Ioulia Tretiakova_

Risk-Based Asset Allocation: Worth the Effort

Thank you for your comments, Nick.

You are absolutely right, one cannot use overlapping intervals to calculate autocorrelation because the metric will be naturally upward biased. The 252-day chart is there for illustration only, as you correctly observed. In autocorrelation calculation for observations to be IID, I used non-overlapping intervals, 21-day to be exact. In addition, to account for kurtosis, 3sd and higher events, such a few trading days in October 1987 were taken out since tail events are not persistent. An alternative, of course, is to use rank correlation.

I also appreciate your observation on the degree of equivalence of trend-following and risk-based investing. We have stared at risk for a long time and at some point I investigated the source of good backtested returns we were observing. A natural hypothesis, as you stated, would be to assume that a significant portion of it comes from trending. However, there are several aspects that make risk-based approach quite different from trend-following.
- Firstly, the turnover is very low (typically we observe 2-3 trades a year in the core portion of our portfolios with about 5-7 ETFs).
- Secondly, correlation to managed futures is in the low 20's (part of it is of course comes from not using short ETFs in the core portion).
- Finally, but most importantly, is that as a return estimation input of our portfolio construction engine we use equilibrium returns, or as a rough approximation, returns that are proportionate to risk. Magnitude of historical returns don't even enter the equation, only return volatility. We thus avoid chasing past returns, something a traditional mean variance allocation tool might do if it uses historical returns.