Quant Investing

大圣电竞比赛赛程手机版

Update: added comparison to SPY-COMP.

This post introduces a quant trading model based on volatility. More specifically it uses the prices of volatility futures contracts based on the SP500 to make risk-on and risk-off decisions that can be used to trade various risk-assets.

Why Volatility?

There is a bunch of research that shows that historical volatility is predictive of future near-term volatility. See here for a good summary of that research. In other words, volatility tends to cluster, low volatility tends to lead to low volatility and higher volatility is associated with higher volatility in the near future. You can think of this as volatility having some of the same characteristics of momentum. Risk-assets, like stocks, exhibit a negative relationship between returns and volatility. Low volatility, higher returns, higher volatility, lower returns. In the quant world these traits have been used for a while in volatility targeting and volatility scaling strategies. Even in the tactical asset allocation world, strategies like Adaptive Asset Allocation (which you can track at Allocate Smartly ) from the Resolve team make use of volatility targeting to improve risk-adjusted returns.

At the other end of the quant world, the short-term, trading oriented side, volatility, in particular volatility futures contracts have been used for a while to actively trade various inverse volatility ETFs, like the now infamous XIV. See here for Volatility Made Simple’s approach and here for QuantStratTradeR’s approach to trading inverse vol ETFs. Both have subscription options if those approaches interest you.

In my approach I use the volatility futures contracts to generate a simple risk-on, risk-off signal for trading a broad range of US stock ETFs, that trades more like a longer term TAA model rather than a short-term trading model.

The Volatility Signal

Like some of the high frequency trading strategies, I use the pricing of short term VIX futures contracts relative to longer term VIX futures contracts to generate a risk-on, risk-off signal. Without giving away the goods, basically, when the short term VIX prices rise above the longer term VIX prices the model goes risk-off. And vice versa, when the VIX prices fall back below longer term VIX prices the model goes back to risk-on. Chart 1 below shows an example of what I’m talking about. It plots the ratio of short-term volatility, as measured by the $VIX index, to the longer-term volatility, as measured by the $VXMT index over the last 5 years. The red line at 1, represents the level where short-term volatility is greater than longer term volatility, i.e. volatility is in backwardation. This would represent a risk-off signal. Similarly, when short-term volatility falls back below some level relative to longer term volatility the model goes back to risk-on mode.

Chart 1: VIX short term futures to VIX long term futures ratio

The Volatility Curve Model

Let’s start by looking at the risk-on risk-off signal history of the model. Data goes back to April 2008. Chart 2 below shows the volatility signal generated by the model; 1 meaning risk-off and 0 meaning risk-on. As you can see from the chart it is possible to generate a pretty stable signal from what looks like a very noisy ratio in Chart 1.

Chart 2: Volatility Model Signal

Now let’s look at a simple application of the model. We use the vol model’s risk-on, risk-off decisions to trade between SPY and TLT. The results from April 2008 are shown in table 1 below. From April 2, 2008 to December 19, 2009 the SPY/TLT model returned 17.91% with a max daily drawdown of -18.6%. The results compare to a SPY buy and hold return over the same period of 9.76% with a -51.5% drawdown.

Table 1: Annual and Compound Returns for SPY/TLT model

Some very nice risk-adjusted returns but what about trading? I wanted a model that did not trade like a high frequency model. Table 2 shows all of the trades since April 2008 for the SPY/TLT model. Basically, there have been 21 trades in 11.5 years. That is not a high number and comparable with many monthly TAA models.

Table 2: Trade history for SPT/TLT volatility curve model

Applications to other risk-assets (stocks, ETFs, bonds)

Of course, the volatility model can be applied to trade many risk-assets. Table 3 below is my tracking sheet for the volatility model where I keep track of several implementations of the model. Results are as of Dec 19, 2019. I also have various benchmarks listed. In every case you get a better risk reward profile than the buy and hold option.

Table 3: Volatility Curve Model Tracking Sheet

As you can see, due to the quick reaction of the volatility signal, the model can be used quite effectively to trade a very leveraged ETF portfolio (TQQQ/TMF model as an extreme example) depending in your risk tolerance. Also, it can be used in a more conservative framework to even trade risky bonds (the HYG/TLT model) with improved results.

Conclusion

Using the volatility futures curve to generate a risk-on risk-off signal for risk assets generates some impressive risk-adjusted returns. Also, despite being a strategy that requires daily monitoring the amount of trading and turnover is not excessive. I’ve been using the model throughout this year and it has also been in beta testing by a few early subscribers. It will be available as part of the Quant Pulse service by the end of this year. I’ll make a separate announcement when it is available.

Update: below is a comparison of the Vol Curve Model signals to the SPY-COMP signals from my Economic Pulse Newsletter over the same period of time, i.e. from April 2, 2008. Reminder, SPY-COMP is a monthly model. It was risk off until the end of April 2009. The other two risk off months were June 2010, and August 2011.

Tagged , ,

23 thoughts on “ Quant Investing: Volatility Curve Model

  1. Hi Paul,

    I really like this piece. Would be interesting to see the comparison to this with some of the more popular TAA strategies to see where it shines (or falls short) vs. the more traditional ones like GEM. Keep up the good work.

  2. A side-by-side comparison of Chart 2: Volatility Model Signal vs. SPY-COMP for the same time period would be really interesting. Peter W.

    1. Thanks Ilya. And thanks for the offer. I’ll keep it in mind as we enter the new year.

      Paul

  3. Paul, very interesting. What signal did you implement for trading? I’ve tried to reproduce your results using a 3-day SMA of the VIX:VIXM crossing 1 but I get a lot more trades than what you’ve listed, and a CAGR closer to 13%. My version makes it look like there’s too much fluctuation around 1, resulting in clusters of whipsaw trades that seem undesirable. And superimposing your list of trades in Table 2 on my implementation suggests you are either using a slightly different values for VIX/VIXM, or using something other than closing prices, or smoothing the ratio with something stronger than SMA(3). Please share any details on this that you’d care to disclose, thanks.

      1. That’s fine, just wanted to make certain that I wasn’t missing something. I can definitely appreciate how this is an important aspect of the model you describe – and one that it makes sense to highlight as a proprietary advantage. Thanks for the response!

  4. Super helpful, thank you Paul.

    Is there a way you could compare to the SPYCOMP performance in Table 3?

    Does this work with other asset classes? For example EFA and EEM. The idea being that – similar to SPYCOMP – the signal in the US influences global markets?

    Happy holidays!
    Ben

    1. Hey Ben, it’s not a 100% apple to apple since SPY-COMP is monthly but if I take SPY-COMP from the end of March 2008 to approx the same end date, SPY-COMP returned about 14.8%, with a DD 0f -19.4%.

      Also, this does not work with other asset classes like EFA or EEM individually. The vix futures curve I’m using is siley based on the SPY.

      What I haven’t done and will do in the coming year, is use relative momentum on top of the vol curve model, i.e. to rotate between SPY and EFA lest say during risk-on.

      Paul

      1. Love it Paul, thank you. I’ll subscribe to your quant newsletter to follow this signal.

        Is there an argument to be made to be short SPY when both the economic and volatility indicators are risk off?

        1. Cool. Don’t know really. There is not enough historical data to make a judgement on that.

          Paul

    1. Risk-on means that the model owns risk assets. Risk-off means the model owns safe assets. For example, in the case of the SPY/IEF option, SPY is held during the risk-on period, and IEF is held during the risk-off period.

      Paul

  5. Just looking at the historical data (although short-term) it suggests if you’re going to bother with leverage in this case, you might as well “go all the way” with a small portion of portfolio assets.

    Is there any potential “this isn’t working anymore” trigger for a strategy like this?

    1. Good question. No idea really, you’d have to see vol spike or drop precipitously with no effect on prices which seems impossible given the VIX futures calculations…Or more realistically the real threat is getting whipsawed to death….

      Paul

  6. Great post – I was confused with the expression “daily drawdown” which sounds like max drawdown in a single day. I know you can’t mean that.

    Have you compared this VIX risk-on risk off with the well known 200 day simple moving average SPY/TLT switching model? I’ll bet your method is better than that but a comparison could be interesting. Thanks for this.

    1. Thanks. Yes, it is max drawdown but just calculated with daily prices. People often also use max drawdown calculated on a monthly price basis. Max drawdown with daily prices is always more than that calculated with monthly prices. So, it’s important to distinguish the two.

      Yeah, the VIX model does much better than SPT/TLT using the 200 day SMA.

      Paul

  7. Paul I just hapenned on this post. Thanks for sharing your research.

    This is an indicator I have been playing around with myself since 2012 when Vance Harwood first wrote about the VIX/VXV as a timing tool for shorting volatility.

    In my testing I have noticed that the 1.0 ratio flip can move up and down in different volatility regimes and therefore a strategy where the trigger wiggles around 0.9 to 1.1 depending on the conditions might work better to eliminate some of the whipsaws. This is something I am working on.

    Another thing I have noticed is that Vol tends to raise slowly at first and by a larger amount when it spikes. Looking at the distance from its recent average allows to trigger a risk-off signal a couple of days early and before the scary spikes in vix.

Comments are closed.