Robin Hewitt

Long only, contrarian, special situations
Robin Hewitt
Long only, contrarian, special situations
Contributor since: 2016
"your repeated claims daily changes are not the source of value".
Where do you see that claim, Silent Trader? Because I'm certainly not saying that. Quite the opposite. What I have said, and I think this may be a source of confusion forever, is that contango is a lagging factor. It does not lead; it merely reports. There's no intrinsic value in the price difference between M1 and M2 per se. That's not creating a tailwind that makes it easier or harder for value to change in a wished-for direction each day. It's only reporting what the market's valuations were, and it can only report that AFTER the market decides. The gain or loss for the day came from the market's changed valuations...and while these may show up as contango, they weren't in any way caused by contango.
If anything, extremely high contango or backwardation are unlikely to persist and are possibly signals that we may have reached a sentiment extreme that's likely to reverse soon.
That's exactly right, Silent Trader. And well put. And yes, it tracks a theoretical index representing a constant one-month synthetic future.
Great question, JimiCarr!
The change in price that occurred during the day gets accounted for each night via the settlement, which is not included here because my claim is that this settlement process is the only way value is added or lost. Merely rolling contracts does not create or destroy value. Regardless how much price difference there is between contract months, all the nightly roll does is shuffle funds around.
That change during the day is set by the market -- specifically, by the market's estimation of the future value of VIX. We don't know ahead of time what the market will decide on any given day. If we end up with M2 higher than M1, we've ended up in contango. But that contango is the result, not the cause, of what the market decided. If contango by itself were adding a "contango power" tailwind or headwind when rolling, we shouldn't have to also include the market's decisions to see that. It should show up somehow just by rebalancing the contract weights.
It seems to me there's confusion between cause and effect in the discussions about a contango tailwind, and that gets expressed as there being a roll yield, a benefit or detriment that happens just from rolling contracts.
It can't be a tailwind if it's an effect rather than a cause. And that's what I'm trying to show. In practice, we won't even know the contango value until after the market tells us what that day's price changes are. Once we realize that, it becomes clear that contango is a lagging, not a leading, factor. It doesn't influence; it only reports.
"You really believe that intraday price changes of the futures have no influence on the value of the portfolio?"
On the contrary, that's the entire source of the value, as is clearly stated in the article. Using that as a strawman for a lengthy attack suggests that you either misread the article or read something into it that simply isn't there.
Thanks, Dane Van Domelen! Yes, I expected this would be controverisal :-)
Thanks, VIX Strategies, and thanks for adding the link as well.
Christopher Lee, That's the real question, isn't it! I'd like to address that in future articles. This one was already quite long.
There are some well-known trading strategies based on the ratio between the VIX and VXV (CBOE's 3-month volatility index) that backtested very well when they were initially published, but have done terribly in recent times. Pretty much every trading strategy has a limited lifespan, but these soured so quickly that I think there's a lesson there....
If you look through the series, you'll see that cash does not steadily decline. It goes up, then comes back down. The total account value is the sum of the cash value plus the value of the M1 and M2 positions. Cash fluctuates (it doesn't just decline) because the weight balances constrain how many non-fractional contracts can be acquired at each rebalancing.
The way I did the rebalancing was to liquidate all contracts into cash, then debit that cash by converting it into new contracts each trade day. There is a moment when the cash does equal the entire account, but that's not what's printed. The printout shows the next iteration's contract counts, their prices, and the residual cash that remained after buying the new round of contracts.
"Using her futures quotes, I calculate a 95% loss in VXX"
Could you let me know more precisely what you calculated? I'd like to replicate whatever it was you did.
"The -54.97% is the average loss per YEAR."
Eli, I think you're right. The index shows a loss around 96% since inception, so they must be reporting a yearly average. Good catch.
"Why do you include cash in your scenario?"
The cash bucket captures the residual that doesn't divide evenly into whole numbers of contracts during rebalancing. That's also where the daily settlements are reconciled. You need to maintain the cash bucket to balance the account. Otherwise you're not tracking all the funds. If you look at one of these ETN's portfolio on any given day, it pretty much always includes a fair bit of cash. Some of it may be serving as collateral...I've never seen that level of granularity reported...but it also holds residuals from rebalancing.
"Everyday m1 contracts are bought and m2 contracts are sold or vice versa. The difference in price is the roll yield."
Silent Trader, I believe you're thinking of these ETNs as holding a constant number of contracts that are rolled each day. That's not how they work.
The total number of contracts varies. They close out a dollar amount of M1 and use that dollar amount to buy an equal dollar amount of M2. Any residual that doesn't divide evenly into a whole number of contracts according to that day's weighting goes into a cash bucket. There's no change in the total value from doing that. It's just moving funds around.
Ignoring things like interest on cash and fees, the only time the value of the ETN changes is during settlement. There's no other way it can change.
On days when XIV goes up in value, that happens during settlement. That money goes into the ETN's cash bucket. Now the total value of the fund has increased -- but no contracts are rolled. The increase is all in cash.
After that change to cash, then it recreates the one-month position by shuffling funds around without changing its total value. A bit more to M2, a bit less to M1, and the rest as cash.
Eli, only if the account balance falls. The balance is the sum of the M1 contracts, M2 contracts, and the cash bucket. You need to update the cash bucket as well as the contract counts each night.
Daniel9999, Since these are futures, it's not exactly a short-and-cover, but the idea is similar. XIV is the seller of future VIX in every contract it holds -- both in M1 and in M2. In the nightly settle, it collects or loses that day's difference since the prior settlement, then redistributes that amount between M1 and M2 contracts. The number of total contracts changes with each rebalancing, so it's not a direct transfer of contracts from M1 to M2; it's a shift in the percentage held -- a little higher percent of M2, and a little lower percent of M1 each trading day.
Here's a thought experiment for those who are still skeptical.
Imagine the "constant contango" scenario and compare the start point of all contracts being in M1, none in M2. M1 has price P, M2 has price (1+c)P. Gradually, some of the weight shifts to M2 each day, while M2 also slides down toward where M1 started. Immediately after M1's expiry, we know two things:
1. All contracts are now in the former M2
2. The former M2 now has price P that the former M1 had when we started
start = finish = P * all contracts.
You end up exactly where you started...same price at the start as at the end, with all the weight in the new M1.
"The XIV certainly makes it money on roll yield(assuming contango)"
Hi herkfsu, Actually, I just posted a worked simulation (with source code) that demonstrates it does not: 50% contango, no change in value. Here's the source code and its output: I was surprised by this myself, but it turns out the nightly rebalancing simply reverses the effects of M1 and M2 declining in value when you apply this constraint.
"1x derivative to rise 800%" You probably want to look at the geometric product rather than just ΔP/Δt. It's not a linear relationship.
That's an interesting question on VXX. Obviously it's been in existence longer than XIV, but even compensating for that, yahoo's price history still shows it as losing far more percentage-wise than XIV has gained. However, if you go to the iPath homepage for VXX, there's a "snaphot" link. As of right now, the most recent snapshot from iPath was on 1/31/2016, and it says the change in VXX's IID since inception is -54.97%. Going by that (which should be the best source) it looks like the performance of VXX is actually pretty close to being the inverse of XIV. I'd have to dig to find out more.
Here's the homepage for VXX:
And here's the VXX snapshot:
Thanks, Gregg. I think what you say makes sense. It will be interesting to do a retrospective...!
Hi Eli, It's not in eq 8 because you cancelled it from both numerator and denominator. However, that was not valid. Had you correctly used dr-1 for time t, you would not have been able to cancel. It's a misrepresentation error. I hope the worked simulation, using actual values, that I posted here: will help show that contango does not affect the value of XIV.
I realized after posting that I'd calculated for VXX, rather than XIV. I don't feel like editing to reflect that, so I'm just noting it here. The important thing is that there's no change to the initial value. It was helpful for me to see how the nightly rebalancing compensates under these constraints, so thanks to Eli for suggesting it.
Hi Eli, I don't need to have the last word. Please feel free to comment further. I did as you suggested and calculated one month of valuations for M1 and M2 sliding down an unchanged contango slope.
Result: No change. Start Value = Ending Value.
Full calculations, with all interim values and the final, unchanged, valuation plus the source code that did the calculations are in my instablog post titled, Worked Example Of XIV For One Month Of "Constant Contango", which you can find here:
I hope this helps! It was a good exercise for me as well. I'd initially thought the account value would change, but then I wondered if perhaps the daily rebalancing would reverse that...and it did.
"Yeah, the shape of the curve is not the critical factor"
Agreed. I got so tired of hearing "it's the contango", and "it's the roll yield", that I just did an article about that -- about how that's incorrect. Even the most minimal assumptions about it being a headwind or tailwind are not correct...and the roll yield concept is a myth. There is no roll yield.
That trade worked well today.
Agree that Tuesday will be interesting!
"If vol spikes, you will lose 100% of your investment"
I think that should be "may lose". The ETN will be extinguished if its account reaches a low enough level that it can no longer maintain its synthetic future.
I'm not seeing how can dr be for the same day, when you define t like this:
"Lower case "t" stands for the current trading day, "t-1" stands for the previous trading day."
If t changes dr, which is also a day counter, changes....
So you're saying M1 and M2 will decrease in value, right? Anytime M1 and M2 both decrease in value, XIV should go up. But that can happen (and frequently does happen) during backwardation as well as during contango, so it's not the c term that's changing XIV, but rather dM1 and dM2. Look, for example, at the term structure on 12/14&15/2015. On both days, it was in backwardation (M1>M2, the definition of c in the equations). XIV went up in value from 23.13 to 25.40, even though c was negative.
Basically what you'd be doing in the case you're describing is setting up a situation where M1 and M2 go down, and additionally that situation just happens to be a contango situation. But you can completely remove (or change) the contango assumption and get the same result because all that really matters is that both dM1/dt and dM2/dt were negative. This is what often happens in backwardtaion -- the entire curve falls down a little, but remains in backwardation.
"Since in equation 1 dr/dt is in both numerator and denominator"
Because the days remaining goes down by one each trading day, dr in the numerator = dr-1 in the denominator. They're not the same value. This actually starts earlier, in eq 1.
One way to handle the changing dr would be to pick one as the base, then calculate the other by adding or subtracting 1. For example, if we take dr at t-1 as the base in eq 1, the denominator would stay the same, but all occurrences of dr in the numerator should be replaced by dr-1.
To calculate the t+1 value of the index, you'll need to also establish how much M1 and M2 change from settle-to-settle, in other words, the vertical shift of your term structure. There are two degrees of freedom here -- a slope, which you're holding constant, and an intercept, which may change from t to t+1. You can't calculate the next value unless you have both.
Jay, do you mean this: "XIV gains because the futures are one day closer to expiration and their premium has gone down" ?
Note that, although M1 and M2 are both one day closer to expiry, the synthetic future that XIV tracks is no closer to expiry than it was the day before.... Its risk premium is still for one month.
Eili, I'm looking at eqs 5 and 6, where dr/dt is in both numerator and denominator, but dr is not the same in each. Using latex notation,
r_(t-1) = 1+r_t
I believe you need to either subscript dr or express it recursively.
Thanks for the careful read, MattRadke! The roll is a bit tricky because the total number of contracts is not fixed. It can change each day. The index it tracks seeks to maintain a notional value each day that's equivalent to the one-month position. Easiest way to think of that is to imagine all the contracts settling to cash each night, then using that amount to buy new contracts according to the next day's weightings. If you walk through an example of doing that, I think you'll find that the total number of contracts has to change to maintain the notional value of a one-month position.
Hi Eli and thanks for your comment.
I believe I see a problem in the derivation. In those equations dr/dt represents the M1 weighting. But...the M1 weighting for day=t will be less than the M1 weighting on day=(t-1). That means the dr/dt's don't just cancel out, which would change your result.
Thanks for elaborating, ICCS.
So...I do think anytime anything (such as XIV) does well, it can become a crowded trade. You probably don't even need a model to say that much.
But it sounds like you're saying something different, namely, that a market that used to be rational has become (permanently?) distorted because people have been trained to always short volatility no matter what. (That last part comes from the Prisoners' Dilemma reference.) Did I understand you correctly?
Thanks, Nathan. Reason I think it's too soon to go short is the same one I mentioned previously...we're in an environment of rising volatility. Today may have been a good day for a quick short, but it's a risky time for it. I used to do a lot of quick trades, but I found they were less profitable than staying with a direction through short reversals.
I'm still waiting. I did get a signal to go long end of Friday, but by Monday UVXY had popped so much that I sat it out. Will re-evaluate next week.