Pricing the Term Structure of VIX Futures
Vol is back with a vengeance. Policy makers don't have the weapons to deal with the shock unleashed by the Corona pandemic. The reason is that all the weapons we do have are either geared to fight demand shortfalls or inflation. We got nothing to fight the massive supply-side disruptions that have been unleashed by the most serious pandemic since the 1918 Spanish Flu. The world economy is already in recession. The markets know. Whence the unrestrained risk-off across asset classes.
VIX is testing the all time high achieved during the western financial crisis. The VIX is truly Wall St's fear gauge. The reason is that dealer banks (Wall St sensu stricto) use it as a proxy to calculate their value-at-risk. When volatility spikes, they must shed risk assets to maintain a constant probability of default. Conversely, when they shed risk off their books, vol spikes.
I figured out a simple strategy to trade volatility. In previous work, I showed that it was possible to achieve abnormal risk-adjusted returns if you have a good model of risk-offs. In that work, we traded the term structure of volatility using ETFs. I had a suspicion that working directly with VIX futures would be even more profitable. That turns out to be an understatement.
The market has returned about 6 percent over the past decade before the present mayhem. I showed how to achieve more than 16 percent at market levels of volatility. Since then the market has cratered. As of yesterday, the average returns on the S&P 500 since 2013 has been 3.94 percent. In what follows, Of course, one can achieve arbitrarily higher returns by piling on risk. The real issue is risk-adjusted returns. In what follows, I will exhibit a strategy to trade the term structure of volatility that has yielded, as of yesterday, 24.08 percent at market levels of volatility.
The basic insight is that dealers live at the short end of the vol term structure. The controlling eigenvector is the slope of the vol term structure. The term structure is usually upward sloping.
During risk-offs, the term structure inverts. This is because dealers live at the short end of the curve.
Over the past three weeks, since all hell broke loose, the vol term structure has been inverted on three days out of four. [We are working in returns space; not prices.] Yesterday, for instance, the contracts expiring soon sported a 30 percent return; while those with longer tenor/maturity yielded half as much.
Since they are marginal investors in your typical volatility trade, the response function of the dealers dictates the term structure of VIX futures. During a risk-on, when dealer balance sheet capacity is plentiful, the term structure is upward sloping ('contango'). Conversely, during a risk-off, when dealers are shedding risk off their books, the term structure is downward sloping ('backwardation'). If you can predict risk-offs, you can thus make a shit load of money by trading the term structure of volatility. This is what we propose to do.
I found a feature that predicts risk-offs out-of-sample. More precisely, I have isolated a signal that predicts a risk-off tomorrow based on information available today. [If you want to know the feature, I'll be happy to tell you in exchange for a cut of your net profits on the trade.] We train the model on the data available from the CBOE. Astonishingly, the signal predicted the big risk-offs on Vol Monday, and China-related vol in 2015, the Fed-induced risk-offs of 2014, and, of course, the recent mayhem. The signal is very kosher — we have good confidence that we have isolated the signal for dealer risk appetite in real time.
We trade the vol term structure as follows. On any given day, we sort the vix futures contracts by tenor and place them in three buckets: Low, Mid, High. Then, if the probability of a risk-off is lower than a certain threshold (a hyperparameter optimized out-of-sample) we sell the Low portfolio and hedge that by going long on the High one. If, on the other hand, the probability of risk-off is higher than the threshold, we go long the Low portfolio of vix futures and hedge that by shorting the High portfolio. So our portfolio is High-Low when we predict a risk-on, and Low-High when we predict a risk-off.
Selling vol is just a highly leveraged long position on the market and High-Low is a hedged version of selling vol. We can illustrate our strategy by comparing its returns to the returns on the S&P 500.
Basically, we make a lot of money on the days we can predict a risk-off. So the secret sauce of our feature is really all there is to it. But this is no small matter. The cumulative returns dwarf those on the market. If you had invested $1 on Jan 1, 2014 in the market, you would have $1.16 today; if you'd invested it in selling vol, you'd have $2.73; if you'd used the hedged selling strategy (Low-High), you'd have $4.09. If, on the other hand, you'd followed the Policy Tensor's strategy, you'd have $34.49. Of course, our present circumstances are particularly conducive to volatility trades that work. But that's the point. It is, in fact, possible to predict risk-offs the day before. Not with certainty. But given the footprint of the dealers in financial markets, with just enough certainty to make it extremely lucrative — especially in highly volatile markets.
The recent performance of the systematic trading strategy is interesting to investigate. What did our algo do since the present convulsion began? We can see that we were selling vol until Feb 24. More precisely, we shorted short-term vix futures and went long on long-term vix futures. Since Feb 25, we have shorted long-term futures and been long short-term futures.
In retrospect, it looks like an easy call. But hindsight is twenty-twenty. The beauty of systematic trading is that one does not put one's hand inside the machine to change anything. Our systematic strategy worked before the mayhem began. It has worked beautifully during the cataclysm. And it'll continue to work as long as the response functions of the dealer remain the same.