Welcome to the third issue of the sixth volume of The Journal of Investment Strategies. We present here two different approaches to the perennial problem of robust portfolio construction, a deep examination of backtesting engines, and a rigorous take on a long-standing technical trading technique.
In the issue’s first paper, “Agnostic risk parity: taming known and unknown unknowns”, Raphael Benichou, Yves Lempérière, Emmanuel Sérié, Julien Kockelkoren, Philip Seager, Jean-Philippe Bouchaud and Marc Potters – the research team at Capital Fund Management (CFM) – present a novel take on the notion of risk parity. They identify their paper’s main goal as the construction of a portfolio that will result in a robust out-of-sample diversification: in other words, a portfolio that truly is equally driven by all of the relevant driving factors. Unlike the more traditional approaches to risk parity, the CFM team focuses on the notion of managing not only measurable and forecastable risks (known unknowns) but also unmeasurable and unforecastable uncertainties (unknown unknowns) when building resilient portfolios. In a signature physicist style, they solve the problem by positing a deeper symmetry requirement – a rotationally invariant symmetry of the portfolio representation through assets – and derive the plausible solution from it with minimal additional assumptions.
This symmetry requirement has a solid reason behind it: without it, we may be able to achieve good diversification with regard to one definition of “fundamental assets” while failing to achieve any reasonable diversification under another definition of fundamental assets. The notion that we might need to think more deeply about the definition of assets is not an idle thought, given that in many cases portfolio managers actually implement their views with derivatives, which can be easily redefined to include or exclude any particular exposure. The results of this paper will likely be most interesting to quants running actively traded portfolios, such as diversified futures trend-following funds, which have recently seen a challenging period in terms of performance. A better portfolio construction methodology will be a welcome development for these investors.
“Black–Litterman, exotic beta and varying efficient portfolios: an integrated approach” is the second paper in this issue. Here, Ricky Alyn Cooper and Marat Molyboga investigate the interrelationships among the Black–Litterman, exotic beta and risk parity approaches to robust portfolio construction. The authors show that these approaches need not be considered as alternatives to each other; instead, they can be considered as being part of an integrated framework, unifying the various insights. In particular, the authors recast the exotic beta portfolios as “views” within the Black–Litterman framework. This part is not controversial at all, as exotic betas represent potential robust outperformance sources, and therefore the exposures to such betas should indeed be consistent with the Black–Litterman approach, in which a portfolio manager is free to express their forecasts with respect to the relative outperformance of certain positions compared with equilibrium returns.
The second contention of Cooper and Molyboga – that one can use risk parity portfolios instead of equilibrium (market cap) ones as a starting point for Black–Litterman methodology – is both more interesting and more controversial. It is interesting because it is quite possible that, as the authors show, such a starting point leads to potentially better end results and better stability. However, it is also controversial because we no longer have as firm a logical foundation as in the classical interpretation. A leap of faith is made to assume that a portfolio with a higher Sharpe ratio, such as a risk parity portfolio, may in fact be acceptable as the Bayesian prior in the Black–Litterman method.
I do not have any practical objections to this claim: it is entirely plausible. However, whether the entire world would consider this an incontrovertible truth remains to be seen and, in fact, if they did, it would probably have such an impact on the market that the risk parity portfolio would become identical to the market cap, ie, it would clear the supply and demand.
Robert Löw, Stanislaus Maier-Paape and Andreas Platen investigate a tricky concept in “Correctness of backtest engines”, our third paper. As most practitioners of investment management, along with many academics, rely on such engines, either commercial or homegrown, for conducting their research, the results of this paper and the authors’ cautionary words should be of great importance for a large audience. The main problem the authors point out is to do with execution assumptions under incomplete information, such as when using compressed open–low–high–close (OLHC) bars and even tick data. They caution against believing the results that assume the strategy algorithm could have actually traded at a trigger price at the time when such a price triggers an order in our strategy. They offer several unit tests that can be incorporated into algorithmic trading software to guard against such creeping inconsistencies, which would then inevitably lead to bad out-of-sample performance. Correspondingly, I believe the issues highlighted by the authors are very important, and that anyone using backtest software should check whether their results are, in fact, affected by these problems.
While I concur with the authors’ focus on the consistency of backtest software, I think these important problems can be mitigated in most settings by following a carefully and conservatively designed algorithmic strategy. For example, if the strategy, as a general rule, includes a time lapse between the decision making and the emission of the trade order, then such a strategy will be more robust to the issues highlighted by the authors. Some more advanced backtesting software vendors actually include this time lapse feature in their software settings, so the researcher does not even need to explicitly account for it in the strategy: it is already taken care of within the backtest module.
In the fourth paper, “Statistical testing of DeMark technical indicators on commodity futures”, Marco Lissandrin, Donnacha Daly and Didier Sornette undertake a deep study of a few of the best-known technical trading indicators: the DeMark Sequential, Combo and Setup trade indicators. It is admittedly uncommon to see a piece of deep, quantitative research covering the decidedly nonacademic topic of technical trading. However, since these indicators are actually quite popular, with many individual and even professional traders relying on them and often just following the output of some software or data subscription vendor, we think a deeper dive into this subject is well justified. The authors construct sophisticated random resampling tests to verify the statistical significance of the indicators. Further, they show that, in the case of commodity futures trading, the manner in which the futures roll is implemented affects the expected performance of the technical trading. With such in-depth analysis, this paper will certainly be a necessary read for many strategists utilizing technical trading methods.
In conclusion, I hope this issue of The Journal of Investment Strategies will find a broad audience, and that readers will not only read the paper that first catches their eye but also look through the other papers in the issue, as such a cross-pollination of ideas is what drives creativity and innovation in the field. It was certainly one of the original objectives of our journal, and I am pleased that we have been able to continue in this vein for over six years.
Arthur M. Berd
Founder and CEO, General Quantitative LLC
This paper offers a new perspective on portfolio allocation, which avoids any explicit optimization and instead takes the point of view of symmetry.
This paper brings Black–Litterman optimization, exotic betas and varying starting portfolios together into one complete, symbiotic framework.
In this paper, the authors provide tools to test the correctness of backtest engines for setups with at most one entry and one exit.
This paper examines the performance of three DeMark indicators over twenty-one commodity futures markets and ten years of daily data.