In this post I introduce an enhancement to the original volatility curve model. The enhanced version of the model introduces a mean reversion component to the original model that seeks to take advantage of extreme high volatility situations in order to enhance returns. Let’s dive right in.
First, let’s take a look at the a long term look of one part of the volatility futures curve. As a reminder, the original vol curve model basically is risk-off when the vol futures curve is inverted and risk-on otherwise.
As you can see from the chart, there are times when the vol futures curve inverts significantly (levels much higher than 1) and during these periods it takes a significant amount of time for the curve to return to normal. 2008 and 2020 are the most extreme examples as you can see by the amount of time the longer moving average spent in the inverted area. During these extreme periods there is an opportunity to take advantage of the situation and look for mean reversion in the curve, which would allow an earlier risk-on entry, thus higher returns. That’s it. That’s the basic concept of the new model. At the same time, the enhanced model tries to keep the number of trades and turnover to a reasonable level. Let’s take a look at the differences between the risk-off periods for the two models. Enhanced model is in yellow, original model is in blue.
As the chart shows the big differences are in 2008 and 2020 but there are some slight differences in the other extreme periods such as 2010, 2011, 2015/16. Also, all the differences come from the risk-on trigger, not the risk-off trigger. OK, now let’s look at some results of the various models. In general, you will see that performance improves but that also drawdowns go up a bit.
The difference in performance is not huge over the longer term but those few percentage points can make huge differences over the long run. In the shorter term, especially in extreme years, like so far in 2020, the differences are much more pronounced. In general, I think the model can make sense for those using the leveraged models (expect SSO/LT). The credit model (HYG/TLT) would also make sense. If you use one of the base models, SPY/TLT or SPY/IEF, I think it makes a bit less sense unless you tend to suffer from FOMO (fear of missing out) during those extreme periods.
Now, let’s look at the number of trades. As you would expect, the enhanced model does generate more trades but not that many. I use the SPY/TLT model as an example below but the number fo trades are independent of which ETFs are used to implement the model. The overall result is 4 more trades over the 13 year period.
Overall, the model is very similar to the original model but it just reacts quicker during extreme periods. This faster reaction time causes some more trading and some higher drawdowns but generates higher returns.
That’s about it for the new enhanced volatility curve model. Besides being a stand alone model, I think it can also be used as a complement to the original model. During normal times of volatility you stick with the original model and when the market enters a period of extreme volatility you can then start to look at the enhanced model. This involves some discretion but is something that I think can be managed and I will do for subscribers.
The model is now available to subscribers of the Quant Pulse Service along with the original model.