Corona-Vol: How Did We Do?
The Standard & Poor's index of market-cap weighted 500 largest US stocks reached an all-time high of 3386 exactly 10 days ago. Since then it has fallen 12.8 percent as panic about the impact of the Corona pandemic has escalated. Volatility has returned with a vengeance. Wall St's "fear gauge", the VIX, has climbed from 14.4 at close of day on Wed, Feb 19, to 40.1 at close of day on Friday, Feb 28. FT reported that the VIX has posted the biggest weekly rise since the depths of the global financial crisis. This is a good time to revisit the performance of our volatility trading strategies.
The data I analyzed in December ended on November 27, 2019. Since then it has been exactly three months. So the timing is doubly fortunate. In what follows, we examine the performance of our strategies until end-of-day Friday, Feb 28, 2020. I compared my tactical strategy to three passive benchmark strategies: (1) holding the SP500 Index, (2) selling volatility unhedged (-VIXY), (3) selling short-term volatility and hedging by going long on medium-term volatility (ie, VIXM-VIXY). As before, we extract the data from the FT website. The tactical strategy relied on predicting the term-structure of volatility based on the observed value of the features on the previous day. That is, we isolate the signal we believe contains predictive information about the vol term structure, and based on that signal, we position ourselves long or short the hedged selling portfolio (VIXM-VIXY) on the next day. We found that this betting strategy amounts to a corner solution in a more general hyperparameter space. So, how did we perform?
In order to check our performance, we split the data set into training sample (until November 27, 2019) and testing sample (November 28, 2019 to February 28, 2020), Then we look at the performance of the four strategies both in-sample and out-of-sample. In sample, the mean daily returns were +4bp (basis points=0.01 percent) for holding the SP500, +28bp for selling vol unhedged, +16bp for hedged selling, and +18bp for our tactical hedging strategy. As we saw then, the superior raw performance of selling vol unhedged comes with dramatically higher risk. The in-sample daily volatility of returns on the SP500 was 91bp, for selling vol unhedged was 400bp, and for both hedged selling and tactical hedging, it was 237bp. The in-sample Sharpe ratios are displayed in Figure 1.
How did we do out-of-sample? In the three months since our last analysis, all the passive strategies have lost money. The mean daily return on holding the SP500 has been -11bp, for selling vol unhedged it has been -56bp, for hedged selling it has been -44bp. But our tactical strategy has yielded a mean daily return of +12bp. The daily volatility of returns on holding the SP500 has been 103bp, for selling vol unhedged it has been 468bp, for hedged selling 309bp, and 312bp for our tactical strategy. Figure 2 displays the out-of-sample Sharpe ratios.
So how did our algorithm ended up making money over the last quarter when the passive strategies were bleeding money? Digging into the numbers, it turns out that we were long the hedging portfolio until last Monday, Feb 24. Then our predictive model signaled a risk-off in the offing and we went short the hedging portfolio for the rest of the week. It is worth looking at what happened over the last week in detail. Figure 3 displays the returns by weekday.
We can see that we did not call the risk off until end-of-day on Monday. But having called it, we made 5.4 percent on Tuesday instead of losing it, lost 1.9 percent on Wednesday instead of making it, but gained 9.7 percent on Thursday, the day of the worst returns on the SP500, and then made 4.1 percent again on Friday. Over the whole week, the market lost 11.7 percent, whereas traders selling volatility unhedged lost 37.1 percent of their investment. Hedged selling offered some protection—it lost 26.4 percent. Meanwhile, our tactical hedging portfolio gained 4.0 percent over the week. This meant that we made 30.4 percent more than hedged selling, 41.1 percent more than selling vol unhedged, and 15.7 percent more than holding the SP500 this week. Put simply, we made a killing while everyone else was losing their shirts.
No one could've predicted the pandemic. And we don't know where it goes from here. The moral of this week is that selling volatility unhedged is a fool's errand; you can lose your shirt very quickly indeed. And while hedging offers some protection, the real gold is in isolating features that allow you to predict risk-offs out-of-sample.