Quant Investing , TAA Investing


Next up on the performance review for 2021 are the Volatility Curve Models that are part of the Quant Pulse service . For a quick background on these models and hoe they work, start with this post where I first introduced the concepts behind the volatility curve model. Also, here are a few other background posts: an extended backtest of the model , combining it with SPY-COMP , and a Q&A on the model. Ok, let’s dive right in.

First, let’s update the historical risk-on, risk-off chart with 2021 data. The chart below shows the risk-on (0 on the chart) and risk-off triggers (1 on the chart), for the two versions of the models that I run, the pure vol model and one combined with the COMP indicator from Economic Pulse. Basically, VOL-COMP is a more conservative version of the vol curve models. As you can see, 2021 was a quiet year for the risk triggers with only one risk-off trigger in 2021 near the beginning of the year. That brings the total number of trades since the beginning of 2008 to a total of 34 over the 14 year period, or 2.42 trades per year. Pretty low, especially considering that the models can react on a daily basis.

Note: in the backtests below I’ve been able to extend the model back in history to 1999 using a synthetic data set. But the real, live volatility futures data only extends back to 2008.

Now, let’s move on to the application of these models using various risk-on and risk-off asset classes. We’ll start out with the leveraged models. These models use one or more equity ETFs during the risk-on period, and one or more government bond ETFs during the risk-off period. For example, the SPYTLT model, invests in SPY during the risk-on periods and TLT during the risk-off period. Some of the models, like VTIVEUTLTBIL, use relative momentum during the risk-on and risk-off period to decide among a choice of ETFs, e.g in this case VTI or VEU during the risk-on period. In the table, below I’ve summarized the performance of the models, separated by Vol Curve and VOL-COMP, over the historical time period (1999 to 2021) and various sub periods. I’ve ranked performance by the 2009 to 2021 period because that is the period with the real volatility curve data.

As you can see from the table, absolute and adjusted performance is impressive pretty much across the board. And as I mentioned above, Vol Curve vs Vol COMP gives you higher returns but with a bit higher risk. Now, let’s look at the application of leverage to these models. The table below summarizes the effect of using leverage on the models above, either through the use of leveraged ETFs or margin (for the backtest I have used the fee structure of the various publicly available leveraged ETFs, or applicable margin rates where no leveraged ETFs were available). As you can see from the table, a lot of care needs to be taken when using leverage. But VOL-COMP, does a nice job of reducing a lot of that leverage risk. For example, two of the VOL-COMP leveraged models have max daily drawdowns comparable to 60/40 with obviously way higher returns. These are definitely not for the faint of heart but can really enhance a portfolio when used judiciously.

That’s about all I wanted to cover today. The volatility models continue to performs very well and as expected.


Tagged , , ,