Site icon Flirting with Models

Ensembles and Rebalancing

This post is available as a PDF download here.

Summary

Two weeks ago, we wrote about the idea of payoff diversification.  The notion is fairly trivial, though we find it is often overlooked.  Put simply, any and all trading decisions – even something as trivial as rebalancing – create a “payoff profile.”  These profiles often fall into two categories: concave strategies that do well in stable environments is maintained and convex strategies that do better in the tails.

For example, we saw that rebalancing a 60/40 stock/bond portfolio earned a premium against a buy-and-hold approach when the spread between stock and bond returns remained narrow.  Conversely, when the spread in return between stocks and bonds was wide, rebalancing created a drag on returns.  This is a fairly trivial and obvious conclusion, but we believe it is important for investors to understand these impacts and why payoff is a meaningful axis of diversification.

In our prior study, we compared two different approaches to investing: strategic rebalancing and momentum investing.  In this (very brief) study, we want to demonstrate that these results are also applicable when applied to different variations of the same strategy.

Specifically, we will look at two long/short trend following strategies applied to broad U.S. equities.  When trend signals are positive, the strategy will be long U.S. equities and short the risk-free rate; when trend signals are negative the strategy will be short U.S. equities and long the risk-free rate.  We will use a simple time-series momentum signal.  The first model (“21D”) will evaluate trailing 21-day returns and hold for 1 day and the second model (“168D”) will evaluate trailing 168-day returns and holds for 14 days (with 14 overlapping portfolios).1  Both strategies implement a full skip day before allocating and assuming implementation at closing prices.

Source: Kenneth French Data Library.  Calculations by Newfound Research.  Returns are hypothetical and assume the reinvestment of all distributions.  Returns are gross of all fees, including, but not limited to, management fees, transaction fees, and taxes.  Past performance is not indicative of future results. 

So, what happens if we create a portfolio that holds both of these strategies, allocating 50% of our capital to each?  Readers of our prior note will likely be able to guess the answer easily: we create a concave payoff profile that depends upon the relative performance between the two strategies.  How, specifically, that concave shape manifests will be path dependent, but will also depend upon the rebalance frequency.  For example, below we plot the payoff profiles for the 50/50 blend rebalanced weekly and monthly.

Source: Kenneth French Data Library.  Calculations by Newfound Research.  Returns are hypothetical and assume the reinvestment of all distributions.  Returns are gross of all fees, including, but not limited to, management fees, transaction fees, and taxes.  Past performance is not indicative of future results. 

If we stop thinking of these as two strategies applied to the same asset and just think of them as two assets, the results are fairly standard and intuitive.  What is potentially appealing, however, is that the same literature and research that applies to the potential to create a rebalancing premium between assets can apply to a portfolio of strategies (whether a combination of distinct strategies, such as value and momentum, or an ensemble of the same strategy).

Below, we plot the annualized return of weekly rebalanced portfolios with different fixed-mix allocations to the 21D and 168D strategies.  We can see that the curve peaks at approximately 45%, suggesting that a 45% allocation to the 21D strategy and a 55% allocation to the 168D strategy actually maximizes the compound annualized growth rate of the portfolio.

If we follow the process of Dubikovsky and Susinno (2017)2 to derive the optimal blend of these two assets – using the benefit of hindsight to measure their annualized returns (7.28% and 7.61% respectively), volatility (17.55% and 17.97% respectively), and correlation (0.1318) – we derive an optimal weight of 45.33%.

Perhaps somewhat surprisingly, even if the correlation between these two strategies was 0.9, the optimal blend would still recommend about 10% to the 21D variation.  And, as extreme as it may seem, even if the annualized return of the 21D strategy was just 5.36% – a full 225 basis points below the 168D strategy – the optimal blend would still recommend about 10%.  Diversification can create interesting opportunities to harvest return; at least, in expectation.

And, as we would expect, if we have no view as to a difference in return or volatility between the two specifications, we would end up with a recommended allocation of 50% to each.

Conclusion

While most studies on rebalancing consider the potential benefits of combining assets, we believe that these benefits are trivially extended to strategies.  Not just different strategies, however, but even strategies of the same style.

In this brief note, we explore the payoff profile created by combining two naïve long/short trend following strategies applied to broad U.S. equities.  Unsurprisingly, rebalancing a simple mixture of the two specifications creates a concave payoff that generally profits when the spread between the two strategies is narrow and loses when the spread is wide.

More interestingly, however, we demonstrate that by rebalancing a fixed-mix of the two strategies, we can generate a return that is greater than either strategy individually.  We believe that this potential benefit of ensemble approaches has been mostly overlooked by existing literature and deserves further analysis.

 


 

  1. While seemingly random, there is nothing particularly magical about these signals.  The holding period was selected to align with approximately 1/12th of the evaluation period, to match the usual 1-year lookback with 1-month holding period.
  2. Dubikovsky, Vladislav and Susinno, Gabriele, Demystifying Rebalancing Premium and Extending Portfolio Theory in the Process (May 20, 2015). Available at SSRN: https://ssrn.com/abstract=2927791 or http://dx.doi.org/10.2139/ssrn.2927791

Corey is co-founder and Chief Investment Officer of Newfound Research. Corey holds a Master of Science in Computational Finance from Carnegie Mellon University and a Bachelor of Science in Computer Science, cum laude, from Cornell University. You can connect with Corey on LinkedIn or Twitter.

Exit mobile version