the next day price could easily differ by 10%. I really liked the analogy in the paper, saying that investing in volatility is like "picking up $100 bills in front of a steamroller." That was exactly my reaction to looking through your spreadsheet. Even with these timing mechanisms, there are some scary drawdowns. A couple of  other comments on the paper: 

 

Looking at figure 9 on page 14: although he didn’t include AGG in the optimization, it looks as though you can achieve almost any level of return on the efficient frontier with just AGG and XIV. Buy and hold (as a part of an overall portfolio) may be a pretty viable strategy.

 

The risks section was very good, and makes it clear why we should only consider any of these in the context of a larger portfolio. Although they framed timing synchronization as a roll-yield issue, I thought it is particularly an issue for VRP (at least since the advent of XIV, because VRP generates so many signals, and those signals sometimes last only a few days – this despite all the smoothing).  The regime change risk is a problem for all the strategies, and I think regime change risk is considerable.

 

On the plus side, you're not trying to predict prices in a very fine-grained fashion. You're just trying to stay out of the way of catastrophic collapse.

 

If you have a background in statistics, don't pass up technical appendix A -- Very interesting.

Edited July 11, 2013 by Peticolas

[PaulCao]

Hi Guys, 

 

I went to an algorithmic trading meetup hosted by Quantopian, the trading backtesting startup that I referenced a few posts ago, 

 

I told the host that their next presentation topic should be on the "VXX VRP trading strategy," and the founders encouraged me to present it. 

 

So I plan to use their backtester to try to improve some of the existing strategy and present the results, my ideas include: 

 

Concrete idea: 

a) implement the "Mojito 2.0" strategy; http://www.godotfinance.com/workingpapers/DynamicVIXFuturesVersion2.xhtml

Which uses IVTS = VIX/VXV  (relation between 1-month VIX and 3-month VXV index); and based on what level IVTS is, another measure of VRP; trade a ratio VXX and VXZ, especially a calendar spread of the short and long-term VIX tracking ETN. 

 

Hedging the VXX/SPY hedging strategy; as mentioned before, there is already one existing algorithm, I want to try this method also from Donniger et al. http://www.godotfinance.com/workingpapers/DynamicVIXFuturesVersion2.xhtml; which again based on their metric of IVTS, trade a spread of VXX and SPY. 

 

c) Short VXX and long XIV only when the VIX term structure is in backwardation, this happened briefly in the recent VIX spike and then subsequent decline, I'm curious if this is a viable trading signal. 

 

Vague idea:

d) Use Google trends on words "fear" and "debt" to trade on VXX, this seems very far fetched: however, there was a Nature paper that traded the ETFs to Google trends apparently with great results: https://www.quantopian.com/posts/google-search-terms-predict-market-movements

 

I'll post back whenever I worked through one implementation. I suspect most of the ideas would end up mediocre or not changing much the status quo performance, but it'll be an exercise/excuse for me to follow through on implementing these ideas, 

 

Best, 

PC

[cwelsh]

Hi Guys, 

 

I went to an algorithmic trading meetup hosted by Quantopian, the trading backtesting startup that I referenced a few posts ago, 

 

I told the host that their next presentation topic should be on the "VXX VRP trading strategy," and the founders encouraged me to present it. 

 

So I plan to use their backtester to try to improve some of the existing strategy and present the results, my ideas include: 

 

Concrete idea: 

a) implement the "Mojito 2.0" strategy; http://www.godotfinance.com/workingpapers/DynamicVIXFuturesVersion2.xhtml

Which uses IVTS = VIX/VXV  (relation between 1-month VIX and 3-month VXV index); and based on what level IVTS is, another measure of VRP; trade a ratio VXX and VXZ, especially a calendar spread of the short and long-term VIX tracking ETN. 

 

Hedging the VXX/SPY hedging strategy; as mentioned before, there is already one existing algorithm, I want to try this method also from Donniger et al. http://www.godotfinance.com/workingpapers/DynamicVIXFuturesVersion2.xhtml; which again based on their metric of IVTS, trade a spread of VXX and SPY. 

 

c) Short VXX and long XIV only when the VIX term structure is in backwardation, this happened briefly in the recent VIX spike and then subsequent decline, I'm curious if this is a viable trading signal. 

 

Vague idea:

d) Use Google trends on words "fear" and "debt" to trade on VXX, this seems very far fetched: however, there was a Nature paper that traded the ETFs to Google trends apparently with great results: https://www.quantopian.com/posts/google-search-terms-predict-market-movements

 

I'll post back whenever I worked through one implementation. I suspect most of the ideas would end up mediocre or not changing much the status quo performance, but it'll be an exercise/excuse for me to follow through on implementing these ideas, 

 

Best, 

PC

There are already numerous firms do exactly that (following this idea has been a hobby of mine for a while).

 

For articles on it see:

 

http://online.wsj.com/article/SB10001424052748704588404575123901508940726.html

http://www.prweb.com/releases/Social-Trading-Forex/CaesarTrade/prweb9957004.htm

 

For actual entities that have implemented this see:

AlphaFlashTrader

NewsTechnologies, LLC (who even has a few patents on it)

Bittext

 

There's even a new hedge fund out that solely trades on twitter analysis.