Optimization and Evolutionary Dead Ends

I have always been somewhat of an armchair paleontologist. As a kid, I read everything I could about dinosaurs and loved collecting fossils.

One of the things that particularly fascinated me was how these massive, dominant creatures could have disappeared so suddenly and completely 66 million years ago. What could have possibly happened that caused such a mass extinction?

The discovery of the Chicxulub crater in the Gulf of Mexico's Yucatán Peninsula in the early 1990s confirmed what many had come to believe, that the proximate cause of the Cretaceous-Paleogene extinction event was a 120-mile wide asteroid that struck the earth with the impact of 10^23 joules, or about 4.5 billion times the explosive power of the Hiroshima atomic bomb.

Several thousand gigatons of material were ejected at velocities exceeding 5 kilometers per second, leading to a cloud of dust, soot, and ash that coated the entire earth within 4 to 5 hours of impact. This super-heated material caused mass wildfires, estimated to have burned 70% of the planet’s forests. The impact likely also led to massive tsunamis, with waves that may have reached up to a mile high and aftershocks that exceeded 11 on the Richer scale.

It was absolutely catastrophic. 75% of all species on Earth, including all non-avian dinosaurs, went extinct nearly instantaneously, at least geologically speaking.  

But at the same time, 25% of life managed to survive.

What was the difference between those species that lived and those nature selected for extinction? Well, almost unilaterally, those that went extinct were specialists – purely carnivores or herbivores. On the other hand, the vertebrates that survived were primarily omnivores, insectivores and carrion eaters.

From an evolutionary perspective, specialization is a double-edged sword.

Take for instance the titanosaurs. These last of the sauropods weighed upwards of 75 tons and ate between 400 and 900 pounds of plants per day. In an era when vegetation was plentiful, this survival strategy allowed the giant herbivore to flourish, spreading to almost every corner of the globe. They were optimized to their environment.

However, when environmental conditions change drastically – like when ash blackens the skies creating an “impact winter” that almost entirely halts photosynthesis – this optimization is precisely what led to their extinction. An enormous, calorie-intensive body with a specific food source was an evolutionary dead end. On the other hand, those animals with a broader palate were more easily able to find food from a variety of sources, and hence, survived. Smaller and nimbler was the path forward.

In an environment that doesn’t resemble the past, optimization can be a death sentence.

Which leads us to modern portfolio theory and mean-variance optimization, or MVO. I’ll spare the math, but MVO is strategic asset allocation method that uses risk, return and covariance assumptions across asset classes to select an “optimal” weight for each asset based upon certain objectives. Those objectives might be maximizing return given a maximum volatility threshold, or minimizing volatility necessary to achieve a return target, but the output is purely mathematically optimized, often using solver or some other iterative process.

Let’s look at a simple four asset mix MVO example below. In this example, we have two risky asset classes, a safe asset, and one in between. There are assumed returns and volatilities for these assets which - it should be noted - often look a lot like the past. This should seem basic, but familiar.

Assuming this investor has a required return of 6.0%, and with a covariance matrix in the background, let’s also assume that these are the weights that generate the lowest volatility with that specific return target.

The simple weighted average volatility of this asset mix would be 11.8%, so the benefit of diversification is evident in the expected volatility of 10.5% - a volatility reduction of 1.3% from having less than perfectly correlated assets in the portfolio.

However, as I was once told by a great investor, there are no future facts. And one thing you can be certain of is that all of these assumptions will be wrong to a certain degree.

So, the question is how does an optimized portfolio do in an uncertain future, one that may be very different from the past? First, let's examine what happens when the biggest position outperforms, but all other asset classes underperform their expectations. Well, in this case, the portfolio return of 6.1% exceeds the required rate return. Success.

But what about another scenario where the opposite happens? How does this portfolio do when the largest position underperforms, even if every other asset class materially beats their return assumptions? It should come as no surprise, but this portfolio performs quite poorly in that environment. In an environment that doesn’t look like the past, that optimal set of characteristics is potentially disastrous.

Now, let’s compare this optimized approach to a simpler asset allocation strategy, namely naïve diversification. With this approach, the four asset classes are simply equal weighted at 25% each. How does this more diversified method fare across the same scenarios?

It obviously doesn’t do as well in Scenario 2 where risky asset 1 outperforms, because it’s less concentrated. But it generates very nearly the same return in base case assumptions, and significantly outperforms in Scenario 3.

In an environment that doesn’t look like the past, that optimal set of characteristics is potentially disastrous, and the simpler, more diversified approach yields acceptable outcomes in all scenarios. On average, across all three, the second approach outperforms the MVO. Put another way, it increases the odds of survival.

An optimization is quite literally best fit to one environment – that’s what it means. But when environmental conditions are not as expected, optimization is what creates fragility. This is why I think mean variance optimizers should more accurately be called error term maximizers. The largest overweight is to the biggest mistake (which often happens because these are high return / high risk assets).

In part due to these challenges, and in search of more robust portfolios, a new approach to portfolio construction called the Total Portfolio Approach – or TPA – has emerged. And it has another interesting evolutionary parallel.

Convergent evolution occurs when animals independently evolve similar features in different geographies or time periods as a response to similar environmental conditions. These features are highly analogous, even though the animals may be highly unrelated.

The classic examples of this are parallel morphologies between ichthyosaurs and dolphins. Ichthyosaurs were marine reptiles that evolved from land dwelling ancestors around 250 million years ago before going extinct around 160 million years later, whereas dolphins – like whales – descended from land mammals about 49 million years ago.

Not only do they resemble each other superficially, but there are 20 anatomical features that are nearly identical in both form and function. Both animals adapted to be fast and nimble in the sea, able to swiftly maneuver to pursue and capture small fish and other marine prey while being able to avoid predation from other larger carnivores.

Similarly, the Total Portfolio Approach has evolved largely independently at numerous long-term institutions around the globe in recent years. From Singapore’s $770 billion sovereign wealth fund GIC to the nearly $600 billion pension CPP on the other side of the world in Toronto, and even the relatively tiny $2 billion Baylor University Endowment in Waco, Texas.

Recognizing the challenges of the traditional approach – such as siloed thinking, superficial diversification and difficulty combining public and private risk metrics – these investors have instead implemented a more holistic and opportunistic approach to constructing portfolios.

Despite having evolved from different ancestors, and certainly not identical in terms of implementation, these approaches all have a series of characteristics in common.

■ They start with goals, very clearly defined investment objectives

■ They employ an integrated process whereby investment opportunities compete for capital holistically

■ And they are dynamic, operating collaboratively in real-time

While not a singular, formulaic strategy, TPA provides several advantages relative to a traditional Strategic Asset Allocation (SAA).

TPA pushes the focus to achieving investment objectives as opposed to beating benchmarks, an important philosophical change I’ve written extensively about. A collaborative process allows the team to access innovative and interesting opportunities that often fall in between proscriptive asset class definitions. This opportunistic approach to accessing fundamental investment return drivers and risk factors as opposed to superficial asset classes also often results in a more truly diversified portfolio.

In essence, TPA captures the process of the Yale Endowment Model as opposed to mechanically replicating the output. It allows collaborative, professional investors to build durable portfolios, increasing the odds that the fund will perform acceptably across a broader range of potential environments and hence survive.

Markets, like nature, are competitive. And as Professor Andrew Lo has argued, they are also adaptive.

Those who refuse to change with the times – those not willing to adapt and compete with the prior version of yourself – may find themselves left behind, outcompeted by someone else taking your place in the ecosystem.

“It is not the fittest that survives; but the species that survives is the one that is able best to adapt and adjust to the changing environment in which it finds itself.”

Otherwise, you might find yourself optimized to a dead end.

The opinions expressed in this article are solely those of the author, and do not necessarily represent those of any entities or organizations.