Since launching this website in May 2020, we’ve been absolutely delighted by the number of people who’ve reached out and had thoughtful discussions with us about our strategies. We’re also super grateful to all our subscribers who’ve stuck around through good months and bad and have also made some pretty useful suggestions on how we run our strategies and asked thoughtful questions along the way.
The bane of any long-term systematic strategy is the urge to look at short-term performance and tinker away till you get a better looking backtest. This is the definition of over-fitting, even though we convince ourselves that our changes are thoughtful and are a better way of capturing the underlying essence of what our strategies are about.
One of the main reasons for starting Beat Passive was fighting this urge to tinker and stick through our process. We decided that we’d only ever officially tune our strategies once per year, and never in a big way, mostly to build upon what we currently have.
In this post, we’ll go over the changes we’re introducing, the rationale behind those changes and what they mean for the backtested performance of the strategy.
VIX in points not percent (Tactical Volatility Strategy)
The first change that we’ve made that makes our Tactical Volatility Strategy much more robust is deciding on a leverage factor dynamically rather than using a static 0.5x that SVXY offers. The change does not affect the actual signal of going short, confused or long volatility but instead only how we use the signal ratio to decide the leverage ratio.
This change was inspired by @benneifert’s constant attempts at hammering into everyone not to think about the VIX in percentage terms but in points terms (different from equity price levels). The reason for this is:
3) Don’t quote implied volatility moves (eg VIX) in percentages and think that’s meaningful. Implied volatility changes have a much more stationary distribution in vol points than in percent. It doesn’t take much to double implied vol from an initial level of 8.— Benn "DJ D-Vol" Eifert (@bennpeifert) January 11, 2020
Basically, the VIX is much more likely to go from 10 to 20 in a day (100% increase and thus a wipeout if you’re short volatility, similar to what happened in February 5th 2018’s XIV episode) vs. going from 30 to 60 in a day. Thus, the initial level of the VIX matters. If we statically use 0.5x leverage that SVXY allows all the time, we are not taking into account this level. With our new dynamically deciding leverage, if we were caught short-vol on February 5th 2018, we would have suffered a ~35% drawdown because of our dynamic leverage of holding less SVXY instead of a 80%+ one that SVXY with full 1x leverage at the time suffered.
Our system generally avoids dangerous regimes and moves to the confused state so we avoided February 5th 2018 regardless, but this new way of dynamically levering gives us even more safety, even if we are stuck on the wrong side at the wrong time. The best bit of this change is that the backtested performance pre- and post- change remains almost the same (2.12 sharpe vs. 2.08 sharpe respectively), but gives us much more confidence that our strategy won’t blow up.
Hedged Long Volatility Exposure (Tactical Volatility Strategy)
All of our strategies sometimes take a long volatility position when correlations across asset-classes go to 1. This normally happens in periods of deep crisis (think Oct 2008, March 2020 etc), where liquidity evaporates and forced redemptions result in declines across asset classes; only cash is king. The only asset-class that benefits, and benefits in a huge convex manner, is long volatility.
During these times of crisis, our strategies try to concentrate into the long volatility position, trying to benefit from the convex exposure. Even though we only ever go into long vol very conservatively (4% of the time within the backtested period between October 2008 till December 2020), a lot of times we get false starts and get whipsawed out of the position. What seems like a ‘crisis’ whimpers down.
In order to protect ourselves against these whipsaws, we made a change to go into both long volatility and long market exposure as a hedge when the signal triggers. The hedge protects us when long vol isn’t really spiking and we don’t lose too much in the whipsaw. On the other hand, in times of real crisis, even though the hedge goes down, the long vol exposure goes up much more (due to its inherent convexity), thus still capturing most of the potential gain.
The newer hedged long volatility exposure is more positive skew and less volatile, while also suffers lower drawdowns due to whipsaws.
Mixed Momentum Ensembles (Dynamic Rebalanced Strategy)
Our previous dynamic rebalanced strategy was based on ensembling together different correlational and volatility estimate lookbacks and strategies to form a portfolio of uncorrelated asset classes weighted in a thoughtful fashion and leveraged appropriately.
We have now ensembled two additional variants into the existing strategy (and each variant uses ensembled signals and parameters within itself) which employ more momentum related metrics and signals to derive the appropriate weight and leverage of the same asset-classes that form the dynamic rebalance strategy.
This new mixed dynamic rebalance strategy retains the sharpe ratio of the previous model
while slightly reducing the maximum drawdown. Additionally, on extended backtests, it performs much better than the older strategy in periods of increasing inflation and/or when correlation estimates start to waver too much, as the momentum based signals kick-in and keep performance in line during those times.
In short, the new ensembled variants make the dynamic rebalance strategy more robust while retaining its original performance.
Over the coming few weeks we’ll update our website with the backtested results of the improved versions of the strategies. We would love it if readers and/or subscribers reach out with any questions and comments about the changes to our strategies.