In this post I present an extended backtest of the Volatility Curve Model. To understand the methodology and investment model behind my analysis you can read the original vol curve post here . The Q&A post on the model can be found here , and the enhanced version of the model here . Let’s dive right in.

The CBOE data on volatility futures goes back to Dec 2007. This, of course, has limited the backtest of any investment model that uses volatility futures. Fortunately, someone has done the work and used the CBOE methodology to calculate volatility futures prices back to 1990. For my analysis, I purchased the dataset from Vance at Six Figure Investing. You can read about his calculations and methodology here and of course you can buy the dataset for yourself. His analysis of the changes in volatility over time is also well worth the read. For the purposes of my volatility curve model, the relevant volatility term structure going back to 1990 is shown in the chart below. A ratio above 1, marked by the red line, represents an inverted curve, i.e. short term volatility futures priced above longer term volatility futures. An inverted curve usually is a sign of a risk-off environment and vice versa.

Now, I take the models I developed on the dataset from 2008 onward and apply the same models, the original and the enhanced model, to the full dataset. For this post my analysis will focus only on the SPY/TLT version of the model. In fact I was limited in my backtest to 1993 since that is the furthest back you can get daily prices for the SPY. I also combined each of the volatility models with my SPY-COMP model to produce two versions of a hybrid VOL-COMP model. So, we put all that together throw i into a mega large complex model, spend weeks validating and checking the data, and you get the following results. The table below shows returns, drawdowns, and number of trades for the various models. For comparison, I use SPY as the benchmark.

As you can see, I’ve broken out the data into various sub periods. Over the full backtest, from 1993 to now, the Volatility curve models (VOL OG and VOL ENH) hold up very well, and combined with COMP they do even better. Drawdowns are about half of SPY and number of trades per year remains reasonable. There are a few key things to note here I think. First, the volatility curve has worked better in more recent times than the past. Performance from 2000 forward is about 2x the SP500, over 2 bull and 2 bear markets. But before 2000, the volatility term structure behaved differently. You can see it in the performance numbers and you can sort of see it in the first chart above. In 2000 something changed. Vance also points this out in his analysis. If we dig deeper into one of the models, we can see that the majority of the underperformance pre-2000 was in 3 specific years, 1996-1998.

The dynamic behavior over time of the volatility curve has changed. Good thing is that it has changed for the better and has been quite effective since 2000, or for the last 18 years. But that also means that we need to be vigilant and watch for any structural change in the behavior of the volatility futures curve.

That’s about it for now. I’ll be following this up with some specific information for subscribers on how to apply these results to the individual models and which are best ones for particular situations.

Hi. Are available details how to generate signals by myself for “Volatility Curve Model” after subscribing to QuantPulse?

Hey Jan,

No. Those are not available.

Paul