Hello!

I got the initial idea for my thesis argument while reading Euan Sinclair's Volatility Trading, where he described a basic strategy of going long (short) VIX front futures when the futures curve is in backwardation (contango) and the expected daily convergence is smaller than -0.1 (greater than 0.1). The reasoning behind that is the lack of predictive power of the basis for the cash VIX, but rather for the VIX futures, which would then converge to the cash VIX level.
I followed it up with the reference, which was Simon and Campasano (2012) (who, I believe, first came up with the idea) paper, and thought of employing copulas to model the dependency between some two futures (or just stick to the spot-futures relationship?) and try to capture some trading opportunities if the curve deviates significantly from the usual shape.
I couldn't really find much in the literature on this particularly. I don't hope to obtain any great results, because it probably would've been already covered if it was exploitable. But still, I found it interesting and am keen to find out what can be done.

My main questions/concerns so far are:

• what two instruments should I use - in the paper the trading was done only on the front month VIX futures, while looking at the relation between the spot and the futures levels and, because of the directional exposure, hedging it with the mini S&P futures. But I thought of modelling the dependency between, say, first two futures and entering offsetting long/short positions, hoping for the curve to return to its usual shape and realizing the roll. Probably without hedging then (?)
• Using VIX levels or returns for modelling the copulas? It seems to not be as straightforward as with equities
• The method used for creating continuous futures price series. And the merit of doing so at all?

I'd gladly welcome any comment, insight, thought, advice.

I don't think the basic strategy is well-defined until you specify a supporting cash level. Even then, there are issues.

For example, suppose you specify that the cash will be rebalanced daily to the level of the VIX index, and offer the resulting portfolio to investors as an ETF. Sounds conservative, right?

How would it have performed historically? If the ETF was, unfortunately, short the VIX front future during the Volpocalypse, (Feb 5, 2018, when the VIX doubled in one day), then the ETF would have been essentially wiped out, and then closed like XIV.

So, what about a cash level of 1.5 x VIX, etc? Obviously, whatever level you choose can potentially lead to the same problem in the future. Of course, there is a risk-return trade-off here. I think you have to resolve this conundrum before proceeding.

My two cents.

I'm afraid I don't completely follow.
This strategy is supposed to only enter trades when it sees a mispricing in the curve (probably it would usually be too steeply upward sloping) and get out when it comes back to the usual state of things. Of course I can't know it in advance, but best case scenario, sticking to the Volpocalypse example, is that it would only enter a trade after the curve gets inverted and bet on its reversal.

In my mind, this strategy, more often than not, doesn't hold any position, although that depends on the confidence level chosen. And even then, it has to be managed, not set-and-forget. Therefore, I didn't think of it as an ETF candidate.

Also, because the trade, if entered, is long-short, some of the loss, in case of VIX spike, would be offset by the long position. Granted, on the 5th Feb 18' it would result in a big loss anyway. I used a chart to illustrate the evolution of the curve over these few days: https://ibb.co/YRQSkdJ.

OK, fair enough -- let me rephrase. 

I took a look at Simon & Campasano

They create a passive strategy with the rationale you indicated and present some performance statistics, which argue that their basic strategy has a positive expectation. So, one thing worth doing would be to see if you can replicate their results through 2011, which is the end of their data. Then, in light of the Volpocalypse, what happens to the statistics given if the study is extended through, say June 2023?

Now back to the ETF notion. The reason I raised the ETF business is that, even if you have a strategy with a positive expectation, it can be a tricky issue incorporating such a strategy into a portfolio. Especially with derivatives. For example, buying call options at (for definiteness) Black-Scholes value has a positive expectation, but a portfolio consisting of only that will go bust rapidly. 

The simplest incorporation (of just about anything) is to imagine a money market portfolio and allocating a (rebalanced) percentage of wealth to the strategy in question. But maybe the natural home for this VIX futures strategy is an equity portfolio. I don't know. But in any event, I didn't see Simon & Campasano address this portfolio question. (They do present a 'cumulative profits' chart but that is not the same). I think it's an important issue. Also, I still think you have to be careful to distinguish historical performance from potential, but not seen yet risks, like VIX not just doubling, but moving even more than that in one day. 

Okay, I see where you're coming from, you made me realize that I need to take a broader perspective into consideration.

I'll definitely check the results of Simon & Campasano from 2011 onwards.

The only mention of actual application of such a strategy I found in Chrilly Donninger's http://www.godotfinance.com/pdf/VIXFuturesBasisTrading_2_Rev1.pdf - 'The Calvados is traded in the Sybil-Fund. It is so far the pick of the bunch. One gets a lot of fun for a medium dose of risk.'

I realized the trickiness of the dissonance between historical and potential returns, but it just seems that there is no true remedy for this problem, is there? It can be tested under various scenarios, but ultimately we can't know what is going to happen. That was my thinking at least.

Thanks for the involvement!

You're welcome. 

Certainly, we can't know precisely what might happen, but there are strategies for simulations of the future that give insights on new possibilities. I am especially fond of the "bootstrap".

For example, I raised the issue of extreme moves in the VIX. Note that the VIX has been calculated back to Jan 2, 1986 (using the original methodology): VXO history . 

That's means you have data showing the VIX during the Oct 19, 1987 crash. So, you can assemble pairs of historical (SPX,VIX) returns from 1986-date. Then, to simulate the future, do bootstrap sampling: make IID draws of paired returns from this historical distribution. (Millions!) 

These draws will include the possibility, for example, of two successive crash days, or worse. So, this is a way of seeing some `worst case' events -- at least worse than has been seen so far -- and perhaps some plausible associated frequencies..