The stark difference between mean-variance optimization and Brandywine’s “Predictive Diversification”

Originally published in the Brandywine Asset Management Monthly Report.


Brandywine just completed our third full year of trading Brandywine’s Symphony Program. Prior to launching the Program in July 2011 we presented our performance expectations. It was our belief that our Return Driver based investment approach and Predictive Diversification portfolio allocation model would enable us to achieve those results with a reasonable level of confidence.

We are pleased to report that our performance has continued to track our expectations. Over the past three years Brandywine’s conservative Symphony Program has achieved a +7.31% annualized return with a Sharpe Ratio of 1.03 and our more aggressive Brandywine Symphony Preferred has produced a +28.26% annualized return with a Sharpe Ratio of 1.03. Each of these performances ranks Brandywine among the top investment managers on either a risk-adjusted or absolute return basis. But more importantly, these performances track right in line with expectations.

While every investment manager is required to disclose that PAST PERFORMANCE IS NOT INDICATIVE OF FUTURE RESULTS, Brandywine has devoted decades to understanding the aspects of investing that contribute to predictive performance. After all, if the past provides NO indication of future performance, then investing is not “investing” at all. It is gambling. In contrast to Brandywine; most academic research, and even many investment managers, focus on creating optimal portfolios – not predictive portfolios. There is an enormous philosophical and practical difference.

The stark difference between mean-variance optimization and Brandywine’s “Predictive Diversification”

Brandywine’s Symphony Program is the result of Brandywine’s 30+ years of investment research and trading. When Brandywine began the development of our Brandywine Benchmark program in the late 1980s, we recruited some top academics and finance practitioners to assist us in developing the portfolio allocation model. That model would manage how much capital to commit to each individual trading strategy and market in our portfolio.

In one test, we provided one academic researcher with performance data from more than ten strategies trading across dozens of markets. He returned with the allocation we were to make to each strategy/market combination. In short, a few strategy/market combinations were recommended to receive large allocations, while most were to receive no allocation at all. When we expressed our concern to the researcher that this concentrated portfolio was unlikely to perform well in the future (it didn’t pass the “sanity” check), his response was that what he provided to us was the “perfect” answer.

What we came to realize was that it was the perfect answer – but to the wrong question. The question the academics and our researchers had been answering was “how do I create the optimal portfolio?” – meaning one that displayed the best risk-adjusted returns on that past data. In contrast, the correct question should have been, “How do I create the most “predictable” portfolio?” – one where future performance will most closely match past performance (either tested or actual). After all, if you have low confidence that the results will repeat, then they are not really useful results at all – even if they are “optimal.” His answer was perfect ONLY if future data, i.e. future market fluctuations, were similar to past data – which of course we know is not going to be the case.

Surprisingly, there is very little (almost no) research on methods for producing the most predictable performance. Instead, decades of research have been wasted on answering the wrong question. Nobel Prizes have been awarded for it. Not surprisingly, people (and academics are people) will devote their efforts to answering the questions for which they receive the greatest reward. And for academics, the Nobel Prize is often perceived as the ultimate reward.

Not so with Brandywine. Our interest is in producing the best possible returns for our clients and us. The realization 25 years ago that we and others were asking the wrong question led to a significant amount of new research that resulted in Brandywine’s “Predictive Diversification” portfolio allocation model.

This model has withstood the test of time. First with the performance of our Brandywine Benchmark Program in the 1990s and continuing today with the performance of Brandywine’s Symphony Program. This model is predicated on the belief that the future will NOT be identical to the past. By solving for predictability, rather than optimizing on past returns, the model is better able to handle the natural changes in market conditions that are detrimental to optimized portfolios. The combination of our Return Driver based investment approach and Predictive Diversification portfolio allocation model is the reason that our actual performance has so closely matched our past performance. In addition to that (and somewhat ironically), Brandywine’s actual performance is more “optimal” (in both absolute and risk-adjusted terms) than portfolios that were constructed with the specific intent of being “optimal.”

The focus on allocation models designed to produce the most optimal performance on past data has only served to distract investors and managers from the most critical aspect of investing – ensuring that future performance tracks past performance as closely as possible. As a result of this focus, investment performance results appear to be random. The top managers over one period fail to perform over subsequent periods. While we most certainly cannot guarantee performance results, we can at least make the statement that Brandywine’s Symphony Program was developed with the prime directive of achieving the most predictable performance possible.

Share this post:
  • Facebook
  • Twitter
  • email