The perils of optimizers (Or: Don’t marry a model, part 2)

Optimizers are widely used in the construction and analysis of investment portfolios. But I’m not a fan.

“Good” is good, but “optimal” is not necessarily better

Optimizers are common in investment. After all, who wouldn’t want the best solution to a problem? So sales teams, portfolio managers, consultants, and others frequently wheel out the optimizer when they want to demonstrate the advantages of their particular solution.

But I’m not a fan. Because while “good” is good, “optimal” is not necessarily better.

The goal of optimization is to find the combination of investments that produces the highest utility (or risk-adjusted return.) In other words, it’s hill-climbing. But investment hills are not usually steep hills, and the top of the hill is the flattest point. Indeed, that’s the way many optimizers identify their target: they don’t directly seek the highest point, they seek flat points and then check if they are high or low ones.

So when you get to the top of the hill, it’s not really much higher than the points around it. For all practical purposes, anywhere in the neighborhood is just as good. My former colleague (and a current board member of New Zealand’s sovereign wealth fund, the NZ Super Fund), Craig Ansley, has likened it to a sign that he once saw on a long, flat, stretch of road somewhere between Ann Arbor and Pittsburgh: the sign announced that he had crossed from the Great Lakes water basin into the Mississippi water basin. The top of a flat hill. Not discernibly higher than anywhere around it—but higher it was if your measurement was precise enough. But Craig also points out that in optimization the inputs are never perfect, “so the flat hill is covered with long grass.”

So the portfolios that are produced by an optimizer might well have no advantage over other, alternative portfolios that are made up of very different investments.

At the top of a flat hill, there’s not a long way down

When you first realize this, it can seem like a bad thing, because it means that a small change in assumptions can give a very different answer about what the optimal solution is. But, really, it’s good, because it also means that the top of the hill with one set of assumptions is not very different from the top of the hill with another. We can be reassured that our result is robust, even if our assumptions are not perfect.

So my wariness of optimizers is not because they never have a role to play. Rather, it’s that their results are easily misinterpreted, and may be represented as meaning more than they do.

I’m reminded of the title of a very old Russell research paper written by Ernie Ankrim and Chris Hensel: “Asset allocation? How about the importance of common sense?” Indeed. Investment is not easy, but the difficulty does not usually lie in the math. It lies in the understanding.