Update on VOL CURVE Model YTD Performance

The VOL CURVE model recently closed out it’s latest trade, that was started by the risk-off signal on Feb 24th, 2020. Here I just wanted to update the results of the model including this latest trade.

First, let’s just look at the current snapshot of where the model’s year to date and overall the long term compared to it’s benchmarks.

As you can see the results are pretty good, year to date, and of course remain very strong over the longer term.  On an absolute return basis the models are outperforming their respective risk indexes, except for the the technology index based models, TQQQ/TMF and QLD/TLT. In those cases, the returns vs risk (drawdowns) are still in favor of the VOL CURVE model. I don’t know many people who would tolerate a 70% drawdown.

Now let’s take a look at an update of the trades for a couple of the models, SPY/IEF, and SPY/TLT. The chart below shows performance of SPY vs the risk-off assets IEF and TLT during the risk-off period from Feb 24th to June 18th, 2020. During this period the SPY returned -2.73% with a -30.4% drawdown while IEF and TLT returned 6% and 8.6% with 0% and -3.5% drawdowns respectively. Pretty darn good and exactly what the model was intended to do.

The most frequent question I received during the risk-off period for the model was why it took the model so long to go back to risk-on when the market bottomed in mid to late March. The answer is that the model is deigned to avoid any really high level of volatility. Since volatility during the market’s recent drawdown and bounce back remained extremely elevated with respect to history, the model avoided going risk on until market volatility returned to a more normal level. That is the gist of it. Of course, that begs the question is there a faster reacting version of the model which would catch market bounce backs sooner. Well, there is but it comes with some very nasty tradeoffs, mainly many whipsaws and a lot of trading, two things I am loathe to introduce into the model.

However, there is something that can be done. Basically, the approach is to introduce a mean reversion component to the model only when volatility reaches extreme levels. The rest of the time the model remains in ‘normal mode’. There is still some more whipsaws and trading but not too bad. The chart below shows the performance of this enhanced model over the same periods as the model above.

As you can see from the chart this mean reversion component would have made a huge difference in 2020. The enhanced model would have gone risk-off on the same date as the base model but back to risk-on on April 14th resulting in a large performance difference over the normal model. Over the longer term, the enhanced model provides better performance over the normal mode but with larger drawdowns and more trades (about 25% more trades). It seems to make more sense for the leveraged models than the non-leveraged models but that depends on your goals and risk-tolerance. I’ll be introducing the enhanced model to subscribers over the next few weeks.

Finally, I have been able to get access to a volatility dataset that allows me to take the model back to 1993. I’m still working on churning out all the data and backtests but the model holds up incredibly well over the full period from 1993 to today. So stayed tuned for an upcoming post on that.

** Note: The VOL CURVE model is part of the QuantPulse membership. If you’re interested in singing up, see here.