We wanted to take a moment to highlight a post from the always smart JW Keuning describing a novel approach for measuring how well a strategy has performed relative to drawdowns (losses): Presenting the Keller Ratio.
Our preferred method for assessing a strategy’s return relative to drawdown has always been the Ulcer Performance Index, but the Keller Ratio offers the unique ability to adjust results based on an investor’s unique risk tolerance. In other words, the “best” strategy will be different for an aggressive versus a conservative investor.
In the spirit of cooperation, below we’ve shown Keller Ratio results for all of the strategies that we track, for investors with three different risk tolerances. See JW’s post to learn how the Keller Ratio is calculated. Note that the values alone don’t mean much – they’re most relevant when compared to one another. Values of zero mean that a strategy didn’t meet that investor’s risk tolerance.
Note how the “best” strategy varies depending on the investor’s risk tolerance. The top strategy for our three hypothetical investors are:
- AGGRESSIVE: Vigilant Asset Allocation
- MODERATE: Stoken’s ACA – Monthly
- CONSERVATIVE: Allocate Smartly’s Meta Strategy
When describing strategies, we always wrap “best” in quotations, as determining the best is a highly subjective exercise. There are so many different possible inputs, and so much is dependent on that unique investor. Having said that, these three strategies rise to the top of the list by a number of different metrics that we track.
Our take:
We like the ability of the Keller Ratio to tailor results to an investor’s unique risk tolerance.
If we had to put our finger on an opportunity to improve the metric it would be to incorporate the entire drawdown record (like the Ulcer Performance Index does), rather than just the single max drawdown.
Max drawdown tends to be driven by single significant events (like October, 1987). That means that designing strategies that simply minimize max drawdown, tends to increase the opportunity for overfitting the data, and thus, the likelihood of disappointing results out of sample. We’ve largely mitigated that problem by adjusting these results for “timing luck”, but most researchers don’t have the data/ability to do that.
Thank you to JW Keuning and Wouter Keller for the novel new idea. Both gentlemen are thought leaders in the tactical asset allocation space and we highly recommend following their work now.
* * *
Calculation notes:
- Not all strategies have the same start date, so to compare apples-to-apples, we limited results to just the previous 20 years (i.e. since April, 1998). There were obviously still plenty of opportunities for sizeable drawdowns during that time.
- We added one additional cool factor: we adjusted results for “timing luck” (i.e. we averaged results for monthly trading strategies across all possible “alternate trading days”). That means these results are a lot less prone to overfitting and much more useful for modelling future performance. If you’re new to the concept of timing luck, start here.
- Had we instead used all data for all strategies, and not adjusted for timing luck, results would have looked like this.